Seven hundred years after the death of Dante Alighieri in 1321, the Divine Comedy continues to be read, taught, quoted, adapted and argued over across the world. This article asks a simple question with a complicated answer: why does a medieval Christian poem about the afterlife still speak so directly to modern readers who may share none of its theology? The argument developed here is that the Comedy survives partly because it is one of the most detailed maps of the inner life ever written. Long before Sigmund Freud named the #unconscious, Dante built a three part architecture of the mind in which what is hidden is punished, what is admitted is worked through, and what is loved is finally integrated. Using a hermeneutic and interpretive method, this study reads Inferno as a topography of #repression, Purgatorio as a sustained process of #mourning and #working_through, and Paradiso as a vision of psychic integration that reaches the limits of language itself. The article draws on Freudian, Kleinian, Jungian, Lacanian and contemporary relational frameworks, while remaining alert to the serious dangers of anachronism raised by current Dante scholarship. It also examines the figure of the guide as an early model of the therapeutic relationship, the poet's exile as a laboratory of loss and creativity, and the modern afterlives of the poem in medicine, popular culture and mental health discourse. The conclusion argues that Dante's legacy is not merely literary but psychological: he gives readers a usable grammar for describing suffering, change and hope. Keywords: Dante Alighieri, Divine Comedy, psychoanalysis, unconscious, mourning, contrapasso, individuation, literary theory, medieval literature, mental health humanities 1. Introduction In the opening lines of the Inferno, a man in the middle of his life wakes up inside a dark wood and cannot say how he got there. He has lost the straight path. He is frightened, disoriented, and unable to describe the place he is in without shuddering. Very few openings in world literature have been quoted so often, and there is a reason for that. The scene is not primarily a piece of theology. It is a description of a psychological emergency, and readers who have never opened a medieval commentary still recognise it instantly. The #selva_oscura, the dark wood, has become one of the standard images through which modern people describe breakdown, burnout, grief, addiction, depression and the sensation of being lost in a life that looked, from the outside, entirely successful. This article takes that recognition seriously and asks what it means. Dante Alighieri died in exile in Ravenna in 1321, which means that the twenty first century reader is separated from him by roughly seven centuries of intellectual history. In that time, the vocabulary available for describing inner experience has been completely rebuilt. We now speak of the #unconscious, of trauma, of attachment, of defence mechanisms, of cognitive distortion. Dante spoke of sin, grace, virtue, providence and the soul. These are not the same languages. Yet when a modern reader follows the pilgrim down through the circles of Hell, up the terraces of the mountain, and out into the light of the spheres, something is communicated that survives the translation. The claim of this article is that what survives is a psychological structure, and that the tools of psychoanalytic criticism, used with care, can make that structure visible. The choice of psychoanalysis as a lens is not arbitrary. Psychoanalysis and the Comedy share a fundamental assumption: that a person is not transparent to themselves, that the truth about a life is often buried, and that recovery requires a descent into what has been avoided before any ascent is possible. Freud built a therapeutic method on the idea that the way out is through. Dante built a poem on the same idea. The pilgrim cannot climb the sunlit hill directly. The beasts drive him back. He must go down first, through the whole catalogue of human self deception, before the mountain becomes climbable. In the language of clinical work, there is no shortcut around the material. In the language of the poem, there is no path to Paradise that does not pass through Hell. At the same time, this article is not an attempt to turn Dante into a Freudian before Freud, or to diagnose a medieval poet with a modern disorder. Contemporary Dante scholarship has become rightly suspicious of readings that flatten the historical Dante into a mirror of the critic. Recent work by Barolini (2022) insists on the disruptive particularity of Dante's thought and warns against interpretive narratives that smooth away difficulty. Brilli and Milani (2024) show how much of what we think we know about Dante's life is in fact Dante's own autobiographical construction, repeated by later biographers as though it were archival fact. Cornish (2022) reminds us that the poem asks to be believed in ways that modern criticism finds hard to accommodate. Any psychoanalytic reading that ignores these cautions will produce a comfortable but dishonest Dante. The approach taken here is therefore double: it uses psychoanalytic concepts as instruments of #close_reading, while treating the historical distance between the poem and the theory as part of the evidence rather than as an obstacle to be brushed aside. The article proceeds in stages. It first reviews the state of Dante studies and the state of psychoanalytic literary criticism, since the encounter between them has a longer history than is usually acknowledged. It then sets out a theoretical framework and a method. The central sections offer a sustained reading of the three canticles as three phases of psychic work. Later sections examine the guide as a model of the therapeutic relationship, exile as a case study in loss and #sublimation, and the modern afterlife of the poem in medicine, education and popular culture. The article closes with a candid discussion of the limits of this approach and with suggestions for students who wish to take the work further. 2. Literature Review 2.1 Dante studies in the seventh centenary and after The commemorations of 2021 produced an unusual concentration of scholarship, and the years since have not slowed the pace. Several strands are relevant here. The first is biographical and methodological. Brilli and Milani (2024) argue that Dante's life must be reconstructed by separating the surviving documents from the poet's own literary self presentation, because the two have been fused for centuries. This matters enormously for any psychological reading, since psychoanalytic criticism has historically been tempted to read the poem as a disguised confession by the man. If the man is himself partly a literary invention, then the confession is at least twice removed from the biographical facts, and the critic must say so. The second strand is theological and philosophical. Turner (2022) presents Dante as a theologian in his own right rather than a poet decorating borrowed doctrine, and Cornish (2022) explores what it means to believe a fiction that claims to be true. These works are a corrective to the older habit of treating the theology as a shell around a secular human core. A responsible psychoanalytic reading cannot simply peel away the doctrine and keep the psychology, because in the poem the two are the same substance. The third strand is embodied and material. Webb (2022) reads the Comedy through gesture, showing how meaning is carried by bodies, hands, faces and postures rather than by speech alone. Callegari (2022) examines food, hunger and social identity. Phillips-Robins (2021) studies liturgical song as a practice that shapes the souls who perform it. These readings are unexpectedly useful for psychoanalytic work, because they demonstrate that Dante thinks about the psyche through the body, which is precisely the terrain of affect, drive and #trauma in modern theory. A fourth strand concerns reception. Luzzi (2024) traces the strange biography of the poem itself, showing how it has been revived, suppressed, worshipped and rewritten across seven centuries. DiMassa (2022) examines its transformation in German Romanticism. Wolf (2022) collects perspectives on the unexpected forms Dante's presence takes. This literature is essential to the question of legacy, because it shows that the Comedy has never had a single stable meaning. Each age has found in it what it needed, and our age is finding a psychology. The Oxford Handbook of Dante, edited by Gragnolati, Southerden and Lombardi (2021), is a useful marker of where the field currently stands. It is deliberately plural in method, treating gender, embodiment, philosophy, politics and reception as equally legitimate points of entry. That pluralism licenses the present study, while also warning it not to claim too much. 2.2 Psychoanalysis and literature On the other side of the encounter, psychoanalytic literary theory has itself changed. Martin (2023) offers a recent map of the field, moving from Freud through Klein and Winnicott to Lacan, Zizek, feminist and queer readings, postcolonial critique, #trauma theory and posthumanism. What emerges from that map is that psychoanalytic criticism today is far less interested in diagnosing authors and far more interested in how texts organise #desire, how they manage what cannot be said, and how reading itself is a relational act. This is the version of psychoanalysis used here. Two further developments matter. The first is the psychosocial turn represented by the collection edited by Frosh, Vyrgioti and Walsh (2024), in which psychoanalytic concepts are used to think about collective as well as individual experience. Khanna (2024), writing in that volume on Freud's essay on mourning and melancholia, shows how the distinction between healthy grief and stuck grief has been extended to think about groups, histories and unresolved political losses. That extension is directly relevant to a poem written by a political exile about a city that had expelled him. The second is the growing body of work in the medical and health humanities that treats the Comedy as a source of clinical insight. Bianucci, Lanini, Donell and Lippi (2023) read the three canticles as a progression of sensation, from bodily pain in Inferno, through sound in Purgatorio, to light in Paradiso, and argue that this sequence mirrors modern understandings of pain and its management. Mandarano, Masciangelo and Armocida (2022) examine Dante's clinical descriptions from an endocrinological perspective. Dumitrescu (2022) explicitly places Dante alongside Freud and Dali, arguing that the Inferno anticipates central psychoanalytic themes. Older but foundational contributions in this line include the classic argument that the Comedy can be read as a medieval #psychoanalysis, an idea that has proved surprisingly durable. 2.3 The gap What is missing from this literature is a synthesis aimed at students. The Dante scholarship is rigorous but assumes years of training. The psychoanalytic literature is powerful but often applies theory to the poem without attending to recent philological caution. The medical humanities work is suggestive but brief. This article attempts to occupy the middle ground: a reading that is theoretically serious, historically honest, and written in language that an undergraduate or a general reader can follow. 3. Theoretical Framework Five bodies of theory inform the readings that follow. They are set out briefly here, in plain terms, because the analysis assumes them. Freudian foundations. The core Freudian claim is that the mind conceals things from itself. Material that is unacceptable is pushed out of awareness through #repression, but it does not disappear. It returns in disguised forms: in dreams, in slips, in symptoms, in compulsive repetition. Dreams work by #condensation, where several meanings are packed into one image, and by #displacement, where energy attached to one thing is transferred to another. The mind is also structured by conflict between wish, prohibition and the demands of reality, which Freud eventually formalised as the relation between #id, #ego and #superego. Two later Freudian ideas are indispensable here: the distinction between #mourning, in which loss is gradually accepted, and #melancholia, in which the lost object is absorbed into the self and attacked there; and the #repetition_compulsion, the tendency to repeat painful patterns rather than remember them. Kleinian and object relations. Melanie Klein described two basic positions of mind. In the paranoid schizoid position, experience is split into idealised good and persecuting bad, and unwanted feelings are expelled by #projection. In the #depressive_position, the person recognises that good and bad belong to the same object and the same self, feels guilt, and is moved toward #reparation. Wilfred Bion added #containment: unbearable feelings become bearable when another mind can hold them and return them in a digestible form. Jungian analytical psychology. Jung read the psyche in symbolic terms. Growth occurs through #individuation, the gradual integration of split off parts of the personality into a larger #the_self. That process requires an encounter with #the_shadow, the disowned aspects of the personality, and with the inner image Jung called the #anima. His insistence that wholeness requires a descent, and that guides appear at the threshold, fits a poem structured as descent followed by ascent. Lacanian theory. Lacan reread Freud through language. The subject enters #the_symbolic_order by accepting language and law, and acquires a permanent #lack. Desire is not a need that can be met but a movement that slides from object to object. What cannot be symbolised at all is #the_real, which appears as trauma and rupture. This is useful for the last canticle, where Dante repeatedly announces that language is failing him. Contemporary relational and trauma informed approaches. Modern clinical work emphasises the relationship as the instrument of change: #attachment, the provision of a #secure_base, the capacity for #mentalization, and the slow building of a coherent narrative out of fragmented experience. Trauma theory adds that overwhelming events are not stored as ordinary memory but as intrusive, timeless, bodily material that resists narration. Both illuminate the poem's obsession with guides, witnesses, and the difficulty of telling what happened. Two concepts from within Dante's own system must sit alongside these. The first is #contrapasso, the principle that the punishment expresses the sin, so that the soul in effect becomes its own condition. The second is #allegory, the medieval conviction that a text carries several simultaneous levels of meaning. Both concepts do work that is structurally similar to psychoanalytic interpretation, and the analysis below exploits that similarity while noting that the resemblance is not identity. 4. Method This is a qualitative, interpretive study in the tradition of #hermeneutics and psychoanalytically informed literary criticism. It does not use quantitative measures, and it makes no claim about Dante's actual mental health, which is unknowable and, in any case, not the point. The method has four steps. First, selection. Passages are selected because they are structurally load bearing, because they have attracted sustained critical attention, or because they generate strong reader identification across cultures and centuries. The dark wood, Francesca, Farinata, Ulysses, Ugolino, the frozen lake, the terraces of pride and envy, the dreams of Purgatorio, the confession before Beatrice, and the closing vision of Paradiso are treated as the spine of the argument. Second, #close_reading. Each passage is read first on its own terms, attending to imagery, structure, sound and gesture, before any theory is applied. This ordering is deliberate. Theory that arrives too early manufactures the evidence it then discovers. Third, theoretical mapping. Psychoanalytic concepts are then brought alongside the reading, and the fit is assessed honestly, including where it fails. Where a concept illuminates, that is stated. Where it distorts, that is also stated. Fourth, contextual checking. Each interpretation is tested against recent Dante scholarship. If a reading requires Dante to be a modern secular psychologist, it is rejected. The aim is a reading that a historian of medieval literature would recognise as unforced, and that a clinician would recognise as psychologically exact. The limitations of this method are considerable and are discussed openly in section 12. 5. The Dark Wood: Crisis as the Beginning of Knowledge The poem begins with a failure of orientation. The pilgrim reports that he found himself in a dark wood, that the straight way was lost, and that he cannot fully say how he entered because he was so full of sleep at the moment he left the true path. Three details in that opening deserve attention. The first is the phrase about the middle of the journey of our life. It is not the middle of my life but of our life. From the very first line, the poem generalises. The crisis is offered as a shared human structure rather than a private complaint. This is one of the reasons the passage has become the standard modern shorthand for the #midlife_crisis, even though Dante means something considerably larger and more theological than a modern reader usually means by that term. The second is the sleep. The pilgrim was full of sleep when he wandered off. He did not choose to get lost. He drifted, and the drift happened below the level of conscious decision. This is an extraordinarily precise description of how people actually arrive at crisis. Almost nobody decides to ruin their life on a specific Tuesday. The damage accumulates in a state of partial awareness, through a thousand small avoidances, until one morning the person looks up and does not recognise where they are. In psychoanalytic terms, the pilgrim has been living under the government of what he does not know about himself, and the wood is what that government finally looks like from the inside. The third is the inability to describe. The pilgrim says the wood is so bitter that death is scarcely more so, and that it is hard to say what it was like. Here the poem records a well documented feature of severe distress: the failure of language. Overwhelming experience resists narration. Modern trauma theory describes this as the gap between the event and the capacity to symbolise it, and clinicians hear it constantly in the form of the sentence, I cannot explain it, you would not understand. Dante does not merely report the wood. He reports the difficulty of reporting it, and that second order difficulty is the more accurate psychological observation. Then come the three beasts: the leopard, the lion and the she wolf. Commentators have argued for centuries about their exact allegorical identities. What is uncontroversial is their function. They block the direct route. The pilgrim can see the hill lit by the sun, he can see the way out, and he cannot take it. Anyone who has watched a person in the grip of a compulsion, an addiction or a depressive episode knows this precise configuration: the destination is visible and the road to it is not passable. Insight alone does not free the patient. Knowing what would help is not the same as being able to do it, and the poem states this fact in the first canto, seven hundred years before behavioural science made it a research programme. The response to this impasse is the arrival of a guide, and the guide immediately prescribes a different route. Virgil tells the pilgrim that he must take another road. The straight ascent is closed. The descent is open. This is the central structural insight of the entire poem and it is also, in a different vocabulary, the central structural insight of depth psychology. There is no reliable path to integration that bypasses the material one is avoiding. The way to the light runs through the dark. Freud's early patients wanted symptom relief without excavation, and he found that the symptoms returned. Dante's pilgrim wants the sunlit hill without the descent, and the beasts drive him back. The two systems reach the same conclusion by entirely different roads. 6. Inferno as a Topography of the Repressed 6.1 The gate, and what it means to enter The inscription over the gate of Hell famously ends by instructing those who enter to abandon all hope. The pilgrim is terrified, and Virgil's answer is instructive: he tells him that here he must leave all cowardice behind, and that they have come to the place where he will see the sorrowful people who have lost the good of the intellect. That last phrase is worth pausing on. The damned have lost the good of the intellect. They have not lost intelligence. Many of them are brilliant, articulate, persuasive and self justifying. What they have lost is the capacity to see themselves truly. This is a definition of #resistance so exact that it could be transplanted into a clinical textbook without alteration. The damned can talk endlessly. They cannot change. This is the first structural correspondence between Hell and the #unconscious. Hell is not a place of unconsciousness in the sense of unawareness. It is a place of frozen awareness, of endless articulate repetition without transformation. Each soul tells its story, often beautifully, and the story is always the same story, told the same way, forever. The technical name for this in psychoanalysis is the #repetition_compulsion. The name for it in the poem is damnation. 6.2 Contrapasso as symptom formation The organising principle of Hell is #contrapasso: the punishment fits the sin because the punishment is the sin, made visible and permanent. The lustful, who allowed themselves to be blown about by passion, are blown about by an eternal wind. The gluttonous, who lived for consumption, lie in filth and rain. The wrathful tear at each other in the mud. The flatterers are immersed in the filth they produced. The sowers of discord are cut apart. The traitors, who froze the bonds of love, are frozen in ice. Read psychologically, #contrapasso is a theory of the symptom. In psychoanalytic thinking, a symptom is not an arbitrary affliction added to a person from outside. It is the person's own conflict, expressed in a compromised form. The symptom says what the person will not say. It carries the meaning that consciousness has refused. If a man cannot acknowledge his rage, the rage does not vanish; it appears as migraine, as insomnia, as a pattern of ruined relationships, as a body that will not stop signalling. The punishment is the disowned content, returning in a form the sufferer cannot recognise as their own. Dante's Hell is a museum of this process. Nobody in it is being punished by an external force in the ordinary sense. They are being punished by what they made themselves into. The eternal wind is what unbounded #desire actually feels like from the inside once it is no longer restrained by anything. The ice is what betrayal actually is: the freezing of the capacity to be moved by another person. In this sense the whole cantica can be read as an unfolding of a single proposition: the state of the soul is the environment of the soul. That proposition is not far from the clinical observation that people construct, and then inhabit, the emotional worlds their defences require. 6.3 Francesca: desire, narrative and the seduction of the story The fifth canto is the most read canto in the poem, and the psychological reason is not hard to find. Francesca da Rimini, blown eternally on the wind with her lover, tells the pilgrim how they came to be there. They were reading a romance together, about Lancelot and Guinevere. They read the passage where the lovers kiss. Their eyes met. The book fell from their hands. They read no further that day. Almost everything psychoanalysis has to say about #desire is compressed into that account. Notice first that the desire is not spontaneous. It is mediated. Francesca and Paolo do not simply want each other; they want each other through a story about other people wanting each other. Their #desire is borrowed, learned, modelled on a text. This is precisely the structure that later theory calls mimetic or mediated desire, and it is also what Lacanian theory means when it says that #desire is always the desire of the Other. It comes to us from outside, formed in the field of language and image, and it is never simply our own private appetite. Notice second what Francesca's speech does. It is exquisite, courteous, rhythmically beautiful, and it never once accepts responsibility. Love seized her. Love led them to one death. Love, in her grammar, is always the subject of the sentence and she is always the object. Her language performs, at the level of syntax, the exact defence that put her in Hell: she is a person to whom things happened. This is #displacement of agency, and it is one of the most common structures in clinical practice. The patient describes a life full of events with no author. Notice third what happens to the reader, and to the pilgrim. Dante the pilgrim faints from pity. Dante the poet has constructed a canto so seductive that generations of readers have wanted Francesca released, have found her punishment cruel, have used her to accuse the poem's God. The canto seduces us exactly as the romance seduced her. We are placed inside the mechanism we are being shown. This is a technical achievement of a very high order, and it produces something close to #countertransference in the reader: our emotional reaction to the material becomes itself part of the evidence about what the material is. The pilgrim's faint is therefore not a failure of nerve. It is the poem showing that identification precedes understanding, and that the first honest response to another person's ruin is to feel how close it is to one's own. 6.5 Ulysses: the drive that will not stop Ulysses appears inside a tongue of flame, and tells how he persuaded his aged companions to sail beyond the known limits of the world, urging them that they were not made to live like brutes but to pursue virtue and knowledge. They sailed for five months, saw a mountain in the distance, and were sunk by a whirlwind. Readers, especially modern readers, tend to admire this. That admiration is the interpretive problem. Ulysses is in Hell, and the poem is quite clear that he belongs there, yet his speech is one of the noblest in the poem. What has gone wrong? A psychoanalytic reading offers one answer. Ulysses represents a #desire that has detached itself from relationship. He abandons his father, his son, his wife. He speaks of virtue and knowledge, but the actual structure of his life is a pure, unstoppable forward movement that consumes everyone attached to it. This is what happens when the drive to know becomes untethered from the capacity to love. It looks like heroism. It functions like a compulsion. Freud's late idea of a drive that pushes beyond the pleasure principle, toward dissolution, is not a bad description of a man who persuades old men to row into the empty ocean because stopping is intolerable to him. The contemporary relevance is uncomfortably direct. A culture that treats limitless expansion, restless achievement and permanent self optimisation as self evidently good will find Ulysses admirable, and will be unable to explain why he keeps arriving, in life after life, at the same whirlwind. 6.6 The lower Hell: dissociation, the frozen lake, and the failure of empathy As the pilgrim descends, the sins become colder, and so does the poetry. The upper circles punish appetites, which are at least human and warm. The lower circles punish deliberate acts of the will: fraud, betrayal, the cold manipulation of other people's trust. At the very bottom, there is no fire at all. There is ice. This is a striking psychological choice. Popular imagination expects Hell to get hotter. Dante makes it colder, and the coldness is the point. The worst state, in this system, is not the state of being overwhelmed by feeling. It is the state of feeling nothing at all about what one does to others. The bottom of the psyche is not rage; it is the absence of relation. Modern clinical language would speak here of dissociation, of affective numbing, of the failure of #mentalization, of the incapacity to hold another person's mind in one's own. The frozen souls of Cocytus cannot even weep, because their tears freeze in their eyes and seal them shut. They have lost the physical capacity for grief. Whatever one thinks of Dante's theology, this is an exact image of what happens to a person who has systematically extinguished the capacity to be affected by the suffering they cause. The episode of Count Ugolino, gnawing eternally at the skull of the man who imprisoned him and his children, brings the cantica to its psychological climax. Ugolino's story is a story of #trauma in its purest form: enclosure, starvation, the death of children, and a final line so ambiguous that it has generated eight hundred years of argument about whether he ate them. What matters for this reading is the structure of his eternity. He tells the story, and he chews. He tells it again, and he chews. He is not being punished with a foreign torment. He is being kept, forever, inside the moment that destroyed him, endlessly recounting it, endlessly re enacting it. That is the definition of traumatic repetition. The event will not become a memory. It remains a present. At the very centre stands Satan, gigantic, weeping, and mechanically beating his wings, which produces the wind that freezes the lake that imprisons him. He is the author of his own paralysis. He is also, importantly, silent. The lowest point of the poem is a creature who cannot speak. The unconscious core of the system, the place where everything terminates, is the place where language stops entirely. Lacanian criticism would recognise this immediately as an encounter with #the_real: not what is hidden and can be revealed, but what cannot be symbolised at all. 6.7 Summary of the Inferno reading The cantica has an internal logic that maps, with real precision, onto a psychology of avoidance: Sins of appetite are punished lightly, because appetite is human. Sins of the will are punished heavily, because they involve the deliberate destruction of relation. Souls repeat their stories without changing them. Articulacy is not insight. The punishment is the state of the soul made external. #contrapasso is a theory of #symptom formation. The deeper one goes, the colder it becomes. Numbness is worse than agony. The centre is mute. What is most repressed cannot be spoken at all. None of this requires Dante to be a proto Freudian. It requires only that Dante was an exceptionally accurate observer of what human beings do when they refuse to know themselves, and that Freud, six centuries later, was another. 7. Purgatorio as the Work of Mourning If the Inferno is the most quoted cantica, the Purgatorio is the most useful one, and it is the part of the poem that clinicians, teachers and readers in recovery return to most often. The reason is simple. Hell describes a condition. Purgatory describes a process. 7.1 Time returns The first thing that changes when the pilgrim emerges onto the shore of the mountain is that time starts again. In Hell there is no morning. There is no schedule, no season, no development. In Purgatory there is a sunrise, and the souls cannot climb at night. There is a beach, a queue, a delay, an order of business. The mountain has a timetable. This is psychologically enormous. Where Hell is timeless, Purgatory is temporal, and the difference between the two is precisely the difference between being stuck and being in treatment. Traumatic material is timeless: it does not age, it does not soften, it arrives in the present tense. Material that has been worked through becomes historical: it acquires a past tense, it can be dated, it becomes something that happened rather than something that is happening. The restoration of time is the first sign that change has become possible. 7.2 The terraces as a therapeutic sequence The mountain is organised by the seven capital vices, and the pilgrim climbs through them in order: pride, envy, wrath, sloth, avarice, gluttony, lust. This order is not arbitrary, and it repays attention. Pride comes first, at the bottom, because it is the hardest and the most fundamental. In the theology, pride is the root sin, the refusal to accept one's actual position in reality. In psychological terms, pride is the defended, inflated self that cannot tolerate need, dependence or error. Nothing else can be treated until this is treated. Any clinician will recognise the sequence: the work cannot begin until the patient can bear to be a person who needs help. On the terrace of pride, the souls walk bent under enormous stones. They are learning, physically, to carry the weight of their own limitation, and to look at the ground rather than above it. The bodily nature of this is exactly what Webb (2022) draws attention to in her study of gesture: the poem believes that the body is where inner change is registered. Crucially, the purging is not simply punitive. Each terrace has three components: examples of the virtue to be learned, examples of the vice to be abandoned, and a physical discipline that reshapes the habit. The souls are not merely suffering. They are practising. They rehearse. They sing. They meditate on images. This is not the psychology of punishment; it is the psychology of learning, and it is remarkably close to what modern therapeutic models would describe as graded exposure, cognitive restructuring and behavioural rehearsal, delivered inside a supportive community. Phillips-Robins (2021) shows how far the liturgical singing on the mountain is a practice that forms the singers, not merely a decoration of the scene. The souls are being changed by what they do together. 7.3 Suffering that is chosen The most important difference between the two cantiche is not the intensity of the pain. Some purgatorial punishments look, from outside, exactly like infernal ones. The proud carry stones. The wrathful walk in choking smoke. The lustful walk through fire. The difference is consent. The souls in Purgatory want to be there. They ask to stay longer. They correct the pilgrim when he pities them. The pain is not imposed on them from outside; it is embraced from inside, because they know what it is for. This distinction is one of the most useful things the poem has to offer a modern reader, and it deserves to be stated bluntly. The same suffering can be either damnation or #psychic_change, depending entirely on whether it is meaningless repetition or meaningful work. Grief that is refused becomes #melancholia, and the mourner turns the loss against the self, endlessly. Grief that is entered becomes #mourning, and it slowly does its work and slowly ends. The pain may be indistinguishable on the surface. The structure is completely different. Khanna (2024), writing on Freud's distinction, emphasises that the difference lies in what happens to the lost object: whether it can be let go, or whether it is swallowed and then attacked inside the self. Dante built an entire mountain on that difference three centuries before Freud named it. 7.4 Dreams The Purgatorio contains three dreams, one for each of the pilgrim's nights on the mountain, and they are among the strangest passages in the poem. A dream of an eagle that snatches him upward. A dream of a stammering, ugly woman who becomes seductive as he looks at her, and who is then exposed by another figure. A dream of two women, one active and one contemplative. The middle dream is worth dwelling on because it is a textbook case of what Freud would call #dream_work. The woman is disgusting at first sight. The dreamer's own gaze transforms her, and the more he looks, the more beautiful she becomes. Then a holy lady appears, and another figure tears the woman's clothing to reveal what is beneath. This is a nearly perfect illustration of how #projection and idealisation operate. The object is not intrinsically attractive; the desire of the looker manufactures the attraction. What seduces is what we have added. When the addition is stripped away, the object is revealed as it is. Whatever Dante meant by this theologically, and he meant something specific about the false goods that hold the human will, the psychological mechanism he describes is one that a modern reader recognises from advertising, from infatuation, from addiction and from politics. We fall in love with what we have projected, and then we call it fate. 7.5 Beatrice, and the hardest scene in the poem At the summit of the mountain, in the earthly paradise, Virgil disappears. He does not say goodbye; the pilgrim turns to speak to him and he is simply gone. And in the place where the pilgrim expects comfort, Beatrice arrives and reproaches him. She names him, the only time in the whole poem his name is spoken. She lists what he did after her death: how he turned away, how he followed false images of good, how he went where he should not have gone. He weeps. He cannot speak. He confesses. He faints. Readers often find this scene disappointing. They expect a reunion, and they get an interrogation. But the scene is the psychological hinge of the entire work, and it is worth being precise about why. First, this is the moment where the pilgrim is finally required to say, in his own voice, what he has done. Everything before this has been observation. He has looked at other people's sins for two entire canticles. Now the material is his. The long tour of Hell turns out to have been preparation for a single sentence about himself. Any reader who has been in a therapeutic process will recognise the sequence: a great deal of intelligent discussion of other people, and then, eventually and unwillingly, the sentence that costs something. Second, the confession is met with love that does not soften the truth. Beatrice does not tell him it was fine. She does not tell him it was unforgivable either. She holds both: she is the person who loves him and the person who tells him exactly what he did. That combination is, in relational psychoanalysis, the definition of a #containment that actually works. A presence that only comforts produces no change. A presence that only accuses produces collapse. Change requires both at once, and the poem stages it in a scene of extraordinary emotional violence and extraordinary tenderness. Third, the pilgrim's shame is described with startling physical accuracy. He cannot lift his eyes. His voice breaks before it reaches his lips. He is described as a child who stands accused, looking at the ground. Dante distinguishes #shame, which attacks the whole self, from #guilt, which attaches to the act, and he takes the pilgrim through shame and out the other side into an act of confession that guilt makes possible. Modern research on the moral emotions has spent decades establishing that shame tends to produce hiding and defensiveness while guilt tends to produce repair. The Purgatorio dramatises that distinction and then stages the passage from one to the other. Finally, the pilgrim is drawn through the river Lethe, which removes the memory of sin, and later drinks from Eunoe, which restores the memory of the good he has done. This double movement is subtle and, psychologically, exactly right. Recovery is not the deletion of the past. It is the reorganisation of memory so that it no longer poisons the present, and so that what was good is available again. That is not forgetting. It is the completion of #mourning. 8. Paradiso and the Limits of Language The Paradiso is the least read cantica and the most psychologically radical one, and the reason is the same in both cases: it is about a state that cannot be described. From the first canto, the poet announces that he cannot tell what he saw. He says that memory itself cannot follow. He compares himself to a man who wakes from a dream and retains the feeling of it but not the content. He says language is not enough, again and again, and he says it more insistently the higher he climbs. The final vision is compared to the effort of a geometer trying to square the circle, an impossible task, and the poem ends not with a description but with a movement: the poet's #desire and will turning like a wheel that is moved by love. This is not a failure of the poet's power. It is the subject of the last cantica. Dante is writing about the point at which experience exceeds symbolisation. Lacanian criticism has a precise name for this: the encounter with what cannot be captured in #the_symbolic_order and around which language circles without ever landing. Modern readers can meet this in far more ordinary places than mystical vision. Grief cannot be said. Certain kinds of joy cannot be said. What happened in the worst hour of a person's life cannot be said. Every clinician has watched a patient reach for words and find that the words are not there, not because the patient is inarticulate but because the experience is not language shaped. What is remarkable about the Paradiso is that Dante does not solve this problem. He performs it. He builds a poetics of failure, in which each attempt at description is followed by an admission that the description was inadequate, and the admissions themselves become the most moving thing in the poem. In psychological terms, this is one of the highest achievements in the entire work: he demonstrates that a person can stand at the edge of what cannot be spoken and remain there, without panic, without collapse, and without pretending to a mastery he does not have. The technical term for that capacity, in modern clinical language, is tolerance of #ineffability, and it is rarer than it sounds. The other great psychological theme of the Paradiso is that individuality survives integration. This is the answer the poem gives to a question that troubles every model of psychic wholeness. If the goal is unity, does the person disappear? Dante's answer is no. Each soul in Paradise remains recognisably itself. Piccarda is still Piccarda. Cacciaguida is still an irritable, proud, loving old soldier. The blessed are not dissolved into a uniform light. They are more themselves, not less. Barolini (2022) has drawn attention to precisely this celebration of difference inside unity as one of the poem's most striking and least conventional moves. For a psychoanalytic reader, this is the poem's account of #integration. The #individuation Jung described is not the erasure of the parts but their reconciliation. The self that emerges at the end of a long analysis is not a smoothed, generic, well adjusted person. It is a more fully inhabited version of the particular, awkward, specific person who began. The Paradiso insists on this, and it insists on it against the obvious pull of its own theology, which might easily have argued that the individual is swallowed by the divine. Dante refuses that. The refusal is one of the reasons the poem still matters. 9. The Guide: Virgil, Beatrice, and the Therapeutic Relationship No aspect of the Comedy has more direct relevance to modern psychological practice than its treatment of the guide. The pilgrim never travels alone, he never chooses his own route, and he never completes a stage without help. This is a poem about being accompanied. Virgil is the guide for the descent and the climb. He is patient, he is firm, he explains, he carries the pilgrim when the pilgrim cannot walk, he rebukes him when he lingers, and he is frightened himself at the gates of the city of Dis and has to wait for help to arrive from elsewhere. That last detail matters enormously. The guide is not omnipotent. There are places where he cannot get them through, and they must simply wait, in fear, until assistance comes from beyond him. Any honest account of the therapeutic relationship must include such moments. Virgil's function maps closely onto several things at once. He is the #secure_base from which the pilgrim explores: the pilgrim keeps checking whether Virgil is still there, and Virgil's presence is what makes the terrifying material survivable. He is a containing mind, receiving the pilgrim's terror, processing it, and handing it back in a form that can be thought about rather than merely suffered. And he is the object of an intense #transference, which is exactly why his disappearance is written the way it is. Virgil leaves at the top of the mountain, without a farewell, at the moment when the pilgrim is about to receive what he most wants. The pilgrim turns to tell him something and finds nothing there, and he weeps, and he calls Virgil's name, and Beatrice tells him not to weep yet, because he will have something else to weep for. This is a description of termination. Every therapy that works ends, and it ends by removing the person who made the work possible. The guide who cannot enter the earthly paradise is the guide whose usefulness has been completed. Virgil represents human reason, and human reason can get a person to the top of the mountain and no further. What lies beyond is not reasonable. It is loved. And so the instrument that carried the pilgrim this far must be laid down, and the loss of it is real and painful and correct. Beatrice is a more complicated figure and has been badly served by sentimental readings. She is not simply the beloved. She is the one who sends Virgil in the first place, at the beginning of the poem, descending into Limbo to arrange the rescue. She is the initiator, not the reward. She is also the one who reproaches him, tests him, refuses him easy comfort, and repeatedly turns his gaze away from herself and toward what she is showing him. That last movement is the crucial one. In the Paradiso, Beatrice does not want to be looked at. She wants him to look where she is looking. When he stares at her, she tells him to turn. When he is finally ready, she takes her place in the celestial rose and leaves him with a different guide, Saint Bernard, and she smiles at him from a great distance. A guide who wants to be the destination has failed. Beatrice's whole function is to be the medium of a movement that goes beyond her. That is a description of good practice which many professions, not only clinical ones, would do well to sit with. It is also a description of a transference that is worked through rather than exploited: the intense attachment is used to make the journey possible, and then it is transformed rather than indulged. Bernard, the final guide, does almost nothing but pray and point. By the end, the pilgrim needs less and less mediation, and the last movement of the poem is one in which he sees for himself. The sequence of guides is therefore a sequence of decreasing dependence, ending in a moment of unmediated experience that the poem cannot describe. 10. Exile, Loss, and the Poem as Reparation Dante wrote the Comedy as an exile. He was condemned in Florence in 1302, on charges he regarded as politically manufactured, and he never returned. He died in Ravenna in 1321. He knew that if he went back he could be killed. It is tempting to build a straightforwardly biographical psychology on this fact, and this article resists that temptation, for the reasons set out earlier: Brilli and Milani (2024) have shown convincingly how much of the standard Dante biography is Dante's own self construction, absorbed uncritically by centuries of readers. What can be said with confidence is not about the man's private psyche but about the structure of the work. The Comedy is a poem in which a man who lost his city builds another city, permanently, in language, and assigns everyone their proper place inside it. He puts his enemies in Hell. He puts himself on the mountain. He puts his beloved in Heaven. He gives judgement, which is the one thing exile took away from him. Psychoanalytically, this is a case study in #sublimation and in what Klein called #reparation. The unbearable loss is not denied and it is not simply endured. It is transformed into a made object that outlives the loss and outlives the maker. The rage is real and it is visible on almost every page: the invective against Florence, against the popes, against the corruption of Italy, is not decorative. But the rage is bound inside a structure. It is put to work. It becomes architecture. There is also a #melancholia in the poem that it never fully resolves, and the honest reader should say so. Dante's Florence is idealised in a way that no real city has ever deserved. His Beatrice, who died young, becomes the vehicle of ultimate truth. The lost objects are not merely mourned; they are elevated to a status that no living city and no living woman could occupy. That is the signature of grief that has been partly transfigured and partly frozen. Luzzi (2024), tracing the reception of the poem, notes how consistently later exiles and prisoners have recognised themselves in it, among them Primo Levi and Antonio Gramsci. They recognised, one suspects, not only the consolation but the wound that the consolation never quite closes. This is precisely why the poem is useful rather than merely beautiful. It does not offer a clean recovery. It offers a life's work built on a loss that never healed, and it argues that such a thing is possible, and that what is made from it can be more valuable than what would have been made without it. That is a claim about the human capacity for repair, and seven hundred years of readers have found it credible. 11. Contrapasso and the Modern Science of Self Punishment One of the most persistent misunderstandings of the Inferno is that it is a fantasy of revenge, in which a bitter poet enjoys the torment of his enemies. There is some truth in that, and it would be dishonest to deny the vindictive pleasure that runs through parts of the poem. But the structure of the punishments tells a more interesting story, and it is a story that modern psychology has independently arrived at. In Dante's system, God does not invent the torment. The sinner does. The soul is placed where it already was. This is stated most clearly by the fact that souls fall to their circle by their own weight, as it were, drawn there by what they are. Hell is not something that happens to a person. It is what a person becomes when nothing corrects them. The internal judge is another point of contact. Freud's #superego is not a wise conscience. It is often a cruel, primitive, punishing internal figure that borrows the voice of authority and uses it against the self, sometimes with an intensity far beyond anything the person's actual crimes deserve. Anyone who has listened to a depressed patient describe themselves has heard a sentence being carried out. The Inferno externalises this internal court and shows it in operation. But here the poem does something that a purely clinical account often cannot. It insists that the court can be left. The whole architecture of the Comedy is a claim that condemnation is not the last word, that judgement is a stage rather than a terminus, and that the same person can pass from the place of endless self accusation into a place where accusation becomes a task and the task eventually ends. Whatever one believes about the theology, the psychological structure of hope embedded in that claim is not trivial, and readers in extreme circumstances have repeatedly reported that it held them. 12. Seven Centuries of Relevance: Dante in the Modern Mind 12.1 Medicine and the health humanities The Comedy has become an unexpectedly active text in clinical education. Bianucci and colleagues (2023) read the three canticles as a graded progression of sensory experience, from bodily pain, through sound and music, to light, and argue that this sequence maps onto contemporary approaches to pain management, in which pharmacological control is combined with music, environment and attention. Mandarano, Masciangelo and Armocida (2022) analyse Dante's descriptions of bodily conditions from a modern endocrinological standpoint. Others have examined his descriptions of sleep, collapse, and what looks strikingly like psychomotor slowing in depressive states. None of this proves that Dante was a physician. It proves something more interesting: that he described bodily and mental suffering with such observational accuracy that specialists, centuries later, can still recognise the phenomena. That is what makes the poem clinically useful in teaching. It gives students of medicine and psychology a language for suffering that is precise without being technical, and it forces them to attend to the person rather than the case. 12.2 The therapeutic imagination The poem's three part structure has been absorbed, often unconsciously, into how modern culture narrates recovery. Descent, work, emergence. Rock bottom, the hard middle, the return. Twelve step traditions, trauma treatment models, memoirs of illness and addiction, and the ordinary language of therapy all reproduce a version of this shape. The influential idea that one cannot recover from what one will not face is Dante's first structural principle, restated. The Purgatorio in particular has become a quiet resource in #mental_health contexts. Its central claims are unusually well suited to the work of recovery: that change takes time and cannot be rushed; that the pain of change is different in kind from the pain of stasis; that community and rhythm and practice matter more than insight alone; that shame must be passed through rather than avoided; and that the goal is not the erasure of the past but its reintegration. 12.3 Popular culture and the reinvention of Hell Luzzi (2024) traces how the poem has been continually reinvented, and the process has not slowed. Dante's Hell has become a permanent feature of the global imagination, detached from its theology and available to anyone. Video games send players through his circles. Films borrow his structure. Television dramas use his sins as episode titles. Novels about detectives use his punishments as a template for murders. Cartoons make jokes about the ninth circle. The phrase abandon hope has become a piece of ordinary English. This #reception_history is not a distortion to be lamented. It is evidence. What is being borrowed, again and again, is not the doctrine. It is the psychological grammar: the idea that there are gradations of moral failure, that punishment can express meaning, that the worst thing a person can do is not the loudest thing, and that the interior life has a shape which can be mapped. A culture that had no use for those ideas would not keep reaching for the map. There is also a more troubling contemporary resonance. Dante's Hell is full of people who are permanently confined to a single version of themselves, endlessly repeating their justification, seen by others only through the worst thing they ever did. It is difficult to read that in the age of the permanent digital record without recognising something. The modern public sphere has developed its own contrapasso, and it is not clear that it has developed a purgatory to go with it. The poem's most urgent lesson for the present century may be structural rather than moral: a culture that builds a Hell without also building a mountain has built only half a system, and the half it has built is the one that does not permit anybody to change. 13. Limitations, Objections, and Honest Doubts A study of this kind is obliged to state its own weaknesses clearly, and there are several. The problem of anachronism. Dante did not have a concept of the unconscious, and he would have rejected the account of the human person on which psychoanalysis rests. His psychology is Aristotelian and Thomist. Sin, for him, is a disorder of the will in relation to the good, not a return of repressed material. When this article says that #contrapasso resembles symptom formation, it is describing a structural analogy, not a historical claim. Turner (2022) and Cornish (2022) both, in different ways, insist that the theological architecture is load bearing and cannot be removed without collapsing the building. That warning is accepted here. The problem of pathologising. There is a long and unhappy tradition of diagnosing dead authors. Speculation about whether Dante had epilepsy, narcolepsy, or a depressive disorder is entertaining but evidentially thin, and it can slide into a reduction of a great work to a symptom. This article deliberately makes no claim about the poet's mental health. The problem of the biographical fallacy. As Brilli and Milani (2024) demonstrate, the life we think we know is substantially the life Dante wrote. Psychoanalytic criticism is unusually vulnerable to this trap, because it wants the author's history and will accept it from whatever source offers it. The contested status of psychoanalysis. Psychoanalytic theory is not a settled science, and many of its claims are difficult or impossible to test. A reader who rejects the theory wholesale can still accept much of what is argued here, because the arguments rest primarily on structural observation about the text, but the framing would have to be rebuilt in other terms. The problem of universality. The claim that the Comedy describes a universal human pattern should be made cautiously. The poem is a European, Christian, medieval, male authored text with a specific political agenda. Its account of women, of Islam, of heresy and of the body is not neutral, and recent scholarship, including Barolini (2022), has done important work on the poem's construction of sexual and racial others. A reading that celebrates the poem's universality without acknowledging its exclusions is not a scholarly reading; it is an advertisement. The honest position is therefore modest. The Comedy is not a psychology textbook and psychoanalysis is not a key that unlocks it. What can be said is that the poem and the theory are both serious attempts to describe the same object, and that placing them side by side makes each of them more visible. 14. Implications and Directions for Further Research For students, several practical implications follow. First, read the Purgatorio. It is the cantica that most rewards psychological attention and the one most often skipped. Any project on Dante and psychology that stops at the Inferno will reproduce the popular misunderstanding it should be correcting. Second, resist the diagnostic impulse. The interesting question is never what was wrong with Dante, but what the poem knows about how minds work. Third, read the recent scholarship before applying the theory, since a psychoanalytic reading that ignores the last five years of Dante studies will be confidently wrong in ways its author cannot see. Several research directions remain open. The relationship between the poem's treatment of gesture and the body, as opened by Webb (2022), and contemporary work on embodied affect and #trauma has barely been explored. The reception of the Comedy among prisoners, exiles and refugees, gestured at by Luzzi (2024), would repay a full psychosocial study of the kind Frosh, Vyrgioti and Walsh (2024) make possible. Comparative work on the poem alongside non European narratives of moral descent and return would test the universality claim rather than assuming it. And the use of the Comedy in clinical education, currently informal and anecdotal, could be examined empirically. 15. Conclusion Seven hundred years after his death, Dante Alighieri remains a working part of the modern mind, and the reason is not that he wrote beautifully, although he did. It is that he wrote accurately. He understood that people arrive at ruin while half asleep, and cannot afterwards explain how. He understood that the way out is not the way that looks shortest. He understood that the articulate defence of one's own conduct is not the same thing as insight, and that a person can talk perfectly for eternity and never change. He understood that the worst state is not the state of feeling too much but the state of feeling nothing about the harm one does. He understood that suffering entered voluntarily is a completely different substance from suffering endured helplessly, even when they look identical from outside. He understood that shame must be spoken to someone who both loves you and refuses to lie to you. He understood that the guide who brought you this far must eventually be given up. He understood that the final thing cannot be said, and he wrote a hundred cantos anyway. Psychoanalysis, arriving six hundred years later with a different vocabulary and a different metaphysics, described a strikingly similar terrain: the buried material that governs a life, the compulsion to repeat what cannot be remembered, the difference between #mourning and #melancholia, the necessity of a relationship in which the truth can be told, and the point at which language gives out. The convergence between the two systems does not prove that either is correct. It suggests that both were looking, honestly, at the same thing. That is the shape of Dante's enduring legacy, and it explains why students who have no interest in medieval theology still stop when they read the first line. The poem begins in a dark wood because that is where the reader is standing. It ends in a light it cannot describe because that is where the reader hopes to arrive. Between the two, it builds the most detailed account of #self_knowledge in Western literature, and it insists, against all the evidence it has assembled about human failure, that the journey can be completed. Seven centuries on, the Inferno is still eternal, and the mountain is still there to be climbed. Hashtags #Dante #Divine_Comedy #Psychoanalysis #Inferno #Purgatorio #Paradiso #Literary_Theory #Medieval_Literature #Freud #Jung #Mental_Health_Humanities #Contrapasso #Mourning_And_Melancholia #Enduring_Legacy #Dante_700 References Barolini, T. (2022). Dante's multitudes: History, philosophy, method. University of Notre Dame Press. Bianucci, R., Lanini, I., Donell, S., & Lippi, D. (2023). Pain and its management: Dante's Divine Comedy. Postgraduate Medical Journal, 99(1170), 259-264. doi:10.1136/postgradmedj-2021-140058 Brilli, E., & Milani, G. (2024). Dante's new lives: Biography and autobiography (M. Maschio & E. Plesnik, Trans.). Reaktion Books. (Original work published 2021) Callegari, D. (2022). Dante's gluttons: Food and society from the Convivio to the Comedy. Amsterdam University Press. Casadei, A. (2021). Dante oltre l'allegoria. Longo Editore. Coggeshall, E. (2023). On amista: Negotiating friendship in Dante's Italy. University of Toronto Press. Cornish, A. (2022). Believing in Dante: Truth in fiction. Cambridge University Press. DiMassa, D. (2022). Dante in Deutschland: An itinerary of Romantic myth. Bucknell University Press. Dumitrescu, M. (2022). Dante, Freud and Dali: What they might have in common? Psychoanalytic reflections on Dante's Inferno. Lumen Publishing House. Frosh, S., Vyrgioti, M., & Walsh, J. (Eds.). (2024). The Palgrave handbook of psychosocial studies. Palgrave Macmillan. doi:10.1007/978-3-031-30366-1 Gragnolati, M., Southerden, F., & Lombardi, E. (Eds.). (2021). The Oxford handbook of Dante. Oxford University Press. Khanna, R. (2024). Freud's "Mourning and melancholia". In S. Frosh, M. Vyrgioti, & J. Walsh (Eds.), The Palgrave handbook of psychosocial studies. Palgrave Macmillan. doi:10.1007/978-3-031-30366-1_39 Luzzi, J. (2024). Dante's Divine Comedy: A biography. Princeton University Press. Mandarano, P., Masciangelo, G., & Armocida, E. (2022). Dante Alighieri (1265-1321): An "infernal" endocrinological observation. Hormones, 21(1), 185-187. doi:10.1007/s42000-021-00336-9 Martin, M. R. (2023). Psychoanalysis and literary theory: An introduction. Routledge. Phillips-Robins, H. (2021). Liturgical song and practice in Dante's Commedia. University of Notre Dame Press. Turner, D. (2022). Dante the theologian. Cambridge University Press. Webb, H. (2022). Dante, artist of gesture. Oxford University Press. Wolf, L. A. (Ed.). (2022). The unexpected Dante: Perspectives on the Divine Comedy. Bucknell University Press. Classical works cited in the text by original date of publication Freud, S. (1900). The interpretation of dreams. Freud, S. (1917). Mourning and melancholia. Freud, S. (1920). Beyond the pleasure principle. Freud, S. (1923). The ego and the id. Klein, M. (1946). Notes on some schizoid mechanisms. Jung, C. G. (1959). The archetypes and the collective unconscious. Bion, W. R. (1962). Learning from experience. Lacan, J. (1966). Ecrits. Alighieri, D. (c. 1308-1321). La Commedia.

This article reads Dante Alighieri's Divine Comedy alongside Abraham Maslow's theory of human motivation. The aim is not to claim that Dante was a psychologist or that Maslow was a theologian. The aim is to show that two very different maps of the #human_person, written six hundred years apart, describe a similar movement: a person who is stuck in unmet needs, who slowly re-orders those needs, and who finally reaches a state of full functioning and self-forgetfulness. Using a comparative hermeneutic method, the study treats the three canticles of the poem as three motivational conditions. Inferno is read as the landscape of permanent deficiency, where lower needs are never satisfied and never abandoned. Purgatorio is read as the work of #growth, where deficiency needs are repaired and desire is retrained. Paradiso is read as the region of #being_needs, peak experience, and self-transcendence. The analysis attends to specific cantos, characters, and structural devices, and it compares them with Maslow's published claims, including his later and less familiar writing on transcendence. The article also takes seriously the modern critique of the hierarchy, especially recent empirical work that rejects strict sequential satisfaction. It argues that this critique actually strengthens the comparison, because Dante's own architecture is not a simple ladder either. The conclusion is that Dante gives psychology a #narrative and moral thickness that the pyramid diagram lacks, while Maslow gives contemporary readers a vocabulary that makes Dante's spiritual ascent intelligible without requiring belief. The study closes with implications for literature teaching, for interdisciplinary humanities research, and for student readers who encounter both figures in survey courses. Keywords: Dante; Divine Comedy; Abraham Maslow; hierarchy of needs; self-actualization; self-transcendence; humanistic psychology; medieval literature; motivation; peak experience. INTRODUCTION 1.1 Two Maps of the Same Country Very few works of literature are as carefully organized as the #Divine_Comedy. Dante built a universe with a floor plan. Sin has circles, penance has terraces, blessedness has spheres. Everything is counted, ranked, and placed. Readers often notice this and stop there, treating the poem as a museum of medieval opinion. But the architecture is not decoration. It is an argument about what human beings want, in what order they want it, and what happens to them when their wanting goes wrong. Six centuries later, an American psychologist proposed something with a similar shape. Abraham #Maslow argued in 1943 that human motivation is organized, that some needs press harder than others, and that a person who is starving does not spend much time worrying about self-respect. He described five families of need: physiological, safety, love and belonging, esteem, and #self_actualization. The idea travelled far beyond psychology. It entered #education and management training, nursing curricula, marketing, and school textbooks, usually in the form of a coloured triangle that Maslow himself never drew (Bridgman, Cummings, & Ballard, 2019). Put the two side by side and something interesting appears. Dante's damned are people who cannot stop chasing a lower good. His #penitents are people learning to want the right things in the right measure. His blessed are people who no longer want anything for themselves at all. Maslow's language of #deficiency_needs and #growth_needs describes that same movement in secular terms. 1.2 The Problem This Article Addresses The comparison is attractive, which is exactly why it is dangerous. It is easy to write a shallow essay that says Inferno equals unmet basic needs, Purgatorio equals esteem, Paradiso equals self-actualization, and then stop. That reading is tidy and mostly wrong. It ignores that Dante's Hell contains proud intellectuals as well as gluttons. It ignores that Purgatory begins with sins of the spirit and ends with sins of the flesh, which is the reverse of the order a naive Maslovian would predict. It ignores that Maslow himself, late in life, said the top of his own hierarchy was not the top (Maslow, 1969; Koltko-Rivera, 2006). This article therefore does something more careful. It asks: where do the two structures genuinely converge, where do they only appear to converge, and where do they contradict each other in ways that teach us something? The goal is a disciplined comparison rather than a decorative one. 1.3 Why This Comparison Matters Now Three reasons make the question timely. First, #Dante studies has had a rich few years. The seventh centenary of the poet's death in 2021 produced a wave of scholarship and public interest, including Pope Francis's apostolic letter Candor Lucis Aeternae (2021), Teodolinda Barolini's collected essays (2022), and Alison Cornish's study of truth and fiction in the poem (2022). Much of this work stresses that the Comedy is about #transformation and #conversion, not merely about doctrine. Second, Maslow's theory is being seriously re-examined rather than simply repeated. Scott Barry Kaufman (2020, 2023) has rebuilt the model around a sailboat rather than a pyramid, replacing the image of levels to be conquered with the image of a hull that must be secure and a sail that must be open. Meanwhile Rojas, Mendez, and Watkins-Fassler (2023) tested the classic assumptions on a large representative dataset and rejected the idea of strict sequential satisfaction. The theory is alive precisely because it is contested. Third, students meet both texts. Dante appears in world literature surveys. Maslow appears in every introductory psychology and education course. Very few teachers connect these two frameworks in the #classroom, and yet students intuitively feel the connection when they read Inferno 5 and recognize an addiction, or read Purgatorio 17 and recognize a theory of #motivation. 1.4 Research Questions The article is guided by four questions. RQ1. To what extent does the moral architecture of the Comedy correspond to a hierarchy of human needs? RQ2. How does Dante represent the failure of motivation, and how does that representation compare with Maslow's account of #deficiency motivation and neurosis? RQ3. Does Dante's Paradiso describe something that Maslow would recognize as #self_transcendence, and if so, what are the limits of that recognition? RQ4. What does each framework gain from being read through the other? 1.5 Structure of the Article Section 2 reviews the literature on both sides and identifies the gap. Section 3 sets out the theoretical framework and the comparative #methodology, including safeguards against anachronism. Sections 4, 5, and 6 present the analysis of the three canticles. Section 7 discusses convergences and divergences. Section 8 offers pedagogical implications. Section 9 states limitations. Section 10 concludes. LITERATURE REVIEW 2.1 Dante and the Psychological Reading of the Comedy Reading the Comedy as an account of #inner_life is not new. The medieval tradition of #allegory and its interpretation already assumed that the journey happened in the soul as well as in the cosmos. What has changed is the vocabulary. Recent scholarship has moved the emphasis away from Dante the systematic theologian toward Dante the poet of process. John Took (2020) argues that the poem is an exercise in what he calls self-confrontation, self-reconfiguration, and self-transcendence, and that Dante writes in order to provoke a first-person change in the reader rather than to transmit information. Mark Vernon (2021) reads the poem canto by canto as a guide to spiritual development, treating Hell as a description of psychological entrapment rather than a legal sentence. Teodolinda Barolini (2022) has long insisted that Dante's narrative choices are not simply dictated by theology, and that reading the poem as an unfolding process, rather than as a fixed doctrinal diagram, reveals the human tensions inside it. Alison Cornish (2022) examines how the poem asks to be believed, which matters here because the poem's claim on the reader is experiential and not merely propositional. Danielle Callegari (2022) reads food and appetite across Dante's works, which is directly relevant to any discussion of #physiological_needs in the poem, because Dante's treatment of hunger is never only about hunger. Explicitly psychological readings also exist. Stephen Haller (2024) analyses the pilgrim's change as a reorganization of a whole web of belief, arguing that the transformation does not come through argument but through experience. That claim is very close to what Maslow said about #peak_experience: that it reorganizes a person's values from the inside, and that no amount of instruction can substitute for it. 2.2 Maslow's Theory: Original Claims and Common Distortions Maslow's 1943 paper made a modest and careful claim. He proposed that needs are relatively prepotent, meaning that an unsatisfied lower need tends to dominate consciousness and organize behaviour. He explicitly said that satisfaction is a matter of degree, that people can pursue several needs at once, that the order can be reversed in particular individuals, and that a person may sacrifice everything for an ideal. Most of that nuance disappeared in transmission. Bridgman, Cummings, and Ballard (2019) traced how the theory acquired its famous triangle in management textbooks, and how the triangle imposed a rigidity that Maslow never argued for. The pyramid suggests a staircase, a scarcity of positions at the top, and a single fixed order. Maslow's own prose is far more fluid. Two later developments matter greatly for the present study. The first is the distinction between deficiency motivation and being motivation (Maslow, 1954/1970). D-needs arise from lack and are quieted by satisfaction. B-needs arise from abundance and are intensified by satisfaction. A person eating from hunger stops eating. A person contemplating beauty wants more beauty. Maslow also distinguished D-love, which uses the other person to fill a hole, from B-love, which delights in the other person's existence. This distinction will do a great deal of work in the analysis below. The second is #self_transcendence. In his later writing, Maslow argued that self-actualization was not the final term, and that the fullest human beings were oriented toward something beyond the self, whether that was a cause, a truth, a community, or the sacred (Maslow, 1964, 1969). Koltko-Rivera (2006) documented that this later revision was largely ignored, and that the textbook version of Maslow froze the theory at an earlier and more individualistic stage. Papaleontiou-Louca, Esmailnia, and Thoma (2022) offer a critical review of Maslow's treatment of spirituality, noting both his genuine interest in transcendent experience and the conceptual looseness with which he handled it. 2.3 The Contemporary Status of the Hierarchy The hierarchy is neither confirmed nor discarded. Tay and Diener (2011) found that the need categories predict well-being across cultures, but that satisfaction does not follow the strict order Maslow proposed. Rojas, Mendez, and Watkins-Fassler (2023) tested four assumptions commonly attached to the theory using a large Mexican dataset and rejected all four, including the assumption of sequential satisfaction and the assumption that the proposed order is the best path to well-being. Montag, Sindermann, Lester, and Davis (2020) linked satisfaction of each need to personality traits and affective systems, showing that individual differences shape which needs dominate. Kaufman (2020, 2023) responds to this by rebuilding rather than rejecting. His Characteristics of Self-Actualization Scale finds that Maslow's descriptive portrait of the self-actualizing person holds together as a coherent factor and predicts #well_being, creativity, and self-transcendent experience. His sailboat model replaces the ladder: security needs form the boat, growth needs form the sail, and a person moves by opening the sail rather than by climbing. That revision is important for this article, because a sailboat is much closer to Dante's structure than a pyramid is. Dante's souls do not climb by conquering levels. They climb because a weight is removed and they rise naturally, like a flame going upward. Beatrice says as much in Paradiso 1: the wonder would be if the pilgrim did not rise. 2.4 The Research Gap Comparisons between Dante and modern psychology exist, but they are dominated by psychoanalytic and Jungian readings, which focus on the unconscious, on archetypes, and on symbolic descent. Humanistic psychology has been much less used, even though it is a far more natural partner: Maslow, like Dante, is interested in the healthy person, in growth, in values, and in the peak of human possibility rather than in pathology alone. No sustained scholarly treatment maps the full Maslovian framework, including its later transcendent revision and its recent empirical critique, onto the complete architecture of the Comedy. That is the gap this article fills. THEORETICAL FRAMEWORK AND METHOD 3.1 Framework The study uses a two-part framework. The first part is the revised Maslovian model, taken not from the textbook pyramid but from Maslow's own writings and their recent reconstruction. Its key components are: the prepotency of unsatisfied needs; the distinction between deficiency and being motivation; the description of self-actualization as full functioning rather than achievement; the account of peak experience; and the late claim that transcendence lies beyond self-actualization. The second part is Dante's own psychology of #love, set out in the middle of the poem, at Purgatorio 17 and 18. Virgil explains that love is the seed of every virtue and of every sin. Natural love, he says, cannot err. Elective love can err in three ways: by aiming at a bad object, by too little vigour, or by too much. This is a motivational theory. It says that the human being is a creature of #desire, that desire is not the problem, and that the problem is always the object, the measure, or the intensity. That claim can be placed next to Maslow's without violence. Both writers assume a positive, active, forward-moving #human_nature. Both assume that dysfunction is a distortion of that movement rather than an alien force. Both refuse to explain the higher by reducing it to the lower. 3.2 Method The method is comparative hermeneutics with a structured mapping protocol. Four steps were followed. Step one: structural mapping. The three canticles, and the internal divisions of each, were tabulated against the Maslovian need categories, noting both fit and misfit. Step two: character analysis. Representative souls were selected from each region and analysed in terms of the need that dominates their speech, the object of their desire, and their relation to time. Selection favoured figures who speak at length, since motivation is disclosed through self-narration. Step three: conceptual comparison. Dante's technical terms, such as contrapasso, disordered love, and transhumanizing, were compared with Maslow's technical terms, such as prepotency, deficiency cognition, and peak experience. Step four: disconfirmation. Deliberate search for evidence against the mapping. This step is essential and is reported honestly in Section 7.2. 3.3 Guarding Against Anachronism Three safeguards were applied. First, no claim is made that Dante anticipated Maslow. Dante's framework is theological, Aristotelian, and Augustinian, rooted in #medieval_thought. He believes in an objective final good. Maslow does not require one. Second, no claim is made that Maslow's theory validates Dante's theology, or that Dante's poem provides empirical support for Maslow. Poetry is not data. Third, the comparison is used heuristically. Its value lies in what it makes visible, not in what it proves. A lens is judged by what you can see through it, not by whether it is the object itself. 3.4 Text and Citation Practice References to the poem are given by canticle and canto number rather than by page, following standard practice in Dante studies. Quotations are kept short and are given in English. The translation consulted is Kirkpatrick (2013). Dante's Italian terms are used only where the English loses something important. INFERNO: THE LANDSCAPE OF PERMANENT DEFICIENCY 4.1 The Dark Wood as Motivational Collapse The poem opens with a man of about thirty-five who has lost the path. He does not know how he entered the wood. He says he was full of sleep at the moment he left the true way. This is not the language of crime. It is the language of drift. Read psychologically, the #dark_wood is a description of what Maslow called a state in which the higher needs have gone silent. The pilgrim is not hungry. He is not in physical danger at that moment. He is not friendless. He is simply unable to move toward anything. Three beasts block him, and the third, the she-wolf, drives him back down toward the low ground. Whatever the beasts allegorize, their function is clear: they prevent ascent. The man cannot climb the hill he can see. Maslow described something similar when he wrote about people who function adequately at the lower levels but who are, in his phrase, not living up to their capacities. He argued that this condition produces a specific kind of unhappiness that has no obvious external cause and that resists external remedy. Dante's opening is the most economical portrait of that condition in Western literature. What breaks the deadlock is significant. The pilgrim does not climb by force of will. Help arrives from outside and from above, through a chain that runs from Mary to Lucia to Beatrice to Virgil (Inferno 2). Any purely self-help reading of the poem dies in Canto 2. This is the first serious divergence from a naive Maslovian reading, and it will return. 4.2 The Gate and the Neutrals: The Refusal to Want Just inside the gate, before Hell proper begins, Dante places the souls who lived without praise and without blame. They chased a banner that has no emblem, and they are stung by wasps. Virgil dismisses them with contempt. Heaven will not have them; deep Hell will not have them either. This is the poem's judgment on the absence of #motivation. In Maslow's terms, these are people in whom motivation never organized itself around anything, not even a bad object. Dante finds this state more contemptible than sin. That evaluative move is worth pausing on. Both Dante and Maslow are, in the end, more disturbed by the person who wants nothing than by the person who wants the wrong thing badly, because only the second person has a self capable of being redirected. 4.3 Limbo: Esteem Without Transcendence Limbo is one of the most Maslovian places in the poem, and one of the saddest. The virtuous pagans live in a noble castle, in a green meadow, with dignified voices. They are honoured. They have status. They speak rarely and softly, and they carry an eternal sighing that comes not from torment but from longing without hope. Everything is satisfied here except the highest thing. Physiological need, safety, belonging, and #esteem are all present. Homer, Horace, Ovid, Lucan, Aristotle, and the rest sit in a company of peers. The pilgrim is even briefly admitted to that company, and he confesses that he will not say more about the honour paid to him. It is a moment of pure esteem satisfaction, and Dante immediately moves past it. Limbo is Dante's answer to the question of what happens when a life is arranged perfectly up to the level of esteem and stops there. The answer is not misery. The answer is a permanent, quiet, dignified insufficiency. Maslow made a very similar observation when he insisted that satisfying the lower needs produces relief, not fulfilment, and that a person who has everything and still feels hollow is not malfunctioning but is simply hearing a higher need that has nowhere to go. 4.4 Incontinence: Deficiency Motivation Frozen Forever The upper circles punish sins of appetite, and they read as a catalogue of deficiency motivation that has become permanent. Lust. Paolo and Francesca (Inferno 5) are blown forever on a wind that never rests. Francesca speaks beautifully. She blames love itself, using the polished vocabulary of courtly poetry, and she never once says the word I did. Her account is a perfect specimen of what Maslow called D-love: love that is a need, that consumes, that is described as something that happened to her rather than something she chose. There is no other person in her speech. Paolo says nothing at all in the whole canto. He weeps. Two people are locked together forever and neither can reach the other. That is Dante's definition of the failure of #belonging: proximity without #communion. The pilgrim faints at the end of that canto. He faints because he recognizes himself. The poem is honest about how attractive deficiency motivation is when it dresses itself in the language of the higher needs. Gluttony. Ciacco lies in freezing filthy rain under a foul sky (Inferno 6). The sin of eating is punished by an environment in which nothing is edible and nothing is comfortable. The soul who lived for the pleasures of the table now lies in mud. Dante makes the point with the mechanism the poem calls #contrapasso: the punishment is not an external penalty added on, it is the sin seen clearly, allowed to be itself with nothing to disguise it. Callegari (2022) shows how carefully Dante thinks about food and the social body, and the point holds here: the glutton's crime is not appetite, it is the reduction of the person to appetite. Avarice and prodigality. In Canto 7 two crowds push enormous weights against each other in a semicircle, colliding, shouting, and turning back. Hoarding and squandering are the same disorder measured from opposite ends. Neither group has a face. Dante says he cannot recognize any of them, because a life spent entirely on possession erases individuality. In Maslow's language, this is what happens when the #safety_needs colonize an entire personality: money as a defence against a fear that no amount of money can settle. Wrath and sullenness. In the marsh of Styx the angry tear each other, while the sullen lie beneath the surface, gurgling a hymn they cannot finish. Two failures of the same #need. The wrathful attack the world; the sullen refuse it. Both are, in Maslow's sense, people whose growth has been swallowed by defence. 4.5 The City of Dis: Fear as a Way of Life Halfway down, the journey stalls. The devils shut the gates of Dis. The Furies appear on the tower and call for Medusa. Virgil covers Dante's eyes with his own hands and warns that if the Gorgon is seen, there will be no going up again (Inferno 9). This is the poem's most explicit statement of the danger of the deficiency world. Medusa turns the viewer to stone. To look directly at a certain kind of despair is to be fixed in it permanently. Reason, in the figure of Virgil, is not enough here either. The gates only open when a heavenly messenger arrives and opens them with a touch. Psychologically, Dis represents the point at which the deficiency needs stop being needs and become an identity. Above the walls, souls are still recognizably people who wanted things too much. Below the walls, sin becomes deliberate, constructed, and defended. 4.6 Esteem Turned Inward: Farinata, Brunetto, Ulysses Dante's Hell is not merely a warehouse of appetites. Its most memorable inhabitants are people destroyed by their pursuit of #esteem. Farinata rises from a burning tomb, chest and brow lifted, as if he held Hell in great contempt (Inferno 10). His first question to Dante is about ancestry. He is standing in a grave that is on fire and he wants to talk about political faction. Beside him, Cavalcante asks about his son and, on a misunderstanding, collapses back into the tomb. The contrast is exact: one man is destroyed by his indifference to relationship, the other by his relationship alone, and neither can help the other, though they share a tomb. Brunetto Latini runs across burning sand and asks Dante to remember his book, his Tesoro, in which, he says, he still lives (Inferno 15). This is a devastating line. A man in Hell defines his survival by his literary reputation. Maslow distinguished between esteem that comes from real competence and esteem that depends on the opinion of others, and he considered the second unstable and, in the long run, corrosive. Brunetto is that instability made eternal. Ulysses is the most difficult case (Inferno 26). He is wrapped in a flame that speaks. He recounts how he persuaded his old crew to sail past the boundary of the known world with an argument about human dignity: you were not made to live like brutes, but to pursue virtue and knowledge. It is one of the most stirring speeches in the poem. It is also, in Dante's judgment, a fraudulent counsel that drowned every man who heard it. Why is this speech condemned? Not because knowledge is bad. Dante's whole poem is a pursuit of knowledge. The problem is that Ulysses uses the vocabulary of the highest need to serve a lower one. He does not sail for truth. He sails because he cannot bear to stop, and he takes other people with him. His voyage is called mad flight. Compare Maslow's own warnings about the imitation of self-actualization: the danger of a person who adopts the language of growth as a cover for ego, and who mistakes restlessness for transcendence. Ulysses is the patron saint of that error. For students, Ulysses is the single most useful figure in the poem, because he shows that self-actualization language can be counterfeited, and that the counterfeit is more attractive than the real thing. 4.7 The Frozen Centre At the bottom, everything is ice. Cocytus holds the treacherous, those who broke the bonds of trust that make any community possible. Belonging is not merely failed here, it is weaponized. Ugolino gnaws the skull of the man who starved him and his children (Inferno 33). His story is the most extreme collision of the lowest and highest needs in the poem. He tells of hunger, of children offering him their own flesh, of biting his hands in grief, of the children misreading the gesture. Dante leaves an unresolved ambiguity in the final line about whether hunger did what grief could not. The whole passage is an argument that the physiological floor of human life is not separable from love, dignity, and meaning. Extreme #deficiency_needs do not simply suspend the higher needs. They corrupt them. At the very centre stands Lucifer, waist-deep in ice, weeping from six eyes, beating six wings. The wings are the point. He beats them to escape, and the wind they generate is what freezes the lake that holds him. This is the most precise image of self-defeating motivation ever produced. The harder he strives, the more firmly he is fixed. He does not speak. Motivation, in the Comedy, is finally measured by the capacity to say something to another person. Lucifer has lost it. 4.8 Summary of the Inferno Analysis Hell is not the place where lower needs are punished. Hell is the place where needs have stopped being a hierarchy at all. Nothing is prepotent because nothing can be satisfied. Time is abolished, so no progress is possible; the damned can see the distant future and are blind to the present (Inferno 10), which is exactly backwards for a creature that grows. In Maslow's terms, Hell is total, permanent deficiency cognition. The souls see everything in terms of what they lack and what they can use. They are, in Dante's own image, souls who have lost the good of the intellect. PURGATORIO: THE WORK OF GROWTH 5.1 A Different Kind of Mountain The tone changes completely on the shore of the mountain. Dante speaks of the sweet colour of the eastern sapphire, of a little boat of the mind hoisting sail. Cato sends the travellers down to the water's edge, where Virgil washes the grime of Hell from the pilgrim's face with dew and girds him with a plain rush. The rush is chosen because it bends when the wave strikes and grows back when it is plucked. That rush is the emblem of the whole canticle. Purgatorio is about flexibility, #habit, repetition, and time. It is the only region of the poem where the sun rises and sets, where souls get tired, where the pilgrim sleeps and dreams. It is the region of process. 5.2 Ante-Purgatory: The Psychology of Waiting Before anyone can begin, they must wait. The excommunicated, the late repentant, the negligent rulers, and the violently killed all spend proportional time on the lower slopes. Manfred, cut down and buried outside the church, tells Dante calmly that no one is so lost that eternal love cannot come back for them. Belacqua sits behind a boulder hugging his knees, too indolent even to look up, and jokes at Dante's expense (Purgatorio 4). He is not despairing. He is simply not ready. The poem does not punish him with rage. It gives him time. This part of the mountain is the poem's account of #safety_needs done properly. A person who has just come out of chaos cannot begin advanced work immediately. There is a period in which nothing dramatic happens and the only task is to become settled enough to begin. Kaufman's sailboat is exactly this: before the sail can be opened, the hull must not be leaking. Ante-Purgatory is hull repair. 5.3 The Gate and the Seven Wounds At the gate (Purgatorio 9) there are three steps: one of polished mirror-white marble, one of dark cracked stone, one of blood-red porphyry. An angel cuts seven P's into the pilgrim's forehead with the point of a sword and tells him to wash these wounds when he is inside. The steps have been read as confession, contrition, and satisfaction. Psychologically they can also be read as a sequence any process of change requires: honest #self_knowledge, acknowledgement of what is broken, and the costly commitment to act. The mirror shows you yourself. The cracked stone is the fracture you must admit. The red step is what it will cost. The wounds are on the forehead. They must be washed one by one, and each is removed by climbing. 5.4 The Order of the Terraces: A Problem for the Pyramid Here is where a naive Maslovian reading breaks down, and where the comparison becomes genuinely interesting. If Purgatory were a Maslovian pyramid, it would begin with the correction of physiological disorders and rise to the correction of spiritual ones. It does exactly the opposite. The mountain begins with pride, envy, and wrath, which are sins of the spirit, and it ends with avarice, gluttony, and lust, which are sins of the flesh. The heaviest weight is at the bottom, and the heaviest weight is pride. Dante is telling us that the deepest disorder in a person is not appetite but ego, and that appetite cannot be healed until ego is. That is not what the textbook pyramid predicts. It is, however, close to what Maslow actually argued in his more careful moments: that the higher needs, once they are active, can reorganize the lower ones, and that a person who has found meaning can bear an astonishing amount of physical deprivation. Empirical work supports Dante here more than it supports the pyramid. Rojas, Mendez, and Watkins-Fassler (2023) found that the sequential path Maslow proposed is not the best path, and that satisfaction of love, belonging, and esteem needs contributed most to well-being in their sample. Dante would not have been surprised. 5.5 The Terraces One by One Pride. The #proud carry stones on their backs and walk bent double, their eyes forced toward the pavement (Purgatorio 10 to 12). What is carved on that pavement is a gallery of the fallen proud. They cannot look up, so they are made to read. Oderisi da Gubbio, an illuminator of manuscripts, gives the great speech on fame: worldly renown is a breath of wind that changes direction and changes name. Cimabue thought he held the field; now Giotto has the acclaim. This is the direct counterweight to Brunetto Latini in Hell. Both men cared about reputation. One is still defending it in a burning desert; the other is learning to let it go while carrying a rock. The difference is the entire difference between deficiency esteem and being esteem. Maslow's self-actualizing person, in his description, is relatively independent of the good opinion of others, not because they are cold but because their sense of worth is no longer negotiable. Envy. The envious sit against a bare cliff in coarse grey cloaks, leaning on each other's shoulders, their eyelids stitched shut with iron wire (Purgatorio 13). They are blind, and so they must depend on one another. Sapia of Siena confesses that she prayed for the defeat of her own city and rejoiced when her fellow citizens were slaughtered. #Envy, in Maslow's framework, is a disease of comparison, and comparison is what happens when esteem is sourced entirely from outside. Dante's cure is exact and beautiful: he takes away sight, which is the organ of comparison, and forces the envious to rely on the very neighbours they resented. The therapy for a broken belonging need is enforced #belonging. Wrath. The wrathful walk in acrid smoke, unable to see anything at all (Purgatorio 15 to 16). In that darkness Marco Lombardo delivers the poem's central defence of #free_will: if everything were determined, then joy and mourning would make no sense. He tells Dante that the heavens begin your movements but that a light is given you to know good and evil, and free will, which, if it endures the strain of its first battles with the stars, wins everything in the end. This is the moral centre of the poem and it is placed in the darkest place on the mountain. Dante is saying that the doctrine of human freedom is most necessary exactly where people cannot see. Sloth. #Sloth is defined here as defective love, love that is too slow. Virgil's discourse on love (Purgatorio 17) sits at the exact structural middle of the entire Comedy: fifty cantos before, fifty after. This placement is not accidental. Dante puts a theory of motivation at the geometric heart of his poem. The slothful run, shouting examples of urgency. Their penance is speed. In Maslow's language, sloth is the failure of the growth impulse itself, the condition in which a person knows the good and does not move toward it. It is the sin of the person in the dark wood at the very start of the poem, which is why the pilgrim falls asleep on this terrace and dreams. Avarice. The avaricious lie face down on the ground, unable to move, hands and feet bound, weeping. They loved earth, so they are given earth and nothing else. Pope Adrian V explains that his conversion came only when he reached the highest office and discovered that the life there did not satisfy (Purgatorio 19). This is a first-person account of the failure of achievement to produce fulfilment. Statius, freed on this terrace, causes the mountain to tremble and every soul to cry out. His liberation is public, and everyone celebrates it. Growth in Purgatory is never a private accomplishment. The mountain shakes for you. Gluttony. The gluttons are emaciated, their faces so hollow that Dante can read the word OMO in the bones of their eye sockets (Purgatorio 23 to 24). They walk past trees loaded with fruit and streams of falling water, and they cannot reach either. Their old friend Forese Donati speaks with the pilgrim about wives, about Florence, about poetry. The conversation is warm, funny, and full of memory. Two men who once wrote insulting poems at each other are now walking together, one starving, both happy. Dante's point is severe and hopeful at once. The appetite is not destroyed. It is stretched until it can hold something larger. Forese says the same hunger that once ruled him now leads him toward the tree. Lust. The last terrace is a wall of flame that runs along the mountainside, and the penitents walk inside it, singing and calling out examples of chastity, embracing one another briefly as they pass in opposite directions (Purgatorio 25 to 26). Guido Guinizzelli, whom Dante calls the father of his poetic style, is here. So is Arnaut Daniel, who speaks in Provencal and asks Dante to remember his pain in due time. The wall of fire is the last obstacle, and Dante will not walk through it. He is terrified. Virgil argues, reasons, and reassures, and none of it works. Then Virgil says one thing: Beatrice is on the other side. Dante walks into the fire immediately (Purgatorio 27). That moment is the strongest single piece of evidence for the argument of this article. Reason cannot move him. Fear cannot move him. A higher #love moves him instantly. Motivation, for Dante, is finally hierarchical in the only sense that matters: the strongest motive is the highest one, once it becomes real. 5.6 Autonomy: The Crown and the Mitre On the far side of the fire, Virgil speaks for the last time. He tells Dante that he has brought him as far as reason and art can go, that the pilgrim's will is now free, upright, and whole, and that it would be wrong not to act according to it. Then he says: over yourself, I crown and mitre you. Virgil crowns him king and bishop of himself. He is now his own temporal and spiritual authority, a figure of complete #autonomy. This is the exact moment of #self_actualization in the poem, and it is a remarkably precise match. Maslow described the self-actualizing person as autonomous, self-directed, resistant to enculturation, and governed by inner rather than outer criteria. Deci and Ryan's later work on autonomy as a basic psychological need (Ryan & Deci, 2020) describes the same condition in different vocabulary: behaviour that is fully endorsed by the self. Note what has happened structurally. The seven P's are gone. The deficiency needs have been repaired. Only now is the person released into self-government. Maslow's central claim, that growth becomes possible when deficiency stops shouting, is dramatized here in narrative form. Note also what Virgil does not say. He does not say that Dante has arrived. He says that Dante can now be trusted to walk without a guide, and then Virgil disappears. Self-actualization is the end of the therapy, not the end of the journey. 5.7 The Earthly Paradise and the Cost of Being Seen At the top of the mountain, in a forest full of birdsong, Matelda walks through the flowers. Then the great procession appears, and Beatrice arrives, veiled, in a chariot. Dante turns to tell Virgil that he is trembling, and Virgil is gone. Beatrice does not comfort him. She names him, sharply, the only time his name appears in the poem, and she prosecutes him. She accuses him of turning after false images of good after her death. Dante stands like a child being scolded. He tries to speak and cannot. He confesses, weeps, and faints. This is one of the most psychologically shocking sequences in literature, and it is essential to the argument. Dante places a total, humiliating collapse immediately after the moment of crowned autonomy. The person who has just been declared free and whole is broken open by being seen by someone who loves him and knows what he did. Any account of self-actualization that treats it as a stable achievement, a personal summit, or a state of settled self-esteem, cannot accommodate this scene. Maslow's own descriptions, at their best, can. He insisted that self-actualizing people accept themselves and others without defensiveness, that they can bear to know unpleasant truths about themselves, and that growth is often preceded by the willingness to be exposed. Dante is then drawn through Lethe, which removes the memory of sin, and later through Eunoe, which restores the memory of every good he has done. This double bath is worth noticing. It is not enough to forget the bad. He must also remember the good, or the self that emerges will be too thin to carry what comes next. He leaves the mountain, in his own words, remade and ready to rise to the stars. PARADISO: BEING NEEDS AND SELF-TRANSCENDENCE 6.1 Beyond the Top of the Pyramid Most popular accounts of Maslow stop at self-actualization. Maslow did not. In his later work he wrote that self-actualization is not enough, that the fully developed person is oriented beyond the self, and that the highest human experiences are transpersonal (Maslow, 1964, 1969). Koltko-Rivera (2006) argued that ignoring this later revision produced a distorted, individualistic Maslow who never existed. If we restore that revision, then the structure of Maslow's mature theory has four stages: deficiency, growth, self-actualization, and #self_transcendence. And the structure of the Comedy has four stages: Hell, Purgatory, the Earthly Paradise, and Heaven. That is a striking parallel, and it becomes more striking when one notices that in both systems the fourth stage cannot be reached by working harder at the third. 6.2 Transhumanizing Paradiso begins with Dante looking at Beatrice while she looks at the sun. He finds himself rising, and he does not know how. He coins a word for what is happening to him: trasumanar, to pass beyond the human. He immediately says the word cannot be explained (Paradiso 1). Beatrice tells him not to be surprised that he is rising, and explains that all things have an inclination toward their proper place, that fire rises, and that the surprise would be if he had stayed down. #Rising, here, is not achievement. It is nature, once the obstacle is removed. This is the deepest structural point of the comparison. Purgatory is effortful; Paradise is not. The climbing is over. What remains is a kind of falling upward. Maslow's B-motivation has exactly this quality: it is not a striving to fill a lack but an unfolding of what is already there. He wrote that a musician must make music and a poet must write, in the same way that a tree must grow. 6.3 The Non-Competitive Hierarchy The single most important theological move in Paradiso is also its most important #psychological one. In the sphere of the Moon, Dante meets Piccarda Donati, whose vows were broken by force. He asks her, with what he thinks is tact, whether she wishes she were higher up. She smiles, and the smile is described as burning with the first fire of love. Then she tells him that if the souls here desired to be higher, their desires would be out of tune with the will of the one who assigns them here, and that charity does not allow such a thing. And then she says the line: in His will is our peace (Paradiso 3). Dante understands immediately that every place in Heaven is Paradise, though grace does not rain down in one measure. This is a hierarchy without competition. Ranks exist. Comparison does not. Nobody wants what somebody else has, and #envy has no foothold. Beatrice explains shortly afterwards (Paradiso 4) that the souls do not actually live in the spheres at all. They all dwell in the Empyrean. The spheres are a pedagogical arrangement, staged for Dante's benefit, because a human mind can only learn through what it can see. The implications for the pyramid are considerable. Dante's hierarchy of the heavens is a hierarchy of capacity, not of privilege, and it exists for the learner, not for the ranked. Every soul is fully satisfied at the level of its capacity. This dissolves precisely the problem that critics have always raised against Maslow's diagram: that it implies scarcity at the top, that it implies self-actualization is a competitive achievement available to few, and that it flatters the successful. If one had to name the single thing Dante can teach modern motivational psychology, this would be it. The highest state of human motivation is one in which #hierarchy no longer produces envy. 6.4 The Being Values Made Visible Maslow described the being values, or B-values, that self-actualizing people report at the peak of experience: truth, goodness, beauty, wholeness, aliveness, uniqueness, justice, simplicity, effortlessness, playfulness, and self-sufficiency. Reading Paradiso against that list is uncanny. Truth is examined explicitly. In the sphere of the fixed stars, Dante is questioned on faith by Saint Peter, on hope by Saint James, on love by Saint John, and he must give reasoned answers like a graduate student in an oral examination (Paradiso 24 to 26). The pilgrim who fainted in Hell now argues. Justice appears in the sphere of Jupiter, where the souls literally spell out words of scripture in the sky and then form an eagle that speaks with one voice while saying I and mine, not we and ours (Paradiso 19). A community of individuals who speak as a single person without losing their individuality is Maslow's description of a healthy synergic culture rendered as literal spectacle. Beauty is Beatrice's smile, which grows more intense at every level and which Dante repeatedly says he cannot describe. Contemplation and playfulness appear in Saturn, where the contemplatives ascend and descend on a golden ladder that reaches beyond sight, and where, alone in all of Heaven, there is silence and no singing, because the music at that height would be unbearable (Paradiso 21). Aliveness, wholeness, and effortlessness are the persistent qualities of the whole canticle. Nothing in Paradiso is a struggle. 6.5 Peak Experience and Its Aftermath Maslow's account of peak experience has four features that matter here: disorientation of time and space; the loss of self-consciousness; the sense that reality is being seen as it is rather than as it is useful; and the failure of language afterward. Every one of these is in the final cantos. In the Empyrean, Dante first sees a river of light with sparks flying out of it and living flowers on its banks. He is told this is a shadow of the truth, and instructed to drink from the river with his eyes. When he does, the river changes shape and becomes round: the Celestial Rose, with the whole company of the blessed seated in its petals (Paradiso 30). His guide changes for the last time. Beatrice takes her seat in the rose, and Saint Bernard takes over. Dante looks up and sees her, far away, and she smiles at him, and turns back to the eternal fountain. He does not follow her. He lets her go. This is the last and hardest renunciation in the poem, and it is done without grief. Bernard prays to the Virgin. The prayer is a masterpiece of the poem, and its logic is that a human being cannot get the last inch of the way on their own strength. Then Dante looks. He describes his eyes becoming clearer as he looks. He sees, bound by love into a single volume, everything that is scattered through the universe. He sees three circles of three colours and one dimension. He tries to understand how a human face fits inside one of them, and his mind cannot do it, and then, in his own words, a flash of understanding strikes and his desire and his will are turned, like a wheel that is moved evenly, by the love that moves the sun and the other stars. The last line of the poem is about motivation. It says the will has become a wheel turning in perfect balance because it is now moved by love. That is not the end of desire. It is the perfection of #desire. 6.6 Ineffability and the Honesty of the Poem Dante spends a great deal of Paradiso saying that he cannot say what he saw. He compares himself to a geometer who cannot square the circle. He says his memory of the vision is like snow melting in the sun, or like the leaves of the Sibyl's prophecy scattered by wind. Maslow was similarly frank. He noted that peak experiences resist verbal report, that people who have them often become inarticulate or resort to poetry and religion, and that this is a property of the experience rather than a defect in the person. The convergence is not superficial. Both writers face the same problem: the highest state of the human mind is exactly the state that language handles worst. Dante's solution is to make the failure of language part of the poem. Yaden and Newberg (2022) survey the modern research literature on self-transcendent experience and note the same recurring feature of ineffability across very different traditions and contexts. DISCUSSION 7.1 Where the Two Frameworks Genuinely Converge Six convergences survive scrutiny. First, both are theories of #desire rather than theories of behaviour. Dante says love is the seed of every action, good or bad. Maslow says needs organize consciousness. Neither reduces the person to conditioning. Second, both hold that human nature is fundamentally positive and forward-moving, and that dysfunction is distortion rather than essence. Dante's damned are not made of a different substance from the blessed. Maslow's neurotics are not a different species from the self-actualizers. Third, both distinguish between motivation from lack and motivation from fullness. Dante's whole moral system depends on whether a love aims at a real good, and in what measure. Maslow's D and B distinction says the same thing with different words. Fourth, both understand that the higher needs are quieter than the lower ones, and are easily silenced. Limbo is Maslow's point about the higher needs being weak, subtle, and easily drowned out, dramatized as an eternal sigh. Fifth, both recognize that reaching the top is not an achievement, a possession, or a status. Piccarda's smile and Maslow's insistence that self-actualization is a process rather than a trophy are saying similar things. Sixth, and most importantly, both end beyond the self. Dante ends with a will that has become transparent to a love larger than itself. Maslow, in his last years, said that self-actualization by itself curdles into selfishness and that the fully human person must be pointed at something beyond. 7.2 Where They Genuinely Diverge Four divergences must be stated plainly, because ignoring them would falsify the comparison. Grace. This is the deep one. Dante's pilgrim cannot start, cannot pass the gates of Dis, and cannot complete the final vision under his own power. Help comes from outside at every critical juncture. Maslow's framework has no equivalent. It has facilitating conditions, good parenting, safety, and a decent society, but the movement is finally the organism's own. A Maslovian reading of the Comedy must either translate #grace into something like unearned relational support, which loses most of what Dante means, or admit that the poem contains a category the psychology does not have. Objectivity of the good. Dante believes there is one final object of all desire, and that a person is well or badly ordered depending on their relation to it. Maslow is descriptive. He observed what healthy people seemed to value and reported it. He did not, and could not, argue that those values are true. This is a real philosophical difference and not a matter of emphasis. Papaleontiou-Louca and colleagues (2022) note precisely this conceptual looseness in Maslow's handling of spirituality: he wanted the fruits of transcendence without commitment to the metaphysics that traditionally grounded them. The community. Purgatory is collective. The souls climb together, sing together, pray for each other, and the whole mountain shakes when one soul is freed. Heaven is a rose, not a summit. Maslow's model, especially in its popular pyramid form, is individual. One person climbs one pyramid. The recent literature has been pushing back on this individualism (Bland & DeRobertis, 2020; Kaufman, 2023), but the pyramid image remains stubbornly solitary. Judgment. Dante's Hell is permanent. Maslow's psychology has no permanent damnation; every person retains the capacity for growth. Readers should not soften this. The poem's power depends on the possibility of a soul that will never change, and no humanistic psychology can accept that possibility. 7.3 What the Empirical Critique Does to the Comparison One might expect that recent findings against the strict hierarchy would weaken this article's argument. They do the opposite. Rojas and colleagues (2023) rejected sequential satisfaction. Dante's Purgatory also rejects it, by curing pride before it cures gluttony. Tay and Diener (2011) found that needs are universal but not strictly ordered, and that social and esteem needs contribute strongly to well-being even when basic needs are unmet. Dante's Ugolino is precisely that finding turned into tragedy: a man starving to death whose real destruction is relational. Montag and colleagues (2020) found that individual differences shape which need dominates. Dante's Hell is nothing if not a catalogue of individual differences in dominant need, which is why every damned soul gets a different circle and a different speech. In short, the pyramid does not survive contact with either the data or the poem. What survives, in both, is the more modest and more interesting claim: that human beings are organized by desire, that some desires can eclipse others, and that a life goes well or badly depending on whether that organization is healthy. 7.4 What Dante Offers Psychology Three things. First, narrative time. A diagram has no time in it. A person's motivational life is a story, with relapses, plateaus, sudden reversals, and long stretches of nothing much. Purgatorio is the only major model of change in Western literature that takes boredom seriously. Second, the moral density of motivation. Psychology tends to treat needs as morally neutral. Dante insists that motivation always has an object, and that the object is not neutral. Ulysses and Beatrice are both objects of a strong desire. They are not the same. Third, the counterfeit problem. Dante shows, in Ulysses, that the language of the highest need can be used in the service of the lowest, and that the counterfeit is nearly indistinguishable from the real thing. Any psychology of #self_actualization that lacks a theory of its own imitation is naive. 7.5 What Maslow Offers Readers of Dante Also three things. First, an accessible entry. A student who cannot use the word sin, and who has no theological formation, can still recognize Francesca as a person in a compulsive relationship, Ciacco as an addict, and Brunetto as someone who mistook reputation for a self. The psychological vocabulary is a #door_into_the_poem. Second, a defence against moralism. Reading Hell as a punishment catalogue makes Dante seem vindictive. Reading it as a description of what unresolved deficiency motivation actually feels like from the inside makes him seem observant. Vernon (2021) makes exactly this move, and it changes the tone of the poem. Third, a way to take Paradiso seriously. Paradiso is the least read of the three canticles, partly because modern readers have no category for sustained joy. The literature on peak and self-transcendent experience (Yaden & Newberg, 2022; Kaufman, 2023) restores that category, and it makes Paradiso legible as a report rather than as decoration. IMPLICATIONS FOR STUDENTS AND TEACHING 8.1 A Practical Reading Framework Students approaching the Comedy for the first time can use four questions at every encounter with a soul. What does this person want? What are they willing to trade for it? What do they say about themselves, and what does the poem show that contradicts it? And what has their wanting done to their capacity to see other people? Applied consistently, these questions turn the poem from a list of names into a study of #moral_psychology. Francesca wants to be innocent. Farinata wants his party's honour. Ulysses wants the horizon. Piccarda wants nothing. That last one is the hardest for a student to sit with, and it is the whole point. 8.2 Cautions for Interdisciplinary Work Three cautions are worth teaching alongside the comparison. Do not flatten Dante into a self-help author. The poem contains judgment, hierarchy, doctrine, and politics, and a reading that removes them produces a much smaller poem. Do not treat the hierarchy of needs as settled science. It is a useful heuristic with weak empirical support in its strong form, and students should know the difference between a model that organizes thinking and a model that has been tested. Do not assume that the parallel proves anything about either. It is an interpretive #lens. Its output is insight, not evidence. 8.3 Assignments That Work Three tasks have proved useful in classroom settings and are offered here as suggestions. Ask students to write the missing terrace: invent an eighth terrace of Purgatory for a modern disordered love, and specify the sin, the penance, the examples displayed, and the beatitude sung at the exit. The exercise forces them to internalize the logic of #contrapasso as therapy rather than revenge. Ask students to rewrite Francesca's speech in the first person, honestly. The difficulty they encounter is the lesson. Ask students to argue, in writing, against the thesis of this article: that Dante and Maslow are not compatible and that the comparison distorts both. The best students find the grace problem on their own. LIMITATIONS This study has four limitations that should be stated openly. It is interpretive, not empirical. No claim of causal or historical influence is made or could be made. It is selective. The Comedy contains more than a hundred named souls, and only a small number are analysed here. A different selection might weight the argument differently, particularly a selection that emphasized the political cantos, which resist psychological reading almost entirely. It reads the poem in translation for the purposes of exposition, and some of Dante's most compressed motivational language, especially in Purgatorio 17 and Paradiso 33, resists English. It engages one psychological tradition. A Jungian, cognitive, or self-determination reading would produce different and possibly better results. Ryan and Deci's account of autonomy, competence, and relatedness (2020), in particular, deserves a full treatment against Purgatorio that this article has only gestured at. CONCLUSION Dante wrote a poem about a man who was stuck, who was helped, who worked hard, who was crowned over himself, who was then broken open by love, and who finally saw something that made his own will turn as smoothly as a wheel. Maslow spent his career trying to describe the difference between people who are merely not sick and people who are fully alive. He began with a hierarchy of needs, and he ended by saying the hierarchy was not enough, and that the fullest human beings were the ones who had gotten past themselves. Neither man completed his project. Maslow died with the transcendence work unfinished, and the world remembered him through a triangle he never drew. Dante died in exile, and the poem that made him immortal was written by a man who never went home. But the two of them, from opposite ends of Western history, kept insisting on the same unfashionable claim: that a human being is not a bundle of appetites to be managed, but a creature made for something, and that everything depends on what that something turns out to be, which is the whole question of #meaning. The pyramid is the wrong picture. So, in the end, is the mountain. The right picture is the one Dante finds at the top, when he stops climbing and finds that he is already being carried: a rose in which every seat is the best seat, where the hierarchy is real and nobody envies anyone, and where the last human question, what do I want, has finally become the same as the question, what is worth wanting. That is the hierarchy of the heavens. It is not a ladder. It is an #ascent that turns out, at the last moment, to have been a homecoming. HASHTAGS #Dante #Divine_Comedy #Maslow #Hierarchy_of_Needs #Self_Actualization #Self_Transcendence #Humanistic_Psychology #Medieval_Literature #Inferno #Purgatorio #Paradiso #Psychology_of_Motivation #Literature_and_Psychology #Peak_Experience #Spiritual_Ascent REFERENCES Alighieri, D. (2013). The Divine Comedy (R. Kirkpatrick, Trans.). London: Penguin Classics. Barolini, T. (2022). Dante's Multitudes: History, Philosophy, Method. Notre Dame, IN: University of Notre Dame Press. Bland, A. M., & DeRobertis, E. M. (2020). Maslow's unacknowledged contributions to developmental psychology. Journal of Humanistic Psychology, 60(6), 934-958. doi:10.1177/0022167817739732 Bridgman, T., Cummings, S., & Ballard, J. (2019). Who built Maslow's pyramid? A history of the creation of management studies' most famous symbol and its implications for management education. Academy of Management Learning and Education, 18(1), 81-98. doi:10.5465/amle.2017.0351 Callegari, D. (2022). Dante's Gluttons: Food and Society from the Convivio to the Comedy. Amsterdam: Amsterdam University Press. Cornish, A. (2022). Believing in Dante: Truth in Fiction. Cambridge: Cambridge University Press. Francis, Pope. (2021). Candor Lucis Aeternae: Apostolic Letter on the Seventh Centenary of the Death of Dante Alighieri. Vatican City: Libreria Editrice Vaticana. Haller, S. F. (2024). Dante's pilgrimage: Transformation as a web of belief. International Journal of Religious Tourism and Pilgrimage, 12(2), Article 9. Kaufman, S. B. (2020). Transcend: The New Science of Self-Actualization. New York: TarcherPerigee. Kaufman, S. B. (2023). Self-actualizing people in the 21st century: Integration with contemporary theory and research on personality and well-being. Journal of Humanistic Psychology, 63(1), 51-83. doi:10.1177/0022167818809187 Koltko-Rivera, M. E. (2006). Rediscovering the later version of Maslow's hierarchy of needs: Self-transcendence and opportunities for theory, research, and unification. Review of General Psychology, 10(4), 302-317. doi:10.1037/1089-2680.10.4.302 Maslow, A. H. (1943). A theory of human motivation. Psychological Review, 50(4), 370-396. doi:10.1037/h0054346 Maslow, A. H. (1964). Religions, Values, and Peak-Experiences. Columbus, OH: Ohio State University Press. Maslow, A. H. (1969). The farther reaches of human nature. Journal of Transpersonal Psychology, 1(1), 1-9. Maslow, A. H. (1970). Motivation and Personality (2nd ed.). New York: Harper and Row. (Original work published 1954) Montag, C., Sindermann, C., Lester, D., & Davis, K. L. (2020). Linking individual differences in satisfaction with each of Maslow's needs to the Big Five personality traits and Panksepp's primary emotional systems. Heliyon, 6(7), e04325. doi:10.1016/j.heliyon.2020.e04325 Papaleontiou-Louca, E., Esmailnia, S., & Thoma, N. (2022). A critical review of Maslow's theory of spirituality. Journal of Spirituality in Mental Health, 24(4), 327-343. doi:10.1080/19349637.2021.1932694 Rojas, M., Mendez, A., & Watkins-Fassler, K. (2023). The hierarchy of needs empirical examination of Maslow's theory and lessons for development. World Development, 165, 106185. doi:10.1016/j.worlddev.2023.106185 Ryan, R. M., & Deci, E. L. (2020). Intrinsic and extrinsic motivation from a self-determination theory perspective: Definitions, theory, practices, and future directions. Contemporary Educational Psychology, 61, 101860. doi:10.1016/j.cedpsych.2020.101860 Tay, L., & Diener, E. (2011). Needs and subjective well-being around the world. Journal of Personality and Social Psychology, 101(2), 354-365. doi:10.1037/a0023779 Took, J. (2020). Why Dante Matters: An Intelligent Person's Guide. London: Bloomsbury Continuum. Vernon, M. (2021). Dante's Divine Comedy: A Guide for the Spiritual Journey. Brooklyn, NY: Angelico Press. Wolf, L. A. (Ed.). (2022). The Unexpected Dante: Perspectives on the Divine Comedy. Lewisburg, PA: Bucknell University Press. Yaden, D. B., & Newberg, A. (2022). The Varieties of Spiritual Experience: 21st Century Research and Perspectives. New York: Oxford University Press.

Across low and middle income countries, the daily work of clinics, health posts, and community health programmes is shaped less by local decision making than by the reporting requirements of distant funders. This article develops a sociological account of that process using the theory of #institutional_isomorphism. It argues that international non governmental organisations, acting as intermediaries between donors and communities, transmit a set of funding mandates that push very different health systems toward a common organisational and clinical form. The article identifies three transmission channels. Coercive pressure operates through grant conditions, logical frameworks, indicator targets, audits, and the constant threat of defunding. Mimetic pressure operates through uncertainty, which encourages local organisations to copy whatever model appears to be winning grants. Normative pressure operates through training curricula, professional networks, guideline authorship, and the shared vocabulary of the global health field. The result is a striking convergence in protocols, registers, job descriptions, and even in the phrasing that a health worker uses when counselling a patient. The article evaluates both the gains of this convergence, which include safer prescribing, clearer referral pathways, comparability of data, and rapid scale up of proven interventions, and its costs, which include ceremonial compliance, indicator gaming, the crowding out of locally generated knowledge, workforce distortion, vertical fragmentation, and abrupt collapse when donor cycles end. Drawing on the concepts of decoupling, epistemic injustice, and audit culture, the article proposes a framework of reflexive standardisation, in which standards are treated as revisable working agreements rather than as fixed moral facts. Practical implications are set out for donors, ministries of health, international NGOs, local organisations, and researchers. The article closes with a research agenda that calls for comparative, longitudinal, and community centred study of how standards are received, resisted, and reworked at the point of care. Keywords: institutional isomorphism, global health governance, non governmental organisations, clinical standardisation, donor conditionality, audit culture, decoupling, health systems, decolonising global health, evidence based medicine 1. Introduction Walk into a rural health facility in almost any country that receives significant development assistance for health, and you will notice something that ought to be surprising but rarely is. The wall charts look familiar. The registers look familiar. The counselling scripts, the flow of patients from screening to testing to treatment, the coloured cards used to track defaulters, the monthly reporting form, and even the motivational posters share a family resemblance with facilities thousands of kilometres away, in countries with different languages, different disease burdens, different professional traditions, and radically different histories. This resemblance is not an accident, and it is not simply the natural convergence of independent actors on the one best clinical method. It is the visible trace of a governance relationship. Much of the money that keeps those facilities open, that pays the community health workers, that supplies the test kits, and that funds the supervision visits, arrives through international non governmental organisations, which in turn receive it from bilateral donors, multilateral funds, and private philanthropies. That money arrives with #funding_mandates attached: a grant agreement, a results framework, a set of indicators, a reporting calendar, an approved protocol, an audit clause. Those mandates travel down the chain and settle, eventually, in the consulting room. The core claim of this article is that #international_NGOs function as engines of organisational convergence in health, and that the mechanism of that convergence is best understood through the sociological theory of institutional isomorphism (DiMaggio and Powell, 1983). Isomorphism describes the process by which organisations in a shared field come to resemble one another, not because resemblance makes them more efficient, but because resemblance makes them more legitimate and therefore more likely to survive. When survival depends on grant income, and grant income depends on demonstrating conformity to a funder's model of good practice, then conformity becomes a survival strategy. The clinic that follows the protocol gets funded. The clinic that improvises does not. I use the term #social_isomorphism deliberately, rather than the narrower organisational isomorphism, because the convergence at stake is not confined to formal structures such as boards, budgets, and departments. It reaches into practice. It shapes how a nurse greets a patient, what she is permitted to record, which symptoms count as data and which are dismissed as noise, how long a consultation should last, when a referral is mandatory, and what counts as a successful outcome. Standardisation of this depth is a social phenomenon, not merely an administrative one. It reorganises relationships between health workers and communities, between patients and their own bodily experience, and between local knowledge and imported expertise. The argument is not that standardisation is bad. That would be a foolish position. Standard treatment guidelines have saved an enormous number of lives. The consolidation of antiretroviral therapy into simple, public health oriented regimens made it possible to treat millions of people using a workforce that could be trained in weeks rather than years (World Health Organization, 2021). #Standardisation reduces the variability that harms patients, allows drugs to be procured at scale, permits data from one district to be compared with another, and makes it possible to detect failures. To argue against standards as such would be to argue against the possibility of a health system. The argument is rather that #standardisation_has_politics. Standards are not found in nature. They are made by particular people, in particular institutions, under particular incentives, and they encode assumptions about what matters, what is measurable, and who is competent to judge. When those people are concentrated in a small number of northern capitals, and when adherence to their standards is enforced by financial dependence rather than by professional deliberation, the result is a form of governance that is powerful, largely invisible, and rarely accountable to the communities it reshapes (Abimbola and Pai, 2020; Khan et al., 2021). This article therefore asks four questions. First, what are the concrete mechanisms through which #donor_conditionality is translated into clinical practice at the community level? Second, what does the resulting convergence gain, and what does it cost? Third, how do local actors respond, given that they are neither passive recipients nor fully autonomous agents? Fourth, what would a more defensible relationship between global standards and local practice look like? The article proceeds as follows. Section 2 sets out the structure of the global health financing field. Section 3 develops the theoretical framework, drawing on institutional theory, resource dependence, world society theory, and the sociology of standards. Section 4 explains the method and the limits of the evidence base. Section 5 describes the architecture of funding mandates in detail, treating the grant agreement as a technology of governance. Sections 6, 7, and 8 analyse coercive, mimetic, and normative isomorphism in turn. Section 9 examines the consequences, both beneficial and harmful. Section 10 turns to local agency, translation, and resistance. Section 11 develops the framework of reflexive standardisation and its practical implications. Section 12 sets out a research agenda. Section 13 concludes. 2. The Field: Money, Intermediaries, and the Long Chain of Delegation 2.1 A field, not a market To understand why clinics converge, it helps to see global health not as a market in which independent buyers and sellers meet, but as an organisational field in the sense used by institutional theorists: a recognised area of institutional life comprising funders, regulators, professional bodies, service providers, and the organisations that supply them with legitimacy and resources (DiMaggio and Powell, 1983; Scott, 2014). Fields have centres and peripheries. They have shared vocabularies. They have gatekeepers who determine what counts as a credible claim. And they have currencies of prestige that are not reducible to money, although money follows them closely. The #global_health_field took its modern shape over the last three decades. Development assistance for health grew rapidly from the 1990s, and its architecture became increasingly plural, with bilateral agencies, multilateral funds, private foundations, faith based organisations, and a dense population of implementing NGOs (McCoy, Chand and Sridhar, 2009). Large disease specific initiatives created dedicated financing channels, each with its own reporting language and its own definition of success. The consequence was a field in which a single district health office might be accountable, simultaneously, to a ministry, to three or four vertical programmes, and to several NGOs, each demanding different data on different forms at different intervals. 2.2 The intermediary position of the international NGO The #INGO occupies a structurally awkward position. It is downstream of the donor and upstream of the community. It must satisfy the donor's demand for legibility, auditability, and attributable results, while also satisfying the community's demand for services that make sense in context. These demands are frequently incompatible, and the INGO manages the incompatibility by absorbing it into its own bureaucracy: it hires monitoring and evaluation staff, grants managers, compliance officers, and technical advisers whose job is to convert messy local activity into clean donor legible outputs (Watkins, Swidler and Hannan, 2012; Krause, 2014). Krause's analysis of humanitarian relief agencies makes the crucial point that these organisations do not, in a practical sense, produce assistance for beneficiaries. They produce projects, and the project is the commodity that is exchanged with donors. Beneficiaries become the raw material out of which a good project is made. Once one sees the #project as the unit of production, many otherwise puzzling behaviours become intelligible: the preference for interventions that can be counted, the reluctance to fund slow institutional work, the short time horizons, the churn of pilots, and above all the need to demonstrate that the organisation is doing the right kind of thing in the right kind of way. Cooley and Ron (2002) described the competitive environment in which NGOs pursue contracts as producing insecurity and short horizons, which in turn generate behaviour that is rational for the organisation but dysfunctional for the mission. #Competitive_grant_seeking rewards organisations that can present themselves as credible carriers of the funder's model. Since the funder cannot directly observe quality of care in a distant village, it must rely on proxies. The proxies are compliance and conformity. The organisation that looks like a proper, professional, protocol following implementer is more likely to be trusted with money than one that looks idiosyncratic, however good its actual work. 2.3 The long chain of delegation Between the taxpayer or philanthropist who provides the funds and the patient who receives care, there may be six or more organisational links: donor agency, prime contractor, international NGO, national NGO, district office, facility, community health worker. Each link introduces its own translation and its own requirement for upward accountability. #Upward_accountability, in which each actor answers to the one above it rather than to the people it serves, is a well described pathology of the sector (Ebrahim, 2003). The consequence is that the further one travels down the chain, the more rigid the mandate becomes. The donor's original policy document may be full of language about local ownership, context sensitivity, and community participation. By the time that document has been converted into a subgrant agreement for a village based organisation, it has typically become a list of activities to be performed, targets to be reached, and forms to be completed by the fifth of each month. Flexibility is consumed at the top of the chain and scarcity of discretion is exported to the bottom. This is one of the quiet cruelties of the system: the actors with the least power to change the rules are the ones most tightly bound by them. 3. Theoretical Framework 3.1 Institutional isomorphism The foundational insight comes from Meyer and Rowan (1977), who argued that formal organisational structures often function as myth and ceremony. Organisations adopt structures that signal conformity to rationalised institutional rules, because such conformity brings legitimacy and resources, even when the structures have little to do with the technical work actually being performed. DiMaggio and Powell (1983) extended this into a typology of three mechanisms by which organisations in a field converge. #Coercive_isomorphism results from formal and informal pressures exerted by organisations on which an organisation depends, and by cultural expectations in the wider society. In global health, this is the domain of grant conditions, procurement rules, audit requirements, accreditation, and the sanction of defunding. #Mimetic_isomorphism results from uncertainty. When goals are ambiguous, when technologies are poorly understood, and when the environment is unpredictable, organisations model themselves on others that they perceive to be successful. In global health, this is the domain of best practice toolkits, model programmes, study tours, and the anxious copying of whatever design won the last major grant. #Normative_isomorphism results from professionalisation. Shared university training, shared professional associations, shared journals, and the movement of staff between organisations all create a common cognitive frame and a common set of expectations about how proper work is done. In global health, this is the domain of technical guidelines, degrees in public health, conference circuits, and the socialisation of national experts into an international epistemic community. Crucially, DiMaggio and Powell insisted that convergence produced by these mechanisms need not improve performance. Organisations may become more similar without becoming more effective. This is the disciplining claim that separates institutional analysis from the assumption that convergence reflects the discovery of the one best way. 3.2 Resource dependence Pfeffer and Salancik (1978) provide the complementary account of power. Organisations depend on their environments for resources, and those who control critical resources acquire influence over organisational behaviour. Dependence is greater when the resource is important, when it is concentrated in few hands, and when alternatives are scarce. For a community based organisation in a low income setting, external funding is often all three at once: essential, concentrated, and without substitute. This produces what may be called #asymmetric_dependence, in which the funder can walk away at low cost and the recipient cannot. Resource dependence explains not only compliance, but anticipatory compliance. Recipients frequently adjust their practices before being asked, in anticipation of what the funder will want. This anticipation is often more powerful than any explicit condition, because it is invisible, unrecorded, and therefore unaccountable. 3.3 World society theory Meyer, Boli, Thomas and Ramirez (1997) argued that states and organisations across the world adopt strikingly similar forms because they are embedded in a world culture that defines what a modern, rational, legitimate actor looks like. From this perspective, the ministry of health in a small state adopts a national strategic plan with the same chapter headings as every other national strategic plan not because it has independently calculated that this is optimal, but because that is what a proper ministry does. #World_culture supplies templates, and organisations enact them. This theory is useful here because it explains convergence that occurs without any direct coercion. Nobody had to force the template. The template circulates as a taken for granted model of modernity. 3.4 Decoupling and its limits If organisations adopt structures for legitimacy rather than efficiency, a gap may open between what they claim and what they do. Meyer and Rowan called this decoupling, and Bromley and Powell (2012) refined it by distinguishing policy practice decoupling, in which the policy is not implemented at all, from means ends decoupling, in which the policy is implemented energetically but has little connection to the outcome it is supposed to produce. #Means_ends_decoupling is the more troubling form in global health, and the more common. Facilities do fill in the register. Supervisors do conduct the checklist visits. Reports are submitted, verified, and audited. The activity is real. What is uncertain is whether any of it improves the health of the population, because the entire apparatus is oriented toward the production of compliant documentation rather than toward the diagnosis and correction of failures in care. 3.5 The sociology of standards and audit Timmermans and Berg (2003) showed that standardisation in medicine is never simply the imposition of a rule on a practice. Standards must be worked into local settings, and in the process they are modified, resisted, and sometimes quietly abandoned. Timmermans and Epstein (2010) later described a world of standards but not a standard world: standards proliferate, but their effects are uneven, and their local translations differ. Power (1997) described the rise of the #audit_society, in which the demand for verification reshapes the activity being verified. Auditing is not a neutral inspection of a pre existing reality. It creates the reality it inspects, because organisations reorganise themselves to become auditable. Espeland and Sauder (2016) and Muller (2018) documented the pathologies that follow: gaming, goal displacement, the neglect of the unmeasured, and the corrosion of professional judgement. Merry (2016) added the crucial point that #quantification is seductive precisely because it appears apolitical. A number seems to speak for itself. In fact, every number carries within it a chain of decisions about categories, boundaries, and inclusion, and those decisions are usually made far from the people being counted. 3.6 Metrics and the global health critique A body of anthropological work has examined what happens when these dynamics meet the specific conditions of global health. Adams (2016) and colleagues argued that metrics have become the dominant language of the field, determining what problems can be seen and what solutions can be funded. Wendland (2016) showed how maternal mortality estimates in Malawi were produced through chains of assumption and adjustment that were invisible in the final confident figure. Biruk (2018) documented, in ethnographic detail, how survey data in Malawi is not simply collected but manufactured through the labour, improvisation, and negotiation of fieldworkers. Storeng and Behague (2017) traced how contested indicators become instruments of advocacy, and how the political need for a clean, mobilising number can override methodological caution. Storeng, Palmer, Daire and Kloster (2019) documented how international NGOs shape national reproductive health policy in ways that are rarely visible in official accounts of country ownership. 3.7 Power, knowledge, and epistemic injustice Shiffman (2014) argued that authority in global health rests substantially on the power to define problems and to make moral claims about them. Sriram and colleagues (2018) mapped the ways power operates in health policy in low and middle income settings, including the invisible power that shapes what can even be thought. More recently, an explicit critique has emerged under the banner of #decolonising_global_health. Buyum and colleagues (2020) argued that the field reproduces colonial hierarchies through its funding structures, its authorship patterns, and its assumption that expertise flows from north to south. Khan and colleagues (2021) proposed reforms to funding, governance, and knowledge production. Chaudhuri and colleagues (2021) warned that reformist roadmaps risk leaving the underlying epistemology intact. Bhakuni and Abimbola (2021) developed the concept of #epistemic_injustice in the field, describing how knowledge produced in and about low income settings is systematically discounted unless it is validated by northern institutions. Abimbola (2021) explored how different kinds of knowledge, from the local and tacit to the universal and codified, are unevenly valued. Naidu (2021) argued that modern medicine itself carries a colonial inheritance that shapes what counts as legitimate care. Taken together, these literatures allow a synthesis: isomorphism in global health is not merely an organisational tendency but a mechanism through which #epistemic_authority is concentrated and inequality is reproduced, even by actors who sincerely intend the opposite. 4. Method and Scope This article is a conceptual and integrative review. It does not report new empirical data. Its purpose is to build a theoretical framework that connects an established body of organisational sociology to an established body of critical global health scholarship, and to derive from that connection a set of practical implications. The literature was assembled by tracing three intersecting bodies of work: classical and contemporary institutional theory; the sociology and anthropology of standards, metrics, and audit; and critical scholarship on global health governance, NGO practice, and decolonisation. Where the article describes patterns of practice, it does so in the generalised and illustrative mode, drawing on published accounts rather than on original fieldwork. Three limitations should be stated at the outset. First, the evidence base is uneven. Ethnographic accounts of clinical standardisation are rich but geographically concentrated, with a heavy weighting toward sub Saharan Africa and toward HIV and maternal health programming. Whether the mechanisms described here operate identically in, for example, South Asian tuberculosis programmes, Latin American primary care, or Pacific island health systems is an empirical question that this article cannot settle. Second, there is a selection problem in the critical literature. Scholars are more likely to publish accounts of dysfunction than accounts of quiet success. A programme in which a standard protocol was adopted, worked well, and improved outcomes without controversy generates no ethnography. The analysis here therefore attempts to give explicit weight to the benefits of standardisation, but the reader should be aware that the underlying literature tilts critical. Third, the article generalises across a heterogeneous category. #INGOs range from small faith based clinics to organisations with budgets larger than the ministries they work alongside. Their behaviour is not uniform. The claims made here are claims about tendencies produced by structural position, not about the character or intentions of any particular organisation or individual. 5. The Architecture of the Mandate To understand how a funding decision made in one country reshapes a consultation in another, one must look closely at the documents. The grant agreement is not paperwork surrounding the real work. It is a technology of governance, and its clauses are the transmission belt of isomorphism. 5.1 The results framework At the centre of most agreements sits a #results_framework, sometimes called a logical framework or logframe. It sets out a hierarchy: inputs produce activities, activities produce outputs, outputs produce outcomes, outcomes contribute to impact. Each level is attached to indicators, each indicator to a target, each target to a means of verification. The logframe appears to be a neutral planning tool. It is not. It imposes a specific ontology on the work. It requires that the intervention be decomposable into discrete, attributable, countable units. Anything that cannot be so decomposed becomes invisible to the framework, and what is invisible to the framework is, for practical purposes, invisible to the funder. Relationship building, trust, the slow accumulation of institutional competence, the political work of persuading a district authority, the emotional labour of accompanying a dying patient: none of these fit neatly, and so none of them are resourced, rewarded, or protected. The framework also fixes causality in advance. By requiring the implementer to state, at the moment of application, exactly how the activity will produce the outcome, it forecloses learning. Deviation from the stated theory of change becomes a compliance problem rather than an intellectual discovery. 5.2 Indicators and targets Indicators are the operational heart of the mandate. They specify not only what must be achieved but what may be counted as achievement. A programme funded on the indicator of number of clients initiated on treatment will organise itself around initiation. A programme funded on retention will organise itself around retention. The difference is not merely emphasis. It changes who is sought out, how they are approached, what is recorded, and what happens when a client's needs conflict with the indicator. #Target_setting compounds this. Targets are typically negotiated upward through the chain and then disaggregated downward, arriving at the facility as a number that must be reached this quarter. Because the facility did not set the number and cannot renegotiate it, and because falling short may cost the organisation its next grant, the target acquires a coercive force out of proportion to its epistemic warrant. 5.3 Standard operating procedures and approved protocols Grants routinely require that services be delivered according to an approved protocol, typically derived from an international guideline and adapted, sometimes minimally, at national level. The protocol specifies the sequence of clinical actions: what is asked, in what order, what is tested, what threshold triggers what response, what is recorded, what is referred. Protocols are, in the main, good things. They encode accumulated evidence and reduce harmful variation. But a protocol embedded in a grant condition differs from a protocol offered as professional guidance. The former is enforced by audit and money. The latter is subject to clinical judgement. When the protocol becomes a condition of payment, deviation becomes fraud rather than discretion, and the health worker's capacity to respond to the patient in front of her is curtailed. 5.4 Data systems and reporting cadence Almost every mandate specifies a #monitoring_and_evaluation system: registers, forms, unique identifiers, digital platforms, verification visits, and a reporting calendar. Data flows upward on a fixed schedule regardless of whether anything meaningful has changed. Two effects follow. First, the volume of documentation required at the point of care can consume a substantial share of clinical time, a phenomenon widely reported by frontline workers. Second, the categories in the register determine what can be perceived. If the register has no field for a condition, that condition does not exist in the data, and therefore does not exist in the planning cycle, and therefore attracts no resources. The register is an instrument of visibility and of erasure at the same time. 5.5 Audit, verification, and the compliance regime Funds are increasingly disbursed against verified results. Verification introduces auditors, spot checks, data quality assessments, and the recovery of ineligible expenditure. The rational response of any organisation facing this regime is to invest heavily in auditability. Staff are hired for their ability to produce defensible documentation. Systems are designed backwards from the audit. This is Power's audit society in its global health form. The audit does not merely observe the programme. It reconstructs the programme in its own image. 5.6 Procurement, branding, and the aesthetics of conformity Mandates typically specify procurement rules, permitted suppliers, branding requirements, and visibility obligations. The donor's logo appears on the vehicle, the register, the T shirt, the wall. These requirements seem trivial. They are not. They are a continuous, material assertion of whose project this is, and they train everyone in the vicinity, including patients, to associate the availability of care with the presence of an external benefactor rather than with a public entitlement. 5.7 Time Finally, and perhaps most consequentially, the mandate imposes a clock. Grants run in cycles, typically three to five years, frequently shorter. Health system development runs on decades. The mismatch between the #funding_cycle and the timescale of institutional development is not a minor inconvenience. It is a structural feature that makes durable capacity almost impossible to build, and it drives the pilot culture in which promising initiatives are launched, evaluated, celebrated, and closed in the same breath. 6. Coercive Isomorphism: The Force of the Contract 6.1 Conditionality as the direct channel The most straightforward mechanism is direct. To receive money, the recipient must sign an agreement. The agreement specifies the model. Non compliance means loss of funds, and in many settings loss of funds means the closure of services on which a population depends. This is #coercion in the sociological rather than the physical sense. Nobody is compelled. The recipient is free to decline. But when the alternative to compliance is the collapse of the only functioning health service in the district, the freedom is formal rather than substantive. Coercive pressure is felt most acutely at the bottom of the chain. A national NGO subcontracted by an international NGO, which is itself subcontracted by a prime, inherits every condition from every level above, plus additional conditions imposed by each intermediary to protect itself. Conditions accumulate as they descend. 6.2 The disciplinary power of the indicator Once targets are attached to money, the indicator becomes a disciplinary instrument. Performance based financing, in various forms, makes this explicit by tying payment to verified outputs. Even where financing is not formally performance based, the shadow of the next grant performs the same disciplinary function. The literature on #indicator_gaming is consistent across sectors and settings. When a measure becomes a target, the behaviour of those measured shifts toward the measure and away from the underlying purpose (Muller, 2018; Espeland and Sauder, 2016). In health, this shows up in familiar forms: recruiting the easiest to reach clients in order to hit enrolment numbers, discharging or reclassifying difficult cases, moving activity between reporting periods, and, in the most damaging cases, recording services that were not delivered. It is important to be precise about the moral character of these behaviours. They are usually not corrupt in the ordinary sense. They are the predictable output of a system that punishes honest shortfall more severely than it punishes dishonest sufficiency. Biruk's (2018) account of how data is cooked in the field is powerful precisely because it refuses to moralise: fieldworkers under pressure, with inadequate resources and impossible quotas, do what is necessary to produce a usable dataset, and everybody up the chain benefits from not looking too closely. 6.3 Certification, accreditation, and the licence to operate Beyond the grant itself, a broader compliance ecology has emerged. Organisations must demonstrate financial management standards, safeguarding policies, procurement systems, data protection compliance, and often formal accreditation before they are eligible for large grants at all. Each of these requirements is defensible on its own terms. Safeguarding matters. Financial control matters. But the cumulative effect is a substantial fixed cost of eligibility, which small and locally rooted organisations cannot easily bear. The compliance ecology therefore functions as a filter, systematically favouring large, professionalised, internationally connected organisations over small, embedded, community rooted ones. This is a quiet but powerful driver of isomorphism: the field converges partly because non conforming organisations are selected out before they ever reach the table (Khan et al., 2021). 6.4 The threat that never has to be made The most efficient coercion is the kind that is never exercised. Recipients learn, quickly, what the funder likes. They shape proposals accordingly. They avoid raising problems that might suggest incompetence. They present difficulties as lessons learned rather than as failures. They do not report the intervention that did not work. This #anticipatory_compliance is invisible in any account of the funding relationship, because it leaves no documentary trace. It is nonetheless the mechanism through which the most consequential adaptations occur, and it is one reason why donors can sincerely believe that they are not imposing anything: from where they sit, the recipient volunteered. 7. Mimetic Isomorphism: Copying Under Uncertainty 7.1 Why uncertainty drives imitation DiMaggio and Powell argued that when means ends relationships are unclear, organisations imitate models they perceive as legitimate and successful. Global health is a near perfect environment for this. Health outcomes are shaped by countless factors outside any programme's control. Attribution is genuinely hard. Feedback is slow and noisy. Under such uncertainty, the safest strategy is not to work out the best thing to do, but to do what the successful organisations are seen to do. 7.2 The toolkit economy The field has developed a vast infrastructure of #best_practice toolkits, implementation manuals, model job descriptions, sample registers, standard curricula, and ready made theories of change. These artefacts are enormously useful. They also make it far easier to copy than to design. An organisation writing a proposal under deadline, competing against dozens of others, will reach for the toolkit. The toolkit already contains language the funder recognises. It already contains indicators the funder accepts. Deviating from it means having to justify the deviation, at cost, with uncertain reward. The rational move is to adopt the model and adapt it later, if ever. 7.3 Pilots, models, and the travelling intervention The field circulates #travelling_models: intervention designs that acquire a reputation in one context and are then exported. A community health worker programme that succeeds in one country becomes the template for a dozen others. A peer support model piloted in one district becomes a national policy in a different country entirely. The circulation is often driven less by evidence of transferability than by narrative appeal. The model that travels is typically the one that is simple enough to describe in a slide, distinctive enough to be branded, and attached to a compelling story of success. Complexity, contingency, and the messy contextual factors that actually made it work do not travel well. 7.4 The homogenisation of the proposal Because proposals are the point at which organisations present themselves for judgement, and because proposals are written in the funder's language, competitive grant seeking produces a striking uniformity of stated intentions. Read enough of them and the same phrases recur: community led, evidence based, gender transformative, sustainable, scalable, integrated, rights based. This is not necessarily cynical. The words often express real commitments. But when the same vocabulary is used by every applicant, the vocabulary ceases to distinguish and becomes ritual. What distinguishes proposals then is the fluency of the performance, which is a function of professional capacity, which is a function of prior funding. #Isomorphism here is self reinforcing: the organisations that already speak the language get the money that allows them to keep speaking it. 8. Normative Isomorphism: The Making of a Shared Mind 8.1 Professional training as a transmission channel The third mechanism is the deepest, because it does not operate through constraint at all. It operates through belief. Health professionals in every country are trained in curricula that draw on international guidelines. Public health cadres, increasingly, hold degrees from a relatively small number of institutions, or from national institutions whose curricula are modelled on them. Continuing professional education is often delivered by, or funded by, the same organisations that fund services. Fellowships, exchanges, and technical assistance visits circulate people through the same nodes. The effect of this is a shared cognitive frame. Practitioners do not experience the international guideline as an imposition, because it is simply what they were taught medicine is. #Professional_socialisation converts an external standard into an internal conviction, and once that conversion has occurred, coercion is unnecessary. Eichbaum and colleagues (2021) examined how global health education partnerships reproduce hierarchies of expertise even when they are designed to be equitable. Naidu (2021) argued that the colonial inheritance of medical education shapes not only what is taught but what is thinkable within it. 8.2 Guidelines and the geography of authorship Guidelines are the crystallised authority of the field. They are produced through committees, systematic reviews, and expert consensus. The evidence they synthesise is drawn overwhelmingly from research conducted, funded, and published under northern institutional auspices, and the experts who sit on the committees are drawn disproportionately from the same institutions (Abimbola, 2019; Bhakuni and Abimbola, 2021). This is not a conspiracy. It is the accumulated result of where research money sits, which journals confer prestige, which languages are read, and which institutional affiliations are trusted. But the effect is that #clinical_guidelines carry, invisibly, the assumptions of the settings in which the underlying evidence was produced: assumptions about laboratory availability, drug supply, staffing ratios, literacy, transport, and the social organisation of family life. When those assumptions do not hold, the guideline does not simply fail to apply. It quietly redefines the local reality as deficient. 8.3 The epistemic community and its vocabulary Global health has a language. To participate in the field one must be able to speak of catalytic investment, of country ownership, of health system strengthening, of the cascade, of differentiated service delivery, of the last mile. This vocabulary is not empty, but it is a membership badge as much as a set of concepts. Those who speak it are heard. Those who do not are, at best, invited to provide local context. Abimbola (2021) has argued that this constitutes a hierarchy of knowledge in which #local_knowledge is admitted only as raw material, to be processed into universal knowledge by those with the institutional authority to universalise. The community health worker who understands, in detail, why families in her village do not attend the clinic possesses knowledge of extraordinary practical value. That knowledge enters the system, if at all, as a quotation in a qualitative annex. It does not enter as an argument that could change the protocol. 8.4 Staff mobility and the carrying of models People move. A national programme officer trained by one INGO takes a post at a ministry. A ministry official joins a multilateral fund. A consultant who designed a model in one country is contracted to design it in another. Each movement carries a model, a vocabulary, and a set of assumptions about what proper practice looks like. Over time, this circulation produces a field that thinks alike, not because anyone was forced to, but because the same people, formed in the same way, keep arriving at the same conclusions. 9. Consequences 9.1 What standardisation genuinely achieves It would be a serious analytical error, and a moral one, to treat standardisation as simply a mechanism of domination. The gains are real and they are large. Reduced harmful variation. Before standard protocols, treatment for many conditions varied enormously between practitioners, often with no clinical justification and sometimes with lethal consequences. Standard treatment guidelines reduce the scope for idiosyncratic and dangerous prescribing. Task shifting and workforce expansion. Simplified, standardised protocols made it possible to shift substantial clinical responsibility from physicians to nurses and from nurses to lay workers, without unacceptable loss of quality. Callaghan, Ford and Schneider (2010) reviewed the evidence for this in HIV care in Africa, and the approach became central to the massive expansion of treatment access. This was only possible because the protocol was simple, explicit, and standard. #Task_shifting is a direct dividend of standardisation. Procurement and supply at scale. Standard regimens permit standard procurement, which permits volume, which permits price reduction. The transformation in the affordability of essential medicines over the last two decades rests on this logic. Comparability and accountability. If every district counts the same thing in the same way, failures become visible. Comparability makes it possible to identify facilities that are underperforming and populations that are being left out. Without some standardisation of measurement, equity claims cannot even be formulated. Protection against local abuse. Standards can protect patients from local hierarchies. A protocol that requires informed consent, or that prohibits the refusal of care to unmarried adolescents, is a resource that a health worker can invoke against local pressure. Universalism is not only an imposition. Sometimes it is a shield. Any critique that cannot account for these gains is not worth taking seriously. The question is not whether to have standards, but who makes them, how they are enforced, and whether they can be revised by those who must live with them. 9.2 Ceremonial compliance and hollow practice The first and most pervasive cost is #ceremonial_compliance. Facilities perform the standard because performance is what is inspected. The register is completed. The checklist is ticked. The counselling session is recorded as delivered. Whether the register reflects care, whether the checklist detected anything, whether the counselling communicated anything, is a separate question, and it is not one the system is designed to ask. This is means ends decoupling in Bromley and Powell's sense. The activity is not fake in the sense of not happening. It is fake in the sense of being disconnected from its purpose. And the disconnection is not the fault of the health workers. It is the predictable result of a system that measures activity because activity is measurable, and that cannot measure meaning. 9.3 The distortion of clinical judgement Protocols enforced by audit narrow the space for judgement. The health worker who recognises that this particular patient needs something other than what the protocol specifies faces a choice between serving the patient and protecting the programme. Where deviation must be documented and justified, and where documentation is costly and justification is risky, the rational choice is to follow the protocol. The cost is borne by the atypical patient: the one whose presentation does not fit, whose circumstances are unusual, whose needs cross programme boundaries. Vertical, disease specific standardisation is particularly poor at handling the person who has more than one thing wrong with them, which is to say, most people. 9.4 The erasure of what is not counted Registers make some things visible and others invisible. Programmes funded to deliver a specific intervention have no category for the problems they are not funded to address. A woman who comes to an HIV clinic with severe depression, or an injury from domestic violence, or an untreated chronic condition, may be counselled, tested, and recorded, and her actual suffering may leave no trace in any system. Over time this shapes not only data but perception. Health workers learn to see what the register asks about. Communities learn what the clinic is for. And the health problems that no vertical programme owns become, in a very practical sense, nobody's problem (Prince and Marsland, 2013). 9.5 Workforce distortion and the internal brain drain Well funded international programmes pay better than ministries. They offer per diems, vehicles, training allowances, and career progression. The predictable result is that skilled national staff move from public systems to donor funded projects, and from clinical roles to monitoring and evaluation roles, because that is where the money and the prestige are. This #internal_brain_drain is one of the most damaging and least discussed consequences of the funding architecture. It hollows out the very institutions that donors claim to be strengthening, and it does so through the ordinary operation of labour markets rather than through any deliberate policy. 9.6 Vertical fragmentation Because funding arrives through disease specific channels, health systems become fragmented into parallel structures: separate supply chains, separate information systems, separate supervision, separate training, separate cadres. A district health team may spend an extraordinary proportion of its time servicing the reporting demands of vertical programmes rather than managing the health of its population. The rhetoric of #health_system_strengthening emerged partly in response to this, but as Storeng (2014) and Hafner and Shiffman (2013) have documented, system strengthening has itself frequently been absorbed into the vertical logic, becoming a set of activities delivered by projects rather than a reorientation of the architecture. 9.7 The transition cliff Grants end. When they do, the standardised apparatus they supported, the salaries, the supervision, the data systems, the commodities, ends with them. Where the programme was genuinely parallel to the public system, there may be nothing left. Communities that were taught to expect a certain standard of care are left with a facility that cannot provide it, staff who cannot be paid, and registers for which no forms are printed. The mandate that enforced conformity for five years does not, in general, take responsibility for the sixth. This is perhaps the sharpest expression of asymmetric dependence: the costs of discontinuity fall entirely on those who did not choose the terms. 9.8 The reproduction of epistemic hierarchy Finally, and most fundamentally, the system teaches. Every cycle in which local knowledge is treated as context and imported knowledge is treated as evidence reinforces a hierarchy in which the capacity to know is located elsewhere. Over decades, this shapes what national institutions believe about themselves, what young professionals aspire to, and where the answers to local problems are assumed to lie (Bhakuni and Abimbola, 2021; Abimbola and Pai, 2020). This is the deepest sense of #social_isomorphism. It is not only that organisations come to look alike. It is that the possibility of thinking otherwise is progressively foreclosed. 10. Local Agency: Translation, Friction, and Resistance The account so far risks portraying local actors as passive. They are not. The literature on standards is emphatic on this point: standards are never simply implemented. They are worked into practice, and in that working they are changed (Timmermans and Berg, 2003; Timmermans and Epstein, 2010). 10.1 Translation and the improvised protocol Between the protocol as written and the protocol as enacted lies a zone of improvisation. Health workers adapt continuously: skipping steps that are impossible given the stock available, adding steps that local circumstances require, reordering the sequence to fit the reality of a crowded waiting room, translating counselling scripts into idioms that patients actually understand. This #translation is usually invisible and frequently unrewarded, because acknowledging it would imply non compliance. Yet it is often the reason the programme works at all. The gap between protocol and practice is not always a failure. Sometimes it is where the intelligence of the system lives. 10.2 Selective compliance Local organisations learn to distinguish between the parts of the mandate that are enforced and the parts that are merely stated. Financial rules are enforced. Reporting deadlines are enforced. Community participation clauses, gender integration requirements, and sustainability plans are frequently stated and rarely inspected. The predictable result is a strategic allocation of effort. Organisations invest in the audited and economise on the aspirational. This is not hypocrisy on their part. It is a rational reading of the actual, as opposed to the declared, priorities of the funder. 10.3 Front stage and back stage Many organisations maintain, in effect, two operations: a front stage that produces donor legible outputs, and a back stage where the actual work of care and community relationship occurs. Swidler and Watkins (2017) described this dynamic in AIDS programming in Malawi, showing how local brokers skillfully manage the expectations of altruistic outsiders while pursuing their own quite different projects of survival and advancement. This should not be read as deceit. It is what happens when two parties with different purposes must cooperate through a single set of documents. The documents record the shared fiction that allows the cooperation to proceed. 10.4 Genuine resistance There is also outright refusal. Ministries decline to adopt guidelines they consider inappropriate. Professional associations contest protocols. Communities refuse interventions they experience as intrusive or disrespectful. Local researchers publish critiques. National governments assert regulatory authority over what may be piloted in their territory. Such resistance is more likely where the state is strong, where domestic financing is significant, and where a confident national professional class exists. It is least likely precisely where dependence is greatest, which is to say, where the need is greatest. #Resistance is a function of alternatives, and alternatives are unequally distributed. 10.5 Hybridisation The most common outcome is neither compliance nor resistance but #hybridisation: a working arrangement in which the imported standard and the local practice fuse into something that is not quite either. The register is completed in a way that satisfies the auditor and also records what the nurse thinks matters. The protocol is followed in its essentials and adapted in its details. The community health worker delivers the scripted message and then says what she actually thinks. Hybrid forms are usually treated as implementation failures. They should more often be treated as evidence. A protocol that is consistently modified in the same way across many sites is telling us something about the protocol, and a system that could hear that would be a system that learns (Sheikh and Abimbola, 2021). 11. Toward Reflexive Standardisation If standards are necessary and standardisation as currently practised is damaging, then the task is not to abolish standards but to change their politics. I propose the term #reflexive_standardisation for an approach with five commitments. 11.1 Standards as revisable agreements, not moral facts The first commitment is epistemological. A standard should be understood as the current best working agreement of a community of practitioners, held provisionally and open to revision in the light of experience, rather than as a fixed truth handed down from a superior source of knowledge. This has a concrete implication: every standard should carry, alongside its content, an explicit account of the evidence it rests on, the assumptions it makes about the setting, and the process by which it can be challenged. A guideline that cannot be argued with is not a guideline. It is a decree. 11.2 Feedback loops that run upward The second commitment is architectural. At present, information flows up the chain in the form of compliance data, and instructions flow down in the form of protocols. What is almost entirely absent is a channel through which the experience of implementation can travel upward with enough authority to change the instruction. Building such a channel is not technically hard. It requires that deviations be recorded rather than concealed, that the reasons for deviation be captured, that patterns of deviation be analysed, and that guideline revision processes be formally obliged to consider them. It requires, in other words, that #adaptation be treated as data rather than as failure. The learning health system literature offers concrete models for this (Sheikh and Abimbola, 2021). The obstacle is not technical. It is that such a channel would require funders to accept that their models are partly wrong, and to fund the discovery of how. 11.3 Financing that matches the timescale of institutions The third commitment concerns money. Short, project shaped, output tied grants produce short, project shaped, output oriented organisations. If the goal is durable health systems, the financing must be long, flexible, and directed toward institutions rather than activities. This means larger unrestricted core grants to national and local organisations, multi year commitments with predictable renewal, budget lines for the unglamorous work of maintenance and supervision, and explicit funding of transition well before transition occurs. It also means accepting a lower resolution of attribution. A funder cannot simultaneously demand attributable results in eighteen months and durable institutional capacity over twenty years. #Attribution and durability are, to a substantial degree, in tension, and the sector has consistently resolved that tension in favour of attribution. 11.4 Redistributing the authority to define The fourth commitment is political. Isomorphism is a symptom of concentrated authority. Diluting the symptom requires diluting the concentration. Practical steps have been proposed at length in the decolonising literature (Khan et al., 2021; Abimbola et al., 2021). They include shifting the majority of implementation and research funding directly to institutions in the countries concerned, requiring that guideline panels include substantial representation from the settings where guidelines will be applied, reforming authorship and peer review practices that discount southern scholarship, funding local research infrastructure rather than local data collection, and dismantling the tiered eligibility requirements that exclude small organisations from consideration. None of this is a matter of goodwill. Funders with good intentions have presided over this system for decades. It is a matter of where the authority to define good practice is formally located, and of building institutions that can exercise it. 11.5 Protecting the space for judgement The fifth commitment is clinical. A protocol should establish a floor, not a ceiling. Health workers must retain the authority to depart from the standard when the patient in front of them requires it, and the system must treat such departures as legitimate exercises of professional judgement rather than as compliance breaches, provided they are recorded and reasoned. This requires a shift in the meaning of #accountability: from accountability as the demonstration of conformity, to accountability as the giving of reasons. The first can be automated. The second requires trust, and trust is the resource that the current architecture is most efficient at destroying. 11.6 Implications for each actor For donors. Reduce the number of indicators. Fund institutions, not projects. Extend time horizons. Publish the assumptions embedded in your models. Create formal mechanisms through which implementers can contest conditions without jeopardising funding. Fund exit years, not exit reports. Accept, and say publicly, that some of what you fund will not be attributable. For ministries of health. Assert regulatory authority over what may be piloted, and over what data may be demanded. Insist on a single national information system into which all partners report. Build domestic financing, however slowly, because the capacity to say no is purchased. For international NGOs. Recognise the structural position you occupy and the isomorphic pressure you transmit. Absorb complexity rather than exporting it downward. Do not impose on subgrantees conditions harsher than those imposed on you. Advocate upward, using the protection your size affords, on behalf of partners who cannot afford to. For local and community organisations. Document adaptation rather than concealing it. Build federations and coalitions that can negotiate collectively, since collective negotiation is the only counter to asymmetric dependence. Invest in the capacity to author, not merely to implement. For researchers. Study what is adapted, and why. Take hybrid practice seriously as evidence. Publish negative and null findings. Refuse extractive authorship. Investigate the mandate itself as an object, since the grant agreement is at present one of the least studied and most consequential documents in global health. 12. A Research Agenda The framework advanced here generates testable propositions and identifies clear gaps. Proposition 1. The degree of clinical standardisation in a facility varies positively with the proportion of its funding derived from external, results tied sources, controlling for disease burden and staffing. This is measurable through comparative facility surveys. Proposition 2. Decoupling increases with the length of the delegation chain. The more intermediaries between funder and facility, the greater the divergence between reported and actual practice. This is testable through matched comparison of records and observation. Proposition 3. Adaptation is most frequent where guideline assumptions about infrastructure are least satisfied. Systematic recording of protocol deviation, coded by reason, would allow this to be mapped, and would simultaneously constitute the feedback mechanism that section 11.2 recommends. Proposition 4. Organisations that receive core, unrestricted, multi year funding exhibit greater programmatic heterogeneity than those funded through short project grants. Heterogeneity here is a proxy for retained discretion. Proposition 5. The severity of the transition cliff varies inversely with the degree of integration between the funded programme and the public system during the funding period. Beyond these, four areas require sustained attention. First, the #grant_agreement itself deserves systematic textual analysis. What do these documents actually say? How have their conditions changed over twenty years? What is the distribution of risk they encode? Second, the experience of the frontline worker under mandate requires far more ethnographic attention, particularly outside the settings and diseases that currently dominate the literature. Third, patient experience of standardised care is almost entirely absent from this debate. We know a great deal about what standardisation does to organisations and almost nothing about what it does to the person on the other side of the desk. Fourth, we need comparative work on cases where the pattern was broken: health systems that resisted isomorphic pressure, funders who genuinely devolved authority, guidelines that were successfully revised from below. Negative cases are the only way to distinguish structural necessity from historical choice. 13. Conclusion The convergence of clinical practice across the world's poorer health systems is often presented as progress: the global diffusion of evidence, the triumph of the best available method over local variation and error. There is truth in this. Standardisation has extended treatment to millions who would otherwise have died, and no serious analysis should pretend otherwise. But convergence is also a sociological fact with a sociological explanation, and that explanation is not primarily about evidence. It is about money, dependence, uncertainty, and the professional formation of the people who make the rules. International NGOs, positioned between funders and communities, transmit mandates downward with increasing rigidity as they descend, and those mandates reshape not only the structure of organisations but the texture of care itself. The mechanism is coercive, mimetic, and normative all at once, and its most powerful expression is the one nobody notices: the standard that is not enforced because it has already been internalised. The costs of this arrangement fall, as costs in this field usually do, on those with the least power to refuse. They take the form of ceremonial compliance, hollowed judgement, invisible suffering, distorted labour markets, fragmented systems, abrupt withdrawal, and, most insidiously, an ongoing lesson in the location of authoritative knowledge, which is always somewhere else. The alternative is not the abandonment of standards but the reform of their politics. Standards can be held as revisable agreements rather than as decrees. Feedback can be built to run upward as well as down. Financing can be matched to the timescale of institutions rather than to the timescale of reporting. The authority to define good practice can be moved closer to the places where practice actually happens. And the health worker's judgement, that irreplaceable, unmeasurable, deeply local capacity to respond to the specific person in front of her, can be treated as a resource to be protected rather than as a source of variance to be eliminated. Whether the field is capable of such reform is an open question. The structural pressures that produce isomorphism are, after all, still fully operative, and they operate on the reformers as much as on anyone else. But the first requirement of change is to see clearly what is happening. The uniformity of those wall charts, in clinics separated by oceans and centuries and languages, is not a coincidence. It is a message about who is in charge. #Global_health will not become just until that message changes. References Abimbola, S. (2019). The foreign gaze: authorship in academic global health. BMJ Global Health, 4(5), e002068. Abimbola, S. (2021). The uses of knowledge in global health. BMJ Global Health, 6(4), e005802. Abimbola, S., Asthana, S., Montenegro, C., Guinto, R. R., Jumbam, D. T., Louskieter, L., Kabubei, K. M., Munshi, S., Muraya, K., Okumu, F., Saha, S., Saluja, D., and Pai, M. (2021). Addressing power asymmetries in global health: imperatives in the wake of the COVID-19 pandemic. PLOS Medicine, 18(4), e1003604. Abimbola, S., and Pai, M. (2020). Will global health survive its decolonisation? The Lancet, 396(10263), 1627-1628. Adams, V. (Ed.) (2016). Metrics: What Counts in Global Health. Durham, NC: Duke University Press. Affun-Adegbulu, C., and Adegbulu, O. (2020). Decolonising global (public) health: from Western universalism to global pluriversalities. BMJ Global Health, 5(8), e002947. Bhakuni, H., and Abimbola, S. (2021). Epistemic injustice in academic global health. The Lancet Global Health, 9(10), e1465-e1470. Biruk, C. (2018). Cooking Data: Culture and Politics in an African Research World. Durham, NC: Duke University Press. Bromley, P., and Powell, W. W. (2012). From smoke and mirrors to walking the talk: decoupling in the contemporary world. Academy of Management Annals, 6(1), 483-530. Buyum, A. M., Kenney, C., Koris, A., Mkumba, L., and Raveendran, Y. (2020). Decolonising global health: if not now, when? BMJ Global Health, 5(8), e003394. Callaghan, M., Ford, N., and Schneider, H. (2010). A systematic review of task shifting for HIV treatment and care in Africa. Human Resources for Health, 8, 8. Chaudhuri, M. M., Mkumba, L., Raveendran, Y., and Smith, R. D. (2021). Decolonising global health: beyond reformative roadmaps and towards decolonial thought. BMJ Global Health, 6(7), e006371. Cooley, A., and Ron, J. (2002). The NGO scramble: organizational insecurity and the political economy of transnational action. International Security, 27(1), 5-39. DiMaggio, P. J., and Powell, W. W. (1983). The iron cage revisited: institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48(2), 147-160. Ebrahim, A. (2003). Accountability in practice: mechanisms for NGOs. World Development, 31(5), 813-829. Eichbaum, Q. G., Adams, L. V., Evert, J., Ho, M. J., Semali, I. A., and van Schalkwyk, S. C. (2021). Decolonizing global health education: rethinking institutional partnerships and approaches. Academic Medicine, 96(3), 329-335. Erikson, S. L. (2015). Secrets from whom? Following the money in global health finance. Current Anthropology, 56(S12), S306-S316. Escobar, A. (1995). Encountering Development: The Making and Unmaking of the Third World. Princeton, NJ: Princeton University Press. Espeland, W. N., and Sauder, M. (2016). Engines of Anxiety: Academic Rankings, Reputation, and Accountability. New York: Russell Sage Foundation. Ferguson, J. (1994). The Anti-Politics Machine: Development, Depoliticization, and Bureaucratic Power in Lesotho. Minneapolis: University of Minnesota Press. Gaudilliere, J. P., Beaudevin, C., Gradmann, C., Lovell, A. M., and Pordie, L. (Eds.) (2020). Global Health and the New World Order: Historical and Anthropological Approaches to a Changing Regime of Governance. Manchester: Manchester University Press. Hafner, T., and Shiffman, J. (2013). The emergence of global attention to health systems strengthening. Health Policy and Planning, 28(1), 41-50. Hirsch, L. A. (2021). Is it possible to decolonise global health institutions? The Lancet, 397(10270), 189-190. Khan, M., Abimbola, S., Aloudat, T., Capers, E., Sriram, V., and Hridi, A. (2021). Decolonising global health in 2021: a roadmap to move from rhetoric to reform. BMJ Global Health, 6(3), e005604. Krause, M. (2014). The Good Project: Humanitarian Relief NGOs and the Fragmentation of Reason. Chicago: University of Chicago Press. Kwete, X., Tang, K., Chen, L., Ren, R., Chen, Q., Wu, Z., Cai, Y., and Li, H. (2022). Decolonizing global health: what should be the target of this movement and where does it lead us? Global Health Research and Policy, 7, 3. Li, T. M. (2007). The Will to Improve: Governmentality, Development, and the Practice of Politics. Durham, NC: Duke University Press. McCoy, D., Chand, S., and Sridhar, D. (2009). Global health funding: how much, where it comes from and where it goes. Health Policy and Planning, 24(6), 407-417. Merry, S. E. (2016). The Seductions of Quantification: Measuring Human Rights, Gender Violence, and Sex Trafficking. Chicago: University of Chicago Press. Meyer, J. W., Boli, J., Thomas, G. M., and Ramirez, F. O. (1997). World society and the nation-state. American Journal of Sociology, 103(1), 144-181. Meyer, J. W., and Rowan, B. (1977). Institutionalized organizations: formal structure as myth and ceremony. American Journal of Sociology, 83(2), 340-363. Mosse, D. (2005). Cultivating Development: An Ethnography of Aid Policy and Practice. London: Pluto Press. Muller, J. Z. (2018). The Tyranny of Metrics. Princeton, NJ: Princeton University Press. Naidu, T. (2021). Modern medicine is a colonial artifact: introducing decoloniality to medical education research. Academic Medicine, 96(11S), S9-S12. Ndlovu-Gatsheni, S. J. (2020). Decolonization, Development and Knowledge in Africa: Turning Over a New Leaf. London: Routledge. Pfeffer, J., and Salancik, G. R. (1978). The External Control of Organizations: A Resource Dependence Perspective. New York: Harper and Row. Power, M. (1997). The Audit Society: Rituals of Verification. Oxford: Oxford University Press. Prince, R. J., and Marsland, R. (Eds.) (2013). Making and Unmaking Public Health in Africa: Ethnographic and Historical Perspectives. Athens, OH: Ohio University Press. Scott, J. C. (1998). Seeing Like a State: How Certain Schemes to Improve the Human Condition Have Failed. New Haven, CT: Yale University Press. Scott, W. R. (2014). Institutions and Organizations: Ideas, Interests, and Identities (4th ed.). Thousand Oaks, CA: Sage. Sheikh, K., and Abimbola, S. (Eds.) (2021). Learning Health Systems: Pathways to Progress. Geneva: World Health Organization, Alliance for Health Policy and Systems Research. Shiffman, J. (2014). Knowledge, moral claims and the exercise of power in global health. International Journal of Health Policy and Management, 3(6), 297-299. Shiffman, J. (2017). Four challenges that global health networks face. International Journal of Health Policy and Management, 6(4), 183-189. Sridhar, D. (2022). Preventable: How a Pandemic Changed the World and How to Stop the Next One. London: Viking. Sriram, V., Topp, S. M., Schaaf, M., Mishra, A., Flores, W., Rajasulochana, S. R., and Scott, K. (2018). 10 best resources on power in health policy and systems in low- and middle-income countries. Health Policy and Planning, 33(4), 611-621. Storeng, K. T. (2014). The GAVI Alliance and the Gates approach to health system strengthening. Global Public Health, 9(8), 865-879. Storeng, K. T., Abimbola, S., Balabanova, D., McCoy, D., Ridde, V., Filippi, V., Roalkvam, S., Akello, G., Parker, M., and Palmer, J. (2021). Action to protect the independence and integrity of global health research. BMJ Global Health, 6(1), e004718. Storeng, K. T., and Behague, D. P. (2017). Guilty until proven innocent: the contested use of maternal mortality indicators in global health. Critical Public Health, 27(2), 163-176. Storeng, K. T., Palmer, J., Daire, J., and Kloster, M. O. (2019). Behind the scenes: international NGOs influence on reproductive health policy in Malawi and South Sudan. Global Public Health, 14(4), 555-569. Swidler, A., and Watkins, S. C. (2017). A Fraught Embrace: The Romance and Reality of AIDS Altruism in Africa. Princeton, NJ: Princeton University Press. Timmermans, S., and Berg, M. (2003). The Gold Standard: The Challenge of Evidence-Based Medicine and Standardization in Health Care. Philadelphia: Temple University Press. Timmermans, S., and Epstein, S. (2010). A world of standards but not a standard world: toward a sociology of standards and standardization. Annual Review of Sociology, 36, 69-89. Watkins, S. C., Swidler, A., and Hannan, T. (2012). Outsourcing social transformation: development NGOs as organizations. Annual Review of Sociology, 38, 285-315. Wendland, C. L. (2016). Estimating death: a close reading of maternal mortality metrics in Malawi. In V. Adams (Ed.), Metrics: What Counts in Global Health (pp. 57-81). Durham, NC: Duke University Press. World Health Organization (2021). Consolidated Guidelines on HIV Prevention, Testing, Treatment, Service Delivery and Monitoring: Recommendations for a Public Health Approach. Geneva: World Health Organization. Hashtags #Social_Isomorphism #Institutional_Isomorphism #Global_Health_Governance #NGO_Funding_Mandates #Clinical_Standardisation #Donor_Conditionality #Audit_Culture #Decoupling_In_Practice #Health_Systems_Strengthening #Decolonising_Global_Health #Epistemic_Justice #Community_Health_Workers #Development_Aid_Critique #Metrics_And_Power #Local_Health_Knowledge

Social work has always worked at the meeting point between private trouble and public structure. Yet much of the profession still explains urban hardship using local causes: family breakdown, poor budgeting, weak community ties, or gaps in service provision. This article argues that such explanations are incomplete in cities that are deeply tied into the world economy. It uses #world_systems_theory, associated most closely with Immanuel Wallerstein and later developed by Christopher Chase-Dunn, Giovanni Arrighi and others, to re-read the everyday problems that #social_work encounters in large, globally connected cities. World-systems analysis treats the modern world as a single capitalist division of labour rather than a collection of separate national societies. It divides that division of labour into #core, #semi_periphery and #periphery zones, and explains persistent poverty as the result of #unequal_exchange and value transfer along #commodity_chains rather than as a national failure of development. The article does three things. First, it restates world-systems concepts in plain language for a social work readership. Second, it applies those concepts to five domains where urban practitioners work daily: labour and livelihood, housing and land, migration and care, health and welfare, and environment and infrastructure. Third, it proposes a five-part practice framework, described here as positional reading, chain tracing, scale bridging, redistributive advocacy, and solidarity building, which allows practitioners to hold individual casework and structural critique together rather than choosing between them. The article closes with implications for #social_work_education, an honest discussion of the limits and criticisms of the theory, and a research agenda. The argument is not that caseworkers should abandon individual help. It is that help offered without a map of the system tends to be repeated forever. Keywords: world-systems analysis; social work; urban inequality; resource distribution; global cities; structural practice; political economy of welfare 1. Introduction A social worker in a large coastal city meets three people in a single morning. The first is a construction labourer who has worked for eleven months, has not been paid for four of them, and cannot leave because his employer holds the paperwork that makes him legal. The second is a domestic worker whose wages support a household two thousand kilometres away, in a country whose own care sector is short of exactly the workers it exports. The third is a young mother who has been moved three times in two years because the building she rents keeps being sold to investors who intend to keep it empty. Standard training gives the practitioner tools for each case. Advocacy with the employer. A referral to a legal clinic. An application for emergency housing. These tools are real and they matter. But if the practitioner asks why these three cases keep arriving, in the same forms, year after year, the standard toolkit falls quiet. The cases are not accidents of bad luck. They are the local, human-sized appearance of a set of relationships that stretch far beyond the city. That is where #world_systems_theory becomes useful. World-systems analysis, developed by Immanuel Wallerstein and colleagues from the 1970s onward, makes a simple but demanding claim. The correct unit of analysis for understanding modern social life is not the nation, the city, or the family, but the capitalist world-economy as a whole (Wallerstein, 2004). Nations, cities and families are positions inside that larger system. Their fortunes depend on where they sit within it. Poverty in one place is not separate from wealth in another. The two are joined by flows of labour, goods, money and value. For #social_work, this is not an abstract point. The profession is one of the main institutions through which societies manage the human consequences of #capital_accumulation. It absorbs distress, distributes small quantities of resource, and, often without meaning to, makes an unequal arrangement liveable enough to continue. If practitioners do not understand the arrangement, they cannot tell the difference between relieving suffering and stabilising the machinery that produces it. This article is written for students. It assumes no prior knowledge of political economy. Its aim is to give a working understanding of world-systems concepts, to show how those concepts explain patterns that urban practitioners already recognise, and to translate the analysis into something a practitioner can actually do on Monday morning. #globalized_urban_centres are the focus because they are where the world-system becomes most visible and most dense: they concentrate capital, they concentrate #migrant_labour, and they concentrate the #social_exclusion that follows from both. The article proceeds as follows. Section 2 sets out the theoretical framework. Section 3 explains the method. Sections 4 to 8 apply the framework to the city and to the five practice domains. Section 9 develops the practice framework. Section 10 addresses education. Section 11 deals with criticisms and limitations, which are serious and must be taken seriously. Section 12 offers a research agenda, and Section 13 concludes. 2. Theoretical Framework: What World-Systems Analysis Says 2.1 The single division of labour The founding move of world-systems analysis is to reject the idea that each country is a separate society travelling along its own development path. In the standard modernisation story, poor countries are simply behind, and will catch up if they adopt the right institutions and policies. World-systems analysis argues the opposite: poor countries are not behind, they are inside. They occupy a particular functional position within one integrated economic system, and that position is what keeps them poor (Wallerstein, 2004). The system began, in this account, in the long sixteenth century in Europe, and expanded until it covered the globe. It has one economy but many states. That is its distinctive feature. A single economic division of labour operates across many political units, which means that capital can move between jurisdictions while labour largely cannot, and that costs can be pushed into places where regulation is weak while profits are booked where returns are safe. This asymmetry is not a bug. It is a central mechanism (Amin, 2018; Robinson, 2020). 2.2 Core, semi-periphery, periphery The system is structured into three broad zones. The #core contains activities that are relatively monopolised, high in value added, and protected by patents, brands, finance, standards and state power. Think of design, research, branding, legal services, financial engineering, and control over technology. The #periphery contains activities that are highly competitive, easily replaced, low in value added, and often physically demanding or environmentally destructive. Think of raw material extraction, basic assembly, garment stitching, and much agricultural work. The #semi_periphery mixes both. It contains states and regions that host some core-like activities and some peripheral ones. It plays a stabilising political role, because it gives the appearance of upward mobility and prevents the system from splitting into a simple two-way conflict (Chase-Dunn and Almeida, 2020). Two points must be stressed for a social work audience. First, these are positions of activity, not fixed labels for countries. A wealthy country contains peripheral activity, and a poor country contains core activity. This is precisely why the framework is useful inside cities. Second, the zones are relational. The #core is core because of what flows to it, not because of some inner virtue. 2.3 Unequal exchange and value transfer The engine of the system is #unequal_exchange. In simple terms, goods and services produced with large amounts of labour and material in one place are traded for goods and services produced with much less labour and material in another. Because wages differ enormously between zones, and because prices do not reflect that difference, value moves systematically from #global_south to #global_north. Recent empirical work has attempted to measure this. Hickel, Sullivan and Zoomkawala (2021) estimate large-scale net appropriation of value from lower-income countries through price differentials in international trade over a sixty-year period. Hickel and colleagues (2022) extend the analysis to material, energy, land and labour flows. Suwandi (2019) documents how modern #commodity_chains, managed by lead firms in the #core, allow those firms to capture most of the final price of a product while the work is done elsewhere. Smith (2016) makes a parallel argument about the wage gap between zones being the hidden foundation of consumer prices in rich countries. Whether or not a reader accepts every number in these studies, the direction of the argument is important. It reverses the usual moral story. Aid does not flow mainly from rich to poor. On the balance of the flows described in this literature, value flows from poor to rich, and aid is a small counter-current. For social work, which has often understood itself as a humanitarian activity carried from the developed to the developing world, this reversal is uncomfortable and necessary. 2.4 Hegemony, cycles and crisis World-systems analysis also has a temporal dimension. It describes long cycles of expansion and contraction, and periods in which one #core power holds #hegemony over the system: productive, commercial, financial and military dominance at once. Arrighi (2010) traces successive cycles of accumulation, each ending in a phase of financial expansion as productive profitability declines. Financialisation, on this reading, is not a new era but a recognisable late stage. The practical relevance is this. Periods of hegemonic decline and systemic crisis are periods of intensified pressure on labour, of welfare retrenchment, of border hardening, and of #austerity. They are also periods in which the future is genuinely open, because systems in crisis can shift. Wallerstein consistently argued that the outcome of such transitions is not determined in advance. 2.5 Antisystemic movements Finally, world-systems analysis has always paid attention to resistance. #antisystemic_movements are the labour, anti-colonial, feminist, indigenous, environmental and urban movements that challenge the logic of the system rather than simply seeking a better place within it. Chase-Dunn and Almeida (2020) treat these movements as a permanent feature of the system, cycling in waves and increasingly linked across borders. For social work, this is the point of connection. The profession is not only an instrument of management. It has also, repeatedly, been a participant in movements (Ioakimidis and Wyllie, 2023). 2.6 Relation to other critical traditions World-systems analysis is not the only structural framework available to social work, and it should be used alongside others rather than in place of them. #structural_social_work in the tradition of Mullaly and Dupre (2019) already insists that personal problems have social causes, and that practice must address both. World-systems analysis extends that insight beyond the borders of the welfare state. #anti_oppressive_practice (Baines, 2022) provides a vocabulary for the interlocking nature of oppression along lines of class, race, gender, disability and sexuality. World-systems analysis contributes the missing geography. #decolonial_social_work and postcolonial scholarship (Kleibl et al., 2020; Tascon and Ife, 2020; Ndlovu-Gatsheni, 2020) insist that the profession's knowledge base is itself a product of empire. Bhambra and Holmwood (2021) show that classical social theory was built while ignoring colonialism, and world-systems analysis is one of the few Western traditions that placed colonialism at the centre from the start. Even so, it has been criticised by decolonial writers for remaining economistic and for underplaying culture, epistemology and indigenous knowledge. That criticism is addressed in Section 11. Nancy Fraser's work provides an important bridge. Fraser (2022) argues that capitalism depends on hidden background conditions, including unpaid #care_work, expropriated nature, and racialised dispossession, which it consumes without replacing. That argument sits comfortably with a world-systems reading and translates directly into social work terms, because those hidden background conditions are exactly where practitioners spend their days. 3. Method This is a conceptual and integrative review, not an empirical study. Its purpose is theory application: to take an established framework from historical sociology and test how far it can illuminate a professional field. Literature was gathered from three bodies of work. The first is world-systems and global political economy scholarship, with emphasis on recent contributions on #unequal_exchange, #financialization and global labour. The second is critical, structural, postcolonial and eco-social scholarship within social work. The third is urban studies, particularly #global_cities research and scholarship on southern urbanism. Sources were prioritised where they were published within roughly the last five years, with a small number of older works retained where they are foundational and have no recent substitute. Selection was guided by relevance to the analytical task rather than by exhaustive coverage, which is the accepted approach in integrative and theoretical review. The limits of this method should be stated plainly. A conceptual article cannot demonstrate causal relationships. It cannot show that a world-systems informed practice produces better outcomes than a conventional one. What it can do is generate a coherent set of propositions that empirical researchers, and practitioners in reflective settings, can then test. Section 12 sets out what such testing might look like. 4. The Globalized City as a Node in the World-System 4.1 Cities as command points and as labour sinks Cities in the contemporary world-economy do two things at once. They serve as command and control points for capital, hosting the finance, legal, accounting, advertising and management functions through which distant production is coordinated. And they serve as sinks for labour, drawing in workers from regions and countries where livelihoods have collapsed. These two functions are not separate. The command function requires the labour function. A financial district cannot operate without cleaners, security guards, drivers, food couriers, nannies, construction crews and care assistants, and it typically cannot afford to pay those workers at rates that would allow them to live nearby. The result is a city that contains both #core and #periphery activity within a few kilometres, sometimes within a single building. Recent urban scholarship has pushed hard against the idea of a single hierarchy of world cities defined from the North. Robinson (2022) argues for comparative urbanism that treats cities everywhere as ordinary and mutually constitutive rather than as ranked copies of a Western model. Bhan (2019) and colleagues writing on southern urbanism insist that informality, incremental building and negotiated tenure are not deviations from a proper urban form but the normal condition of most urban humanity. Simone (2022) describes the improvised, relational practices through which people hold urban life together in the spaces the formal economy has abandoned. UN-Habitat (2022) reports that a very large share of the world's urban population lives in conditions that formal planning does not recognise. These correctives matter, but they do not dissolve the world-systems point. They refine it. The city is not a rung on a ladder. It is a node in a network, and different parts of the same node occupy different positions in the world division of labour. 4.2 Core-periphery relations inside the city Perhaps the single most useful move a social work student can make with this theory is to apply the core-periphery distinction downward, to the internal geography of the city. Consider a typical large globalised urban centre. It has a district of towers where high value-added, monopolised activity takes place. It has districts of consumption serving that first district. It has labour accommodation zones, often at the edge, often poorly served by transport, sometimes barely acknowledged in city marketing. It has informal settlements or overcrowded older housing where undocumented and low-wage workers live. It has suburbs where a professional middle class enjoys a version of #core life. Value flows between these zones exactly as world-systems analysis predicts it flows between countries. The cleaner's labour subsidises the profitability of the office. The delivery rider's unpaid waiting time subsidises the low price of the meal. The domestic worker's twelve-hour day subsidises the professional couple's two careers. The construction worker's body subsidises the tower itself. This is #spatial_inequality with a mechanism, not just a pattern. The practical value for a practitioner is immediate. Instead of asking only "what does this client need?", the practitioner can ask "which position in the urban division of labour does this client occupy, and what flows out of that position?" The second question predicts the first with surprising accuracy. 4.3 Financialisation of urban space Housing has increasingly become an asset class rather than only a place to live. Investment funds, sovereign wealth vehicles and individual speculators purchase residential property as a store of value. Harvey (2020) situates this within the broader tendency of surplus capital to seek outlets in the built environment. The consequences are familiar to any urban practitioner. Rents rise faster than wages. #gentrification pushes long-standing residents outward. Buildings stand empty in cities with waiting lists for shelter. #displacement becomes routine, and with it the loss of school places, clinic registrations, informal childcare arrangements and neighbourhood networks. #housing_precarity is thus not a housing problem alone. It is a form of accumulation that produces casework. 4.4 Infrastructure, environment and the urban metabolism The globalised city consumes materials, energy, water and food drawn from far outside its boundaries, and exports waste and emissions in return. Hickel and colleagues (2022) quantify these material flows at the international level, and the pattern reappears at the urban scale. #climate_injustice follows the same geography: the districts that consume least are frequently the districts most exposed to heat, flooding, air pollution and poor sanitation. #eco_social_work has begun to take this seriously, building on Dominelli's green social work tradition, but the field still tends to treat environmental harm as a separate specialism rather than as a routine dimension of every case involving health, housing and work. 5. Domain One: Labour, Livelihood and the Urban Working Poor 5.1 The commodity chain reaches the caseload Take a food delivery rider in any large city. The rider is at the end of a chain that includes a technology firm headquartered elsewhere, investors in several countries, a restaurant whose margins are squeezed by platform commission, and a customer who expects a low price. The rider carries the risk. The rider owns the vehicle, pays for fuel and repairs, absorbs the cost of accidents, and is paid by task, not by time. The #platform_economy did not invent this arrangement. It formalised and scaled it. The International Labour Organization (2021) documents the global spread of digital labour platforms and the associated erosion of the employment relationship. #gig_workers are legally classified in ways that place them outside most protections while remaining fully inside the discipline of the firm. For social work, the significance is that the client's problem cannot be located in the client. Debt, exhaustion, family conflict, poor health and irregular attendance at appointments are downstream of a payment model. A caseworker who addresses only the downstream effects will meet the same rider again in eight months. 5.2 Informality is not the absence of a system Around the world, a very large share of urban employment is informal. It is easy to read #informal_economy as a residual category, a place where people end up when the real economy has no room for them. World-systems analysis suggests the opposite. Informality is functional. It provides cheap goods and services that subsidise the reproduction of the formal workforce, it absorbs surplus population without cost to the state, and it supplies flexible labour at short notice. Understanding informality as functional rather than residual has direct practice implications. It means that formalisation programmes that ignore the interests served by informality will keep failing, and it means that practitioners should treat informal workers' associations as legitimate collective actors rather than as an obstacle to orderly service delivery. 5.3 Wage suppression, migration and the labour reserve Cities in the world-economy compete partly on the price of labour. That price is held down by the availability of a reserve, that is, workers who can be brought in when needed and sent away when not. #migrant_labour serves this function in almost every globalised urban centre, whether the migrants are internal, regional or international. The mechanisms differ by region. In the Gulf states, employment has long been governed by sponsorship systems that tie a worker's legal status to a single employer. Hanieh (2018) analyses the political economy of the region and the role of temporary migrant labour within it, and Kanna, Le Renard and Vora (2020) complicate simple accounts by examining how these arrangements are lived, contested and reproduced. Parrenas (2021) documents the conditions of migrant domestic workers in the Arab states in detail, showing how legal structure, household privacy and gender combine to produce dependency. Fernandez (2020) traces the migration of Ethiopian domestic workers and the recruitment infrastructure behind it. Elsewhere the mechanism is different but the function is the same: internal migration under restrictive registration systems, seasonal agricultural schemes, student visas that permit limited work, or asylum regimes that leave people without the right to work at all. The practitioner's core insight should be this. The vulnerability of the migrant worker is not a cultural trait, a language problem or a personal misfortune. It is produced by a legal architecture that exists because it lowers the cost of labour. #labour_exploitation is designed, not accidental. 5.4 What structural labour practice looks like If labour is the mechanism, then labour institutions are part of the response. This is unfamiliar territory for many social workers, especially those trained in clinical settings, but it is not new. The profession's own history includes settlement house organisers, union welfare officers and community advocates. Concretely, structural labour practice in a globalized urban centre may include: mapping the actual chain a client sits in and identifying who holds the money; documenting patterns of wage theft rather than resolving them one by one; connecting clients to #trade_unions or workers' centres where they exist and helping to build them where they do not; supporting collective claims for a #living_wage; and giving evidence to inquiries and regulators. These are not additions to social work. They are #macro_practice, and they belong to the profession. 6. Domain Two: Migration, Care and Social Reproduction 6.1 Global care chains When a woman leaves a lower-income country to care for children in a higher-income city, care does not appear from nowhere. It is transferred. Her own children are cared for by a grandmother, an older sibling, a poorer neighbour, or nobody. #global_care_chains describe this transfer, and they are among the clearest illustrations of #unequal_exchange in everyday life. The exchange is unequal in a precise sense. The care that the migrant provides is highly valuable to the receiving household, which could not otherwise sustain two professional careers. The care that is lost in the sending household is not compensated. The difference is captured as a saving by the receiving household and, indirectly, by the employers of both professionals, who are relieved of the cost of a shorter working day. #social_reproduction is the technical name for all the work that produces and maintains human beings: feeding, cleaning, nursing, teaching, comforting, and managing crises. Fraser (2022) argues that capitalism systematically undervalues this work while depending on it absolutely. Because it is overwhelmingly performed by women, and increasingly by racialised migrant women, #gendered_labour and #racialization are not side issues but structural features. 6.2 Remittances as a policy substitute #remittances now exceed official development assistance for many countries. They are frequently celebrated as a development success. World-systems analysis invites a colder reading. Remittances are the privatised replacement of a public function. They transfer the cost of social protection in the sending country onto the bodies of its emigrants, who work long hours in dangerous conditions abroad so that their families can pay for healthcare and schooling that the state does not provide. This is why the same countries that export care workers frequently have severe shortages of care workers at home. The #brain_drain in nursing and medicine is the sharpest example, but the principle applies across skill levels. 6.3 Practice implications For social work, several implications follow. First, assessment must include the transnational household. A client's income, obligations, decisions and stress levels cannot be understood by looking only at the people who live in the flat. A worker sending sixty per cent of a low wage abroad is not managing money badly. She is meeting an obligation created by a system that has withdrawn from her family's country. Second, the profession should be wary of framing migrant clients primarily through vulnerability and rescue. Anti-trafficking frameworks, while sometimes necessary, can individualise a structural relationship and can end in deportation rather than protection. Practitioners should ask whether an intervention increases a worker's power or merely removes her from view. Third, social work in receiving cities has an obligation to the sending regions. This is not sentimental. It is a recognition that the receiving city's care economy is subsidised by them. 7. Domain Three: Welfare, Health and the Distribution of Public Resource 7.1 The welfare state as a core institution Comprehensive #welfare_state provision has historically been concentrated in #core zones and was built during a period when those zones captured a large share of world surplus. This is not to dismiss the achievement, which was won by organised working-class movements at real cost. It is to notice the material conditions that made it affordable. Once those conditions change, the pressure on welfare is immediate. Capital mobility, tax competition, and #debt discipline restrict what states can spend. In peripheral and semi-peripheral countries, #structural_adjustment programmes explicitly required reductions in public spending, the introduction of user fees, and the privatisation of services. The results in health, education and #social_protection are documented across decades. In core cities, the same logic arrives under the name #austerity. Services are cut, thresholds for eligibility rise, and the residual service is targeted at the most acute cases. Garrett (2021) documents the effects of this on the profession itself: rising caseloads, narrowing discretion, and a shift from support to risk management. 7.2 NGOisation and professional imperialism Where the state retreats, non-governmental organisations enter. This is often presented as an efficiency gain and sometimes as an expression of civil society. World-systems analysis suggests a more careful reading. #NGOisation refers to the process by which social provision is transferred to project-funded, donor-dependent organisations whose priorities are shaped by funders located in the #core. Funding cycles are short. Reporting requirements are heavy. Programmes must be demonstrably measurable, which biases them toward individual behaviour change and away from collective political action. Staff who could be building durable local institutions spend their time writing bids. #professional_imperialism is the related problem of exporting social work models, curricula, ethical codes and assessment instruments from core countries to peripheral ones without adaptation. Kleibl and colleagues (2020) document this at length, and Sewpaul and Henrickson (2019) discuss the tensions involved in constructing global ethical statements that do not simply universalise a Western particular. The point is not that international collaboration is wrong. It is that collaboration between unequal partners tends to reproduce the inequality unless it is deliberately structured otherwise. 7.3 Health inequities and the social gradient #health_inequities follow the same map. Marmot (2020) documents the persistence of a social gradient in health, in which each step down the socioeconomic ladder corresponds to worse outcomes, and shows how #austerity widened the gap. The #social_determinants of health are, in world-systems terms, simply the local expression of one's position in the division of labour: what work you do, what you breathe, what you can afford to eat, whether you can rest, and whether you can afford to stop working when ill. For urban social work, this means that health referrals without material change are frequently symbolic. Advising a client with respiratory illness to avoid air pollution, when she lives beside a freight route because that is where the rent is affordable, is not clinical guidance. It is a description of her class position delivered as advice. 7.4 Child welfare and the inequalities gradient Perhaps the most challenging body of recent evidence for social work concerns children. The Child Welfare Inequalities Project and related work (Bywaters et al., 2020; Bywaters and Skinner, 2022) demonstrates a strong association between deprivation and the likelihood that a child will be subject to a protection plan or taken into care. #child_welfare_inequalities are not evenly distributed; they track poverty closely. Featherstone and colleagues (2018) argue for a social model of child protection that takes this seriously, and Krumer-Nevo (2020) develops a poverty-aware paradigm in which practitioners work with, rather than around, the material conditions of family life. Read through a world-systems lens, this evidence is even more uncomfortable. If poverty in globalised cities is produced by the position of certain workers in the world division of labour, and if child protection intervention tracks poverty, then the child protection system is, in part, an instrument that processes the human consequences of value transfer. That is a hard sentence to write and a harder one to sit with. It is not an argument for inaction. It is an argument for locating action at the level where the cause lies. 8. Domain Four: Land, Housing and Urban Resource Distribution 8.1 Land as the oldest question #land_rights sit beneath almost every urban conflict. Who may build, who may stay, who may be moved, and who captures the increase in value when a district becomes desirable: these questions determine the shape of the city. In many globalised urban centres, land is the primary vehicle for accumulation. Public land is sold to private developers. Infrastructure investment raises adjacent land values, and that increase accrues to owners rather than to the public that funded the infrastructure. #displacement follows, and the displaced are re-housed, if at all, at the periphery, where transport costs consume the wages of the people who service the centre. 8.2 Informal settlements and the politics of recognition Where formal housing is unaffordable, people build. Informal settlements are a rational response to a market failure, and they house an enormous share of urban humanity (UN-Habitat, 2022). They are also, routinely, targeted for clearance in the run-up to major events and investments. Social work's role in clearance has been ambiguous and sometimes shameful. Practitioners have been used to enumerate residents, to persuade them to accept inadequate compensation, and to manage the distress of removal. Ioakimidis and Wyllie (2023) place such episodes in a longer history of the profession's complicity, alongside its history of resistance. Honest engagement with that history is a professional obligation, not an academic exercise. 8.3 The redistributive alternatives There are policy instruments that redirect land value toward the public. They include land value capture, public land banking, community land trusts, permanent affordability covenants, rent regulation, vacancy taxation and large-scale public housing. Whether any of these is achievable in a given city is a political question, not a technical one. The relevance for practitioners is that #policy_practice in housing is not a distraction from frontline work. It is the only route by which the frontline work stops repeating. A social worker who has spent five years placing families in temporary accommodation possesses evidence that no researcher can easily obtain. Presenting that evidence to a council, a select committee, an inquiry or a coalition is #advocacy in its most useful form. 9. A World-Systems Informed Framework for Urban Social Work Practice The preceding sections describe a problem. This section proposes a way of working. The framework has five components. It is intended to be usable at any level of practice, from a single interview to a citywide campaign. 9.1 Component one: positional reading Positional reading means asking, at the start of every assessment, where this person sits in the division of labour, and what flows out of that position. The questions are simple. What work does this person do, formally or informally? Who ultimately profits from that work, and where are they? What legal status governs the person's presence and employment, and who benefits from that status? What obligations does the person carry to people elsewhere? What is the person's position with respect to land: owner, tenant, occupier, or none of these? This is not a replacement for standard assessment. It is an additional layer, and it takes perhaps ten minutes. Its value is that it converts a list of problems into a structure. #power_analysis becomes routine rather than exceptional. 9.2 Component two: chain tracing Chain tracing means following the #commodity_chains in which clients are embedded, upward toward the point where value is captured. If a caseload contains twenty garment workers, chain tracing asks: which brands? Which intermediaries? Which contracts? If a caseload contains fifty delivery riders, it asks: which platform, which pay algorithm, which classification, which insurer? If a caseload contains construction labourers, it asks: which developer, which main contractor, which subcontractor, which recruiter in which sending country, and what fee did the worker pay to get here? Practitioners are often the only people in a city who hold this information across many cases. Aggregating it, anonymously and ethically, converts individual complaints into evidence of a pattern. Patterns can be acted upon. Individual complaints usually cannot. 9.3 Component three: scale bridging Scale bridging means holding the individual and the systemic together without collapsing one into the other. There is a real danger in structural analysis, and it should be named. A practitioner who has understood that the world-economy causes their client's suffering may conclude that casework is futile. This is a serious error. The client in front of you needs the rent paid this month, the medication obtained this week, the child kept safe tonight. Structural insight does not licence abandonment. #praxis means doing both, and doing them consciously. Scale bridging in practice means: deliver the immediate help fully and competently, and simultaneously record what the case reveals about the system; treat every act of individual help as also a piece of evidence; and be honest with clients about what the service can and cannot change, because false promises are their own form of harm. 9.4 Component four: redistributive advocacy Redistributive advocacy means directing professional effort toward #redistribution of resource rather than only toward improved allocation of scarcity. Most social work advocacy operates within a fixed budget. It seeks to get this family the place, this client the payment, this district the clinic. That is necessary. But if the budget is itself a product of #resource_distribution decisions taken at higher scales, then advocacy must sometimes address those decisions: taxation, land value capture, labour regulation, migration status, and the funding of #universal_basic_services. This is legitimate professional activity. Ethical codes across jurisdictions commit the profession to #social_justice and to challenging unjust policy. The gap is not permission. It is training, time and organisational courage. 9.5 Component five: solidarity building Solidarity building means treating those affected as agents rather than as recipients, and building alliances across the boundaries the system creates. It includes supporting workers' organisations, tenants' unions, migrant associations, disabled people's organisations and residents' committees. It includes #transnational_solidarity, because the same firms operate in the sending and receiving countries and the same regulatory gaps are exploited in both. It includes #community_development in its original political sense, as described by writers who insist that development is done by people, not to them. It also requires humility. Practitioners are usually not the leaders of these movements, and should not attempt to be. The professional contribution is technical, evidentiary and infrastructural: access, records, credibility, and the capacity to translate lived experience into language that institutions cannot ignore. 9.6 Summary table Component Core question Typical action Level Positional reading Where does this person sit in the division of labour? Extended assessment including work, status, obligations, land Micro Chain tracing Who captures the value of this person's labour? Aggregated, anonymised mapping of employers, platforms, contractors Mezzo Scale bridging How do I help now and change later? Full casework plus systematic recording of structural evidence Micro and macro Redistributive advocacy Who decides the size of the budget, not just its allocation? Submissions, testimony, coalition policy work Macro Solidarity building Who are the agents of change here? Support to unions, tenant groups, migrant associations Mezzo and macro 10. Implications for Social Work Education 10.1 The missing political economy Most social work curricula contain modules on human development, mental health, law, ethics, research methods and practice skills. Few contain a serious module on political economy. Students therefore graduate able to describe inequality with precision and unable to explain it. A world-systems informed curriculum would add, at minimum: the history of colonialism and its relationship to contemporary welfare arrangements; the structure of global production and #commodity_chains; the political economy of migration; the fiscal foundations of the #welfare_state; and the history of the profession's own entanglement with empire and with the state. Bhambra and Holmwood (2021) provide a route into the first of these. Kleibl and colleagues (2020) provide it for the profession specifically. Chase-Dunn and Almeida (2020) provide an accessible entry to world-systems analysis itself. 10.2 Teaching method Content alone will not do it. #critical_pedagogy is required, in which students examine their own position within the system rather than treating it as something that happens to other people. Practical exercises might include: mapping the commodity chain behind an object in the classroom; tracing the ownership of a local housing development; interviewing a worker in a service the students use daily; or comparing the legal rights of two workers doing similar jobs on different visas. #reflexivity here means more than reflecting on feelings. It means asking what one's own salary, mobility and professional status depend on. 10.3 The risk of demoralisation Teaching structural analysis to students who will soon enter under-resourced services carries a real risk. Students may conclude that nothing they do matters. Educators have a duty to teach the analysis alongside the history of what movements have actually achieved: the eight-hour day, public housing, universal health systems, the abolition of legal indenture in various forms, and the extension of rights to groups previously excluded. None of these was granted. All were won, often with social workers among the organisers. Hope, in this framing, is not optimism. It is the recognition that systems in crisis are also systems in motion. 11. Criticisms, Limitations and Honest Tensions An article that only praised its own framework would be poor scholarship. The criticisms of #world_systems_theory are substantial, and students should know them. 11.1 Structural determinism The most common criticism is that the theory is too structural. If everything is explained by the position of an actor within the world-economy, then agency disappears, and so does the possibility of meaningful action at any scale below the systemic. This is a genuine risk in application, even if the founding texts explicitly reject determinism. The response for social work is the scale-bridging principle described above. Structure constrains; it does not dictate. People and movements act within constraints, and those actions have real effects on real lives, including effects the system did not intend. 11.2 Economism World-systems analysis has been criticised for reducing culture, religion, gender, race and knowledge to reflections of economic position. Decolonial scholars in particular argue that the coloniality of knowledge and being is not simply a by-product of the economic system (de Sousa Santos, 2018; Ndlovu-Gatsheni, 2020). Feminist scholars have argued that #social_reproduction cannot be treated as a residual category. The honest position is that world-systems analysis needs supplementation. It gives an excellent account of the flows of value. It gives a weaker account of the flows of meaning. Social work, which works with meaning constantly, must use both. 11.3 The unit of analysis problem If the world-system is the unit of analysis, it is not obvious what a practitioner in one city is supposed to do. The theory operates at a scale far above the scale of professional action. This is a fair criticism of the theory as a practice guide, and it is the reason this article devotes a full section to translation. 11.4 Empirical disputes The measurement of #unequal_exchange is contested. The studies cited here (Hickel, Sullivan and Zoomkawala, 2021; Hickel et al., 2022) rest on particular assumptions about prices, value and counterfactuals, and other economists dispute both the method and the magnitude. Students should read the critiques as well as the claims. The general direction of value transfer is widely accepted among heterodox economists; the specific quantities are not settled. Similarly, the tripartite division of core, semi-periphery and periphery is a heuristic rather than a precise classification, and different scholars place different countries in different zones. 11.5 The risk of a totalising story Finally, there is a professional risk. A practitioner armed with a complete explanation of everything may stop listening. Clients are not illustrations of a theory. They are people with their own accounts of their own lives, and those accounts frequently do not fit the model. Theory should sharpen attention, not replace it. 12. A Research Agenda If the argument of this article is worth anything, it should generate testable questions. Several are proposed here for students seeking dissertation topics. First, the caseload composition question. Do social work caseloads in globalised urban centres cluster around identifiable positions in the urban division of labour, and can those clusters be mapped onto specific #commodity_chains? This is answerable through anonymised administrative data combined with occupational coding. Second, the intervention effectiveness question. Does structural, chain-traced advocacy produce durable improvements in client circumstances compared with individual casework alone? This is difficult to test but not impossible, particularly through comparison of services that do and do not employ policy practice. Third, the professional identity question. How do practitioners themselves understand the causes of the problems they encounter, and does exposure to political economy change what they do? This is amenable to qualitative and longitudinal study. Fourth, the transnational question. What happens in sending communities when care workers migrate, and what obligations, if any, do receiving-city services acknowledge toward them? This requires collaborative, multi-sited research with partners in sending regions, and should be designed to avoid the extractive research relationships that postcolonial scholars rightly criticise. Fifth, the education question. Does a curriculum that includes political economy produce graduates who practise differently, and if so, for how long before organisational pressure erodes the difference? Sixth, the environmental question. How do #climate_injustice and urban environmental exposure interact with the positions described here, and what does #eco_social_work look like when it is grounded in a world-systems account of who consumes and who suffers? 13. Conclusion This article began with three cases in a single morning: an unpaid construction worker, a domestic worker supporting a household abroad, and a young mother displaced by investment. Conventional social work would treat these as three unrelated problems requiring three interventions. #world_systems_theory treats them as three appearances of one thing. The construction worker is unpaid because his legal status was designed to hold down the cost of building the city. The domestic worker's household abroad depends on remittances because the state that should support it was disciplined into withdrawal. The young mother is displaced because housing in her city has become a vehicle for storing surplus capital. In each case, a flow of value runs out of the client's life and into somewhere else. Naming this does not make casework unnecessary. The rent still has to be paid. But naming it does change what a professional is for. If practitioners understand themselves only as repairers of damaged individuals, they will spend their careers repairing the same damage. If they understand themselves as witnesses to a structure, holders of evidence, and allies of the people producing the wealth of the cities they live in, then the same daily work becomes something else: part of a longer effort to change the conditions that make the work necessary. #social_work did not create the world-system, and it will not, by itself, transform it. But the profession sits at a rare vantage point. It sees the human interior of the machine. Very few institutions do. To see clearly and to say nothing is a choice, and it is not a neutral one. The task, then, is not to choose between the individual and the structure. It is to refuse the choice. References Amin, S. (2018). Modern Imperialism, Monopoly Finance Capital, and Marx's Law of Value. New York: Monthly Review Press. Arrighi, G. (2010). The Long Twentieth Century: Money, Power and the Origins of Our Times. New edition. London: Verso. Baines, D. (ed.) (2022). Doing Anti-Oppressive Social Work: Rethinking Theory and Practice. 4th edition. Halifax: Fernwood Publishing. Bhambra, G. K. and Holmwood, J. (2021). Colonialism and Modern Social Theory. Cambridge: Polity Press. Bhan, G. (2019). Notes on a Southern urban practice. Environment and Urbanization, 31(2), 639 to 654. Bywaters, P., Scourfield, J., Webb, C., Morris, K., Featherstone, B., Brady, G., Jones, C. and Sparks, T. (2020). The Child Welfare Inequalities Project: Final Report. Coventry: Nuffield Foundation. Bywaters, P. and Skinner, G. (2022). The Relationship Between Poverty and Child Abuse and Neglect: New Evidence. Coventry: Nuffield Foundation. Chase-Dunn, C. and Almeida, P. (2020). Global Struggles and Social Change: From Prehistory to World Revolution in the Twenty-First Century. Baltimore: Johns Hopkins University Press. Chancel, L., Piketty, T., Saez, E. and Zucman, G. (2022). World Inequality Report 2022. Cambridge, MA: Harvard University Press. de Sousa Santos, B. (2018). The End of the Cognitive Empire: The Coming of Age of Epistemologies of the South. Durham: Duke University Press. Dominelli, L. (2012). Green Social Work: From Environmental Crises to Environmental Justice. Cambridge: Polity Press. Featherstone, B., Gupta, A., Morris, K. and White, S. (2018). Protecting Children: A Social Model. Bristol: Policy Press. Fernandez, B. (2020). Ethiopian Migrant Domestic Workers: Migrant Agency and Social Change. Cham: Palgrave Macmillan. Fraser, N. (2022). Cannibal Capitalism: How Our System Is Devouring Democracy, Care, and the Planet, and What We Can Do About It. London: Verso. Garrett, P. M. (2021). Dissenting Social Work: Critical Theory, Resistance and Pandemic. Abingdon: Routledge. Hanieh, A. (2018). Money, Markets, and Monarchies: The Gulf Cooperation Council and the Political Economy of the Contemporary Middle East. Cambridge: Cambridge University Press. Harvey, D. (2020). The Anti-Capitalist Chronicles. London: Pluto Press. Hickel, J. (2020). Less Is More: How Degrowth Will Save the World. London: William Heinemann. Hickel, J., Dorninger, C., Wieland, H. and Suwandi, I. (2022). Imperialist appropriation in the world economy: Drain from the global South through unequal exchange, 1990 to 2015. Global Environmental Change, 73, 102467. Hickel, J., Sullivan, D. and Zoomkawala, H. (2021). Plunder in the post-colonial era: Quantifying drain from the global South through unequal exchange, 1960 to 2018. New Political Economy, 26(6), 1030 to 1047. International Labour Organization (2021). World Employment and Social Outlook 2021: The Role of Digital Labour Platforms in Transforming the World of Work. Geneva: ILO. Ioakimidis, V. and Wyllie, A. (eds.) (2023). Social Work's Histories of Complicity and Resistance: A Tale of Two Professions. Bristol: Policy Press. Kanna, A., Le Renard, A. and Vora, N. (2020). Beyond Exception: New Interpretations of the Arabian Peninsula. Ithaca: Cornell University Press. Kleibl, T., Lutz, R., Noyoo, N., Bunk, B., Dittmann, A. and Seepamore, B. (eds.) (2020). The Routledge Handbook of Postcolonial Social Work. Abingdon: Routledge. Krumer-Nevo, M. (2020). Radical Hope: Poverty-Aware Practice for Social Work. Bristol: Policy Press. Marmot, M. (2020). Health Equity in England: The Marmot Review 10 Years On. London: Institute of Health Equity. Mullaly, B. and Dupre, M. (2019). The New Structural Social Work: Ideology, Theory, and Practice. 4th edition. Oxford: Oxford University Press. Ndlovu-Gatsheni, S. J. (2020). Decolonization, Development and Knowledge in Africa: Turning Over a New Leaf. Abingdon: Routledge. Parrenas, R. S. (2021). Unfree: Migrant Domestic Work in Arab States. Stanford: Stanford University Press. Piketty, T. (2020). Capital and Ideology. Cambridge, MA: Harvard University Press. Robinson, J. (2022). Comparative Urbanism: Tactics for Global Urban Studies. Hoboken: Wiley Blackwell. Robinson, W. I. (2020). The Global Police State. London: Pluto Press. Sewpaul, V. and Henrickson, M. (2019). The (r)evolution and decolonization of social work ethics: The Global Social Work Statement of Ethical Principles. International Social Work, 62(6), 1469 to 1481. Simone, A. (2022). The Surrounds: Urban Life Within and Beyond Capture. Durham: Duke University Press. Smith, J. (2016). Imperialism in the Twenty-First Century: Globalization, Super-Exploitation, and Capitalism's Final Crisis. New York: Monthly Review Press. Suwandi, I. (2019). Value Chains: The New Economic Imperialism. New York: Monthly Review Press. Tascon, S. and Ife, J. (eds.) (2020). Disrupting Whiteness in Social Work. Abingdon: Routledge. UN-Habitat (2022). World Cities Report 2022: Envisaging the Future of Cities. Nairobi: United Nations Human Settlements Programme. Wallerstein, I. (2004). World-Systems Analysis: An Introduction. Durham: Duke University Press. Hashtags #WorldSystemsTheory #SocialWorkResearch #UrbanInequality #GlobalSocialWork

Psychological tests are among the most powerful instruments in the human sciences. They decide who receives a diagnosis, who is admitted to a university, who is judged fit to work, and whose suffering is treated as real. Yet most of the tools in daily use were built inside a narrow slice of the world and were validated on people who were mostly Western, educated, industrialised, rich and democratic. When those tools travel across borders, they carry their assumptions with them. This article argues that the resulting harms are not only technical problems of translation or norming, they are also problems of #epistemic_injustice: the systematic devaluing of some people as knowers and as makers of meaning. Drawing on recent work in the philosophy of psychiatry, cross cultural neuropsychology, psychometrics and global mental health, the paper does four things. First, it explains the concept of epistemic injustice in plain language and shows how testimonial, hermeneutical and contributory injustice each appear at specific points in the assessment process. Second, it traces how bias enters the full life cycle of a test, from construct definition to feedback and policy use, rather than entering only at the moment of translation. Third, it reviews the psychometric machinery that the field uses to detect bias, including measurement invariance and differential item functioning, and it reports the live scholarly disagreement about how much these methods can actually deliver. Fourth, it proposes a staged, participatory framework for developing #culturally_responsive_assessment tools for international populations, together with practical guidance on cognitive interviewing, interpreter mediated testing, norming, interpretation and governance. The paper closes with an agenda for students and early career researchers who want to build fairer instruments. The central claim is simple. A test is fair only when the people it measures have had a real say in what it measures, how it asks, and what its scores are allowed to mean. Keywords: epistemic injustice, psychological testing, cultural adaptation, measurement invariance, differential item functioning, international students, migrants and refugees, test fairness, global mental health, participatory psychometrics Introduction Every year, millions of people sit a psychological test in a language, a format and a cultural world that is not fully their own. An international student in a counselling centre fills in a depression questionnaire written for a suburban campus in another hemisphere. A refugee is given a memory task built around word lists that assume twelve years of formal schooling. A migrant worker completes a personality inventory whose items were designed to detect traits that matter to employers in a very different labour market. In each case a number is produced. In each case that number travels further and faster than the story behind it. This article is about what happens when the number is wrong, and more importantly, about why the wrongness so often goes unnoticed. The usual explanation is technical. The test was not properly translated. The norms were not local. The sample was too small. These explanations are true, and this paper will treat them seriously. But they are also incomplete. Behind the technical failures sits a deeper pattern in which certain people are treated as less credible when they describe their own minds, and certain communities are never invited into the room where the concepts are defined in the first place. Philosophers call this #epistemic_injustice, and the concept has moved rapidly from philosophy into psychiatry, nursing and mental health research over the last few years (Kidd, Spencer, and Carel, 2022; Fisher, 2023; Newbigging, Salla, Schon, and King, 2024). The stakes are not abstract. #Psychological_testing sits at the entrance to many of the most consequential doors in modern life. Educational placement, clinical diagnosis, disability support, immigration and asylum decisions, custody assessments, employment selection and access to medication all rely, at least in part, on standardised scores. When those scores misfire for people from #international_populations, the result is not a small statistical annoyance. It is a person who is denied care, a child who is misplaced in a special education stream, an asylum claim that is doubted because a trauma narrative did not follow the expected shape, or a competent adult who is described in a report as impaired. The purpose of this article is to give students and early career researchers a clear map of the problem and a usable path forward. It is written in simple English on purpose. The topic is often discussed in dense philosophical prose or in equally dense statistical notation, and both dialects exclude people. If the argument of this paper is that exclusion from knowledge production is a form of injustice, then writing in a way that excludes readers would be a strange way to make it. 1.1 What this article argues The article makes five connected claims. First, #test_bias is not only a psychometric fact, it is also an epistemic and political one. A biased item is the visible tip of a longer chain of decisions in which some people's ways of describing distress, attention, memory or personality were treated as the standard and others were treated as noise. Second, bias enters the test life cycle at many points, not only at translation. If we only fix translation, we fix the least important link. Third, the statistical tools that the field uses to detect bias, above all #measurement_invariance testing and #differential_item_functioning analysis, are necessary but they are not sufficient, and there is currently a serious and unresolved argument in the literature about how far they can be trusted for cross cultural comparison (Wilson, Bowden, Byrne, Joshua, Marx, and Weiss, 2023; Fischer, Karl, Luczak-Roesch, and Hartle, 2025; Kusano, Napier, and Jost, 2025). Students should know that this debate exists, because textbooks often present invariance testing as a settled solution. Fourth, culturally responsive assessment requires #participatory_psychometrics, meaning that members of the population being measured take part in defining the construct, writing the items, judging the wording and interpreting the results. This is not a courtesy. It is a validity requirement. Fifth, the same logic must be applied to the new generation of digital and algorithmic assessment tools, or the field will simply automate the injustices it has spent decades trying to name. 1.2 Who counts as an international population The phrase international populations is used here in a deliberately broad way. It includes international and exchange students, labour migrants, refugees and asylum seekers, expatriate professionals, second generation minority communities, Indigenous peoples inside settler states, and the very large populations of the Global South on whom instruments from the Global North are routinely deployed. These groups are not the same. A doctoral student from Nigeria studying in Germany and a Rohingya refugee in a camp clinic face very different versions of the problem. But they share a structural position. They are being measured by tools whose logic was decided somewhere else, and they usually have no channel through which to contest that logic. The article uses the term #cultural_responsiveness rather than cultural competence. Competence suggests a finished state that a clinician can acquire and then own. Responsiveness suggests an ongoing relationship in which the tool and the tester keep adjusting to the person in front of them. The second framing is more honest about how little any of us ever finish learning about a culture that is not our own. 1.3 Structure of the paper Section 2 explains epistemic injustice and its varieties. Section 3 gives a short and unflattering history of psychological testing, because the present cannot be understood without it. Section 4 walks through the test life cycle and locates injustice at each stage. Section 5 explains the psychometric toolkit in plain language and reports the current controversy. Section 6 examines the specific situations of international students, migrants and refugees. Section 7 sets out the proposed development framework. Section 8 covers methods for practice, including interpreter mediated assessment and the cultural formulation interview. Section 9 turns to digital and algorithmic assessment. Section 10 addresses ethics, governance and training. Section 11 offers illustrative composite cases. Section 12 gives limitations and a research agenda. Section 13 concludes. Epistemic Injustice: A Working Vocabulary 2.1 The core idea The philosopher Miranda Fricker introduced the term epistemic injustice to describe a wrong done to someone specifically in their capacity as a knower (Fricker, 2007). The idea has since become one of the most widely used concepts in the philosophy of psychiatry, with a very large volume of published work appearing in the last few years alone (Pantazakos and Arnaud, 2024). Its appeal is easy to understand. Mental health care is unusually dependent on what people say about their inner lives, and it is unusually well supplied with reasons to doubt them. Three varieties matter for #psychological_assessment. 2.2 Testimonial injustice #Testimonial_injustice happens when a listener gives a speaker less credibility than they deserve because of a prejudice about the group the speaker belongs to. In mental health settings this is close to routine. Professionals hold what has been described as epistemic privilege, since their scientific and medical vocabulary is treated as more credible than the service user's own account of their experience (Fisher, 2023). Add an accent, a translated sentence, an unfamiliar way of describing sadness, and the credibility discount deepens. In testing, testimonial injustice appears in small and easily missed moments. A test taker says that a question does not make sense to them, and the examiner records this as evasiveness or poor comprehension rather than as evidence that the item is badly designed. A patient says the memory task is strange because nobody in their community has ever been asked to memorise an unrelated list of grocery items, and this is scored as a deficit rather than as an unfamiliar task demand. The person's own testimony about the instrument is available, and it is discarded. 2.3 Hermeneutical injustice #Hermeneutical_injustice is subtler. It occurs when a group's experiences cannot be properly understood because the shared pool of concepts does not contain the resources needed to make sense of them. The person is not simply disbelieved. They cannot even be understood, because the words are missing. Service users may lack the vocabulary to interpret their own mental health experiences in a way that professionals will recognise, and this gap then feeds back into a further loss of credibility (Fisher, 2023). For international populations this is the central problem. Diagnostic categories and test constructs are not neutral containers. They are historically specific ways of carving up experience. When a person's distress does not fit the available carving, the mismatch is attributed to the person rather than to the carving. Distress that is expressed through the body, through moral or spiritual language, through the family rather than the individual, or through idioms that have no clean English equivalent, has nowhere to land on a standard symptom scale. The scale then reports a low score, and the low score is read as absence of distress. 2.4 Contributory injustice and epistemic violence A third variety, developed after Fricker, is #contributory_injustice. Here a dominant group has access to the marginalised group's interpretive resources but refuses to use them, preferring its own. This is different from hermeneutical injustice because the concepts do exist. Somebody simply will not pick them up. When a research team knows that a community explains persistent low mood in terms of spiritual affliction and social rupture, and nonetheless writes a scale that asks only about sleep, appetite and anhedonia, the omission is a choice. Related to this is the notion of #epistemological_violence, which describes what happens when empirical results are interpreted in a way that constructs the other as inferior, even though other interpretations of the same data were equally available. Group differences on a test are a favourite site for this move. A difference is found, and the deficit explanation is chosen over the instrument explanation, though nothing in the data required that choice. 2.5 Why the concept is contested Students should also know that the concept is not accepted uncritically. Some scholars argue that epistemic injustice has been stretched so widely in psychiatry that it now covers almost any communication failure, and that it should be tightened to focus on the failure of professionals to take up first person testimony about what an experience is actually like (Pantazakos and Arnaud, 2024). Others point out that clinicians themselves can be victims of epistemic injustice, particularly when a practitioner discloses their own mental illness and finds their colleagues discounting them (Daly and Buedo, 2024). The concept is a tool, not a verdict. It is most useful when it is applied to specific practices, and #psychometrics is a field full of specific practices. 2.6 From philosophy to measurement The bridge from the philosophy to the psychometrics is this. A test is a machine for converting testimony into numbers. Every item is a question whose answer is taken as evidence. Every scoring rule is a decision about which answers count. Every norm table is a claim about who the relevant comparison group is. If a group's testimony is systematically mistranslated at any of these points, the machine will produce a number that is wrong in a patterned, non random way, and the wrongness will be invisible because it now wears the uniform of objectivity. #Score_interpretation is where philosophy and statistics finally meet. A Short and Uncomfortable History 3.1 The founding assumptions Modern testing grew out of a period in which the measurement of human ability was closely tied to projects of ranking populations. Intelligence testing in particular was used to support claims about the relative worth of national and ethnic groups, and immigration policy in several countries drew on test results whose linguistic and cultural bias is obvious in hindsight. Diagnostic categories were sometimes invented specifically to pathologise resistance to oppression, and modern commentators have drawn a direct line from those historical inventions to contemporary patterns of misdiagnosis in racialised communities (Okoroji, Mackay, Robotham, Beckford, and Pinfold, 2023). This history is not raised here in order to score points against dead scientists. It is raised because the assumptions that produced those results have not fully disappeared. The assumption that a construct defined in one place is a universal feature of the human mind, the assumption that a difference in scores reflects a difference in the underlying trait, and the assumption that the culture of the test maker is the neutral baseline against which others are variations, are all still operating. They are simply expressed more politely. 3.2 The WEIRD problem The best known modern statement of this problem is the observation that the human subjects of psychological research are overwhelmingly drawn from Western, educated, industrialised, rich and democratic societies, while conclusions are stated as if they applied to humanity. The consequences for #test_development are direct. If the item pool, the pilot sample, the factor structure, the reliability estimates and the norms all come from the same narrow population, then the instrument is not a measure of the human mind. It is a measure of that population's mind, presented as a measure of the human mind. Cross cultural neuropsychology has been unusually honest about this. Neuropsychological assessment rests on behavioural norms and values that developed in Western settings and may be biased for people whose cultural experience differs, and cultural and educational biases have been repeatedly identified within neuropsychological tests, particularly where normative data exist only for speakers of the host language or where the patient has had limited formal schooling (Nielsen, Franzen, Watermeyer, Jiang, Calia, Kjaergaard, Bothe, and Mukadam, 2024). Quality of education, not merely years of education, also matters, which means that two people with identical schooling on paper may not be comparable at all. 3.3 The response of professional bodies The field has responded, and the response has been real if uneven. The European Consortium on Cross-Cultural Neuropsychology was established in 2019 and has issued a position statement recommending that clinicians take account of linguistic factors, literacy, education, migration history, acculturation and other cultural factors. Notably, the consortium argues against race based norms as a solution, and instead calls for the development, validation and standardisation of more widely applicable tests that account for variation between individuals, together with better clinical training and guidelines for interpreter mediated assessment (Franzen and colleagues, 2022). That last point deserves emphasis because it is often misunderstood. Building separate norm tables by race is not #test_fairness. It can entrench the idea that the groups are essentially different, while leaving the biased instrument itself untouched. 3.4 Diagnostic manuals and cultural concepts Diagnostic systems have also moved. The introduction of the #Cultural_Formulation_Interview and of the notion of cultural concepts of distress into the fifth edition of the major diagnostic manual represented an attempt to build cultural inquiry into the assessment process rather than leaving it to individual sensitivity. Cultural concepts of distress cover cultural syndromes, cultural idioms of distress and cultural explanations of cause, and the interview is intended to let people describe their problems in their own words before those words are mapped onto professional categories (Aggarwal, Chen, and Lewis-Fernandez, 2023). Studies have shown that the interview can be used to elicit social stressors, social supports and the social determinants of health, and that it can even be used to work with concepts that patients bring in from popular culture and that clinicians may not initially recognise as meaningful (Ioannou and colleagues, 2023). The limits are equally instructive. Implementation research in the Netherlands concluded that using the interview properly requires a fundamental rethinking of the intake assessment, shifting it from a symptom oriented process to one that is centred on context and person (Baarnhielm, Rohlof, and DeMarinis, 2024). In other words, a culturally responsive instrument cannot simply be added to an unchanged system. The system resists it. 3.5 The lesson The historical lesson for a student who wants to build fairer tools is not that the field is irredeemable. It is that good intentions have repeatedly been defeated by structure. Individual clinicians have been culturally sensitive for a century. The instruments have remained biased anyway, because the bias is built into the objects, the norms and the workflows, not only into the attitudes of the people using them. Changing attitudes without changing objects produces sympathetic professionals administering unfair tests. Where Injustice Enters: The Test Life Cycle It is tempting to think of cultural bias as a translation problem. Translation is where most adaptation budgets are spent, and translation is the stage that funders and ethics boards ask about. But translation is only one of at least ten stages, and it is not the stage where the most serious damage is done. This section walks through the life cycle and names the injustice at each point. 4.1 Stage one: defining the construct Someone decides what is being measured. They decide that the thing called depression consists of low mood, loss of interest, sleep disturbance, appetite change, fatigue, guilt, concentration problems and thoughts of death. They decide that intelligence consists of a general factor with several broad abilities beneath it. They decide that a good employee is conscientious. These decisions are made inside a scholarly tradition, and they are made by people who are almost never members of the populations to whom the instrument will later be exported. This is where #contributory_injustice and hermeneutical injustice do their deepest work, because everything downstream inherits the definition. No amount of careful translation can rescue a construct that was never the right shape for the people being measured. If a community understands persistent distress as a rupture in relationships and obligations, and the instrument defines it as an individual mood state, then a perfect translation will produce a perfectly measured irrelevance. 4.2 Stage two: generating the items Items are written by people, and people write from their own lives. An item about enjoying social gatherings assumes a particular social world. An item asking whether the respondent feels like a failure assumes a self that can be evaluated as a unit and that is expected to have a personal trajectory. A memory task using names of common objects assumes that those objects are common. A visual reasoning task using a paper and pencil format assumes familiarity with the conventions of two dimensional representation and with the very idea of solving abstract puzzles for a stranger's benefit. Task familiarity is not a trivial matter. Performance depends on whether a person has ever done anything resembling the task before. Where the test relies on speeded responses, on the assumption that faster is better, or on the assumption that guessing is acceptable, cultural expectations about accuracy, deference and the appropriate way to behave with an authority figure will affect scores in ways that have nothing to do with the trait being measured. This is #construct_irrelevant_variance, and it is the technical name for a moral problem. 4.3 Stage three: translation and linguistic adaptation This stage is the most professionalised. Forward translation, back translation, expert committee review, reconciliation and pretesting form a well established pipeline, and detailed practical guidance for novice researchers now exists in the literature on cross cultural validation of measurement instruments. The pipeline is genuinely useful and it should be followed. But #back_translation has a well known weakness. A back translation can look excellent while the item remains culturally meaningless, because back translation checks semantic correspondence between two texts, not the psychological equivalence of the item for the two populations. If an item asks whether the respondent has felt blue, and the translation into another language produces a literal colour term that is then back translated correctly, the pipeline reports success and the item is nonsense. This is why translation should always be followed by cognitive interviewing, which asks real members of the target population to think aloud as they answer, and which frequently reveals that respondents are answering a different question from the one the researchers believed they were asking. 4.4 Stage four: piloting and item analysis At this stage items are dropped for poor statistical performance. It is worth pausing on what that means. An item that behaves oddly in the new population is often an item that is culturally important, because it is touching something that the original construct did not anticipate. The standard psychometric reflex is to delete it. The result is a cleaner scale that has been progressively purified of everything culturally distinctive, so that it converges back on the original instrument. The field then congratulates itself on having demonstrated that the construct is universal, when what it has actually done is discard the evidence against universality. This is one of the sharpest current criticisms of routine practice. Removing items to achieve statistical comparability changes the meaning of the construct being measured, and any comparison based on the surviving items becomes open to alternative interpretations (Fischer, Karl, Luczak-Roesch, and Hartle, 2025). #Item_deletion is not a neutral technical operation. 4.5 Stage five: norming Norms decide what counts as normal. If the reference sample is drawn from the host country's majority population, then every member of the minority or migrant population is being compared to a group they do not belong to. If the reference sample controls for years of education without considering the quality or the interruption of schooling, the correction is inadequate. If no local norms exist at all, the clinician is left improvising, and the improvisation will usually favour the interpretation that the professional finds most familiar. Studies in Europe have found that appropriate norms were not available for some, and in some clinics none, of the tests being used with diverse patients. The absence of norms is not experienced by the system as a crisis. It is experienced as an inconvenience, and the test is administered anyway. That difference in institutional response is itself a form of #epistemic_injustice, because it reveals whose measurement error the system is prepared to tolerate. 4.6 Stage six: administration The testing session is a social encounter with a power gradient. The examiner may not speak the test taker's first language. An interpreter may be present, and interpreter mediated assessment introduces its own well documented complications, since interpreters may be unfamiliar with neuropsychological terminology and procedures, since access to interpreter services varies enormously between countries, and since neuropsychologists are frequently required to conduct such assessments without ever having been trained to do so (Nielsen and colleagues, 2024). The politics of language choice matter too. Assigning an interpreter from a community in conflict with the test taker's own can silence the person completely. Stereotype threat, unfamiliarity with the testing situation, fear of the consequences of the assessment, especially where immigration status or child custody is at stake, and simple exhaustion all shape performance. None of these appear in the score. 4.7 Stage seven: scoring Scoring rules encode expectations. Verbal fluency tasks reward the ability to generate many words in a category, but the categories themselves vary in salience across environments. Similarity tasks reward abstract categorical responses over functional ones, and preference for functional reasoning is associated with different educational traditions rather than with lower ability. Scoring keys that were derived from one population are then applied to another without revision, and the mismatch is quietly converted into a deficit. 4.8 Stage eight: interpretation Here the clinician turns numbers into a story. This is the moment of greatest epistemic risk, because interpretation is where the deficit explanation and the instrument explanation compete, and the deficit explanation is usually more convenient. It requires no revision of the tool, no additional data collection, and no admission that the professional's instruments do not work. #Clinical_interpretation should therefore always include an explicit consideration of the alternative hypothesis that the score is an artefact. 4.9 Stage nine: feedback Most discussions of test bias stop at interpretation. But the person is usually told something, and what they are told matters. Feedback delivered in a language the person does not fully command, using categories they do not accept, without any invitation to correct the account, completes the circuit of testimonial injustice. The person's own understanding of their situation is not simply ignored. It is formally overwritten by a document. 4.10 Stage ten: institutional use Finally, the score enters an institution. It becomes an admission decision, a diagnosis, a treatment plan, a legal finding or an employment outcome. At this stage the number has been fully separated from its origins. Nobody downstream knows that the norms were wrong, that the interpreter was untrained or that three items were culturally meaningless. The uncertainty has evaporated and the authority remains. This is why fixing tests requires fixing the reporting conventions that surround them, not only the tests themselves. The Psychometric Toolkit and Its Limits This section explains, in plain language, the statistical machinery the field uses to detect bias. It also reports an argument that is currently unresolved and that students will not usually find in an introductory textbook. 5.1 Equivalence: four questions in order Before any comparison across groups is legitimate, four questions have to be answered. The first is conceptual equivalence. Does the construct exist in both groups, and does it mean the same thing? This is not a statistical question. It is answered through ethnography, interviews, focus groups and consultation with the community, and it must be answered first. A great deal of cross cultural research skips it entirely and moves straight to translation, which is like checking whether the map is correctly printed without checking whether it is a map of the right country. The second is linguistic equivalence. Do the words mean the same thing? This is the translation question described earlier. The third is #metric_equivalence. Do the items relate to the underlying construct in the same way in each group? Statistically, this is asked by testing whether the factor loadings are the same across groups. The fourth is scalar equivalence. Given the same level of the underlying trait, do people in different groups produce the same score? Statistically, this asks whether the item intercepts are the same. Without scalar equivalence, mean scores cannot be compared, and yet mean scores are compared constantly. 5.2 Measurement invariance The formal procedure that tests the last two questions is #measurement_invariance testing, usually conducted through multi group confirmatory factor analysis. The logic is a ladder. First the same factor structure is fitted in every group, which is configural invariance. Then the loadings are constrained to be equal, which is metric invariance. Then the intercepts are constrained to be equal, which is scalar invariance. Then, if desired, the residual variances are constrained, which is strict invariance. At each step, model fit is examined. If fit worsens badly when a constraint is imposed, that constraint fails, and the comparison it would have licensed is not permitted. The idea has a clean interpretation. Establishing invariance is a test for the presence of bias, in the sense that two individuals with the same underlying ability who come from different populations should obtain the same score on the test (Wilson and colleagues, 2023). If they do not, something other than the trait is influencing the score. The state of the art in this area has been reviewed thoroughly, covering the historical development of invariance methods, the methodological challenges and the open problems (Leitgob and colleagues, 2023). Students planning cross cultural comparisons should read that literature before collecting data, not after. 5.3 Differential item functioning A closely related tool is #differential_item_functioning analysis, usually called DIF. DIF asks a simple question at the level of the single item. Among people who have the same level of the underlying trait, do members of different groups have different probabilities of endorsing this item or answering it correctly? If yes, the item is functioning differently, and it is a candidate for bias. DIF has produced genuinely useful findings. In studies of everyday functioning across European countries, items such as using a washing machine, making appointments or playing card and board games showed differential functioning, indicating that subtle cultural bias can be present even in scales that look entirely innocuous. Nobody sitting in a research meeting would predict that a washing machine could bias a cognitive measure. The data revealed it. This is a good example of why statistical bias detection remains indispensable even after the strongest participatory design work has been done. 5.4 The evidence that constructs do generalise It is important to be balanced. The evidence is not uniformly negative. A systematic review of the factorial invariance of cognitive ability assessments across cultures, covering fifty seven original studies, found results that were broadly supportive of the generalisability of cognitive ability constructs, and compatible with the well established taxonomy that organises intelligence into a hierarchy of broad abilities (Wilson, Bowden, Byrne, Joshua, Marx, and Weiss, 2023). Related work has found strong or strict factorial invariance for a widely used children's intelligence scale across several countries. What does this mean? It means that the deep structure of certain cognitive constructs appears to hold up reasonably well across the populations that have been studied. It does not mean the tests are fair. A test can measure the same construct in two groups and still disadvantage one of them badly through unfamiliar task formats, inappropriate norms, language demands and the conditions of administration. #Structural_validity and fairness are not the same thing, and conflating them is one of the most common errors in this literature. 5.5 The argument the textbooks do not mention Here is the controversy. Two recent lines of criticism have challenged the routine use of invariance testing in cross cultural work, from opposite directions. The first argues that invariance testing is being used badly and that its results are widely misinterpreted. Removing items to achieve invariance reduces the available variance, biases the invariance tests themselves, and changes what the construct means. Invariance testing is also somewhat circular, because a latent variable cannot be independently verified within the model. The proposed remedy is to bring external and predictive validity into the process from the beginning, rather than treating invariance as an internal accounting exercise (Fischer, Karl, Luczak-Roesch, and Hartle, 2025). The second, and more radical, argues that strict invariance standards are rarely appropriate for comparative research in psychology at all. Invariance is very easily violated in large international studies, which has led some scholars to discourage cross national study of well known constructs entirely, and practitioners are now caught between contradictory recommendations, unsure whether strict standards actually impede culturally conscious research (Kusano, Napier, and Jost, 2025). If the bar is set so high that no cross cultural comparison ever clears it, then the effect is not fairness. The effect is silence, and silence about non Western populations is exactly the outcome the field has been trying to escape. 5.6 What a student should take from this Three practical conclusions follow. First, run the invariance and DIF analyses. They catch things that no amount of good intention will catch. Second, do not treat a failed invariance test as a reason to abandon the comparison, and do not treat a passed invariance test as a certificate of fairness. Report the result, interpret it, and be honest that the field disagrees about what it means. Third, and most important, remember that all of this machinery operates downstream of construct definition. Invariance testing can tell you whether an instrument behaves the same way in two groups. It cannot tell you whether the instrument was ever asking the right question. Only the people being measured can tell you that, and only if you ask them. This is the point at which #psychometrics needs #qualitative_methods, and where the epistemic and the statistical arguments finally converge. 5.7 A note on causal thinking A useful development in recent years has been the effort to bring explicit causal reasoning into cross cultural generalisation, which forces researchers to state what they are assuming about the mechanisms that link culture, exposure, task and response (Deffner, Rohrer, and McElreath, 2022). This is a considerable improvement on the older practice of gathering data in several countries and comparing means with no theory at all about why the groups should differ. If a researcher cannot say what causal path they think links the group variable to the score, then the comparison is not a scientific claim, it is a table. International Populations: Three Situations The general argument becomes clearer when applied to specific groups. The three examined here are chosen because they are common in university and clinical settings and because they show different faces of the same problem. 6.1 International students International students are a large, growing and often invisible clinical population. They typically sit at the intersection of several risk factors. They are separated from their support networks, they are under financial and academic pressure, they may hold visas whose renewal depends on academic performance, and they are frequently reluctant to seek help because mental illness carries different meanings and different consequences in their home communities. Campus counselling services routinely screen with brief symptom questionnaires. These instruments were developed and normed largely on domestic student populations. Several problems follow. The first is #symptom_expression. Distress that presents through the body, through headaches, stomach problems, fatigue or sleep disturbance, may be reported honestly and still produce a low score on a scale weighted towards cognitive and affective symptoms. The student is distressed. The instrument says they are fine. The second is the reference period and the response format. Scales that ask about the last two weeks assume a person who can locate their experience on a calendar of standard weeks, and rating formats assume a comfort with graded self quantification that is culturally variable. Extreme response styles and acquiescence bias differ systematically across cultures, and they contaminate scores in ways that are not visible in a total. The third is the meaning of the assessment itself. A student who suspects that a disclosure could reach their sponsor, their embassy or their department will not disclose. The resulting low score is not a measurement of their mental state. It is a measurement of their justified caution. The fourth is the feedback loop. Because international students score lower on standard screening instruments, services conclude that they need fewer resources, allocate fewer resources, and see fewer international students, which is then read as confirmation that the need is low. #Under_detection becomes self confirming. 6.2 Migrants and refugees in clinical assessment For migrants and refugees the stakes are usually higher and the instruments are usually more consequential. Cognitive assessment in this population has been the subject of sustained work by cross cultural neuropsychologists, and the picture that emerges is sobering. Language barriers have repeatedly been identified as one of the main challenges to cross cultural neuropsychological assessment in migrant populations, and clinicians are commonly required to conduct interpreter mediated assessments without any relevant graduate training (Nielsen and colleagues, 2024). Assessment in a patient's non native language is likely to produce biased results, which is why interpreter mediated assessment is often necessary despite its own difficulties. Formal education is the single most important confound. Many tests assume literacy, familiarity with pencil and paper tasks, comfort with timed performance and experience with abstract classification. A person with little or interrupted schooling may perform poorly on such tasks while functioning perfectly well in their own life. Diagnosing dementia in this situation is genuinely difficult, and the profession has responded with the development of cross cultural test batteries designed for use across languages and educational backgrounds, including instruments validated for the assessment of mild cognitive impairment across European populations and multicultural scales developed specifically to be less dependent on formal schooling (Delgado-Alvarez and colleagues, 2023; Fernandez, Arriondo, Folmer, Vaiman, Leite, and Hardy, 2022). Trauma assessment adds a further layer. Standardised post traumatic stress instruments were developed in particular clinical traditions and they assume that the person will narrate their experience in a specific order and with specific emotional markers. Where a survivor's account is fragmented, or organised around shame or moral injury rather than fear, the instrument may record a lower score, and in asylum contexts a lower score can be read as evidence of fabrication. Here #testimonial_injustice acquires legal force. 6.3 Indigenous and Global South populations The third situation is the largest and the least often discussed in student training. Instruments developed in high income countries are exported at scale to low and middle income countries, often through global mental health programmes that aim to expand access to care. The intention is good. The epistemic effect is complicated. Critics have long argued that the export of psychiatric categories to the majority world risks displacing local frameworks of understanding, and recent work has extended this critique to implementation science, calling for approaches that are dialogical, situated and oriented towards justice rather than towards the efficient delivery of pre packaged models. Concrete work on epistemic injustice in the mental health care of Indigenous populations has begun to appear, with implications drawn explicitly for global mental health practice (Faruk, 2025). The pattern documented is consistent. Local healers, local categories and local explanations of suffering are treated as obstacles to be managed rather than as knowledge to be incorporated. When an instrument is then adapted, the adaptation is linguistic and cosmetic, and the underlying model remains untouched. This is #contributory_injustice operating at the level of an entire health system. A Framework for Developing Culturally Responsive Assessment Tools This section sets out a staged framework. It is intended to be practical enough for a masters thesis and rigorous enough for a funded project. The stages map onto the life cycle described in section four, and the guiding principle is that at every stage there is a decision, and at every decision the community being measured should have a voice. 7.1 Stage one: establish shared governance before anything else Before a single item is written, form a group that includes members of the target population, and give that group real authority, not advisory status. Real authority means the power to veto items, to reject the construct, and to control how findings are reported. If the group can only comment while the researchers decide, then the project has reproduced the structure it claims to be dismantling. This is #participatory_design applied to psychometrics. Lived experience involvement and leadership are widely proposed as a way of combating epistemic injustice, precisely because they ensure that the views of the people at the centre of an issue can inform decisions (Okoroji, Mackay, Robotham, Beckford, and Pinfold, 2023). But involvement can easily become tokenistic. The test of authenticity is whether the community has ever changed a decision that the researchers did not want changed. Practical steps include agreeing a memorandum of understanding, agreeing in advance on data ownership, agreeing on authorship, agreeing on what happens if the findings are unflattering to a funder, and budgeting to pay community members properly for their time. Unpaid participation by marginalised people in a funded research project is itself an epistemic and economic injustice. 7.2 Stage two: map the construct from the ground up Do not begin with an existing instrument. Begin with the phenomenon. Use free listing, semi structured interviews, focus groups and observation to build a picture of how the population in question actually understands and expresses the domain in question. Ask what the local words are, who is consulted when a person has this problem, what causes it, what makes it worse, what makes it better and what would count as recovery. This produces a #construct_map. Compare that map to the construct as defined in the existing instrument. Three things can happen. Some elements overlap, and those elements are candidates for shared items. Some elements are present in the existing instrument but absent locally, and these should be flagged as potentially irrelevant. Some elements are present locally but absent from the instrument, and these are the most valuable finding in the entire project. They are the evidence of hermeneutical gap. 7.3 Stage three: decide honestly between adaptation and construction At this point make an explicit decision, and document it. There are three options. Adaptation takes an existing instrument and modifies it. This is appropriate when the construct map largely overlaps and when comparability with international data is a genuine priority. It is the cheapest option and it is the one that funders prefer. Augmentation keeps the core of the existing instrument and adds a locally developed module. This preserves comparability while capturing local content, and it is often the best compromise. The augmented module is scored separately and reported alongside the core score, never averaged into it. Construction builds a new instrument from the ground up. This is appropriate when the construct map diverges substantially. It is expensive, it is slow, and it produces a tool that cannot easily be compared with international data. Sometimes that is exactly right, and the inability to compare is a finding rather than a failure. The important thing is that the decision should follow the construct map, not the budget. #Cultural_adaptation that was chosen because construction was unaffordable should be described as such in the limitations section. 7.4 Stage four: generate and translate items collaboratively Items should be drafted by mixed teams that include community members, and drafted in the target language where possible rather than translated into it. Where translation is necessary, follow the established pipeline of forward translation by at least two independent translators, reconciliation, back translation and expert committee review, and then treat the output as a draft rather than as a finished product. Pay attention to who the translators are. A translator who is a highly educated professional from the same country as the target population may be culturally very distant from a rural, low literacy respondent. Translation is a class relation as well as a language relation. 7.5 Stage five: cognitive interviewing This stage is non negotiable and it is the one most often skipped. Sit with members of the target population, one at a time, and ask them to answer each item while thinking aloud. Then ask them what they thought the item meant, how they arrived at their answer, what other answers they considered and whether the response options fitted what they wanted to say. #Cognitive_interviewing regularly reveals that respondents are answering a different question from the one that was asked, that response options are not ordered as the designers assumed, that particular words carry moral weight that suppresses honest answering, and that the recall period is not usable. Twenty cognitive interviews will teach a research team more about their instrument than a thousand completed questionnaires. 7.6 Stage six: pilot, and treat item removal as a substantive decision Pilot the instrument, run the item analysis, and then treat every proposed item deletion as a decision with content, not as a statistical formality. Ask why the item is misbehaving. Take the question back to the governance group. If an item is central to the local construct but statistically awkward, consider whether the problem lies with the item or with the model being fitted to it. Document the items that were dropped and why, and publish that documentation. #Transparent_reporting of deleted items is one of the cheapest and most powerful reforms available. 7.7 Stage seven: build norms that are fit for the actual population Wherever resources allow, collect local normative data. Where they do not, be explicit about which reference group is being used and about what that implies. Consider using regression based norms that model the effects of education, literacy, language dominance, acculturation and age simultaneously, rather than crude group based tables. Follow the professional consensus that argues against race based norms and in favour of instruments and norms that account for variation between individuals (Franzen and colleagues, 2022). Where norms genuinely do not exist, the ethical response is not to administer the test anyway and interpret it as if norms existed. The ethical response is to say, in the report, that the instrument cannot be interpreted normatively for this individual, and to rely on other sources of evidence. #Norm_transparency should become a standard element of every psychological report. 7.8 Stage eight: validate against outcomes that matter locally Validity is not an internal property of a set of correlations. Ask whether the scores predict something that the community itself recognises as important. Does a high score identify the people whom the community regards as suffering? Does it predict who seeks help, who recovers, who is able to work, who is supported? External and predictive validity should be built into the design from the beginning rather than added afterwards, which is a point that the psychometric critics themselves have pressed (Fischer and colleagues, 2025). 7.9 Stage nine: design the report, not only the test Decide in advance what the score will be allowed to mean and write that into the reporting template. A culturally responsive report should state the language of administration, the interpreter arrangements, the norms used, the known limitations of the instrument for this population, and the alternative interpretations that were considered and rejected. It should include the person's own account of their difficulties in their own words, and it should record any disagreement between the person and the assessor. #Reporting_standards are where epistemic humility becomes institutional practice rather than personal virtue. 7.10 Stage ten: return the findings to the community Give the results back. Present them to the governance group, to participants and to the wider community in an accessible form, and invite challenge. Publish the instrument and its manual openly where licensing allows, so that the population that helped build it is not then required to pay for access to it. Methods for Practice The previous section addressed test builders. This section addresses the clinician or researcher who has to use existing tools tomorrow morning, in an imperfect system, with the instruments that are actually on the shelf. 8.1 Start with the person's own account Before administering anything, ask the person what the problem is, in their own words, and record those words. Ask what they call it, what they think caused it, what they fear, what they have already tried, whom they have consulted and what they expect from this encounter. This is the logic of the #Cultural_Formulation_Interview, which is available as a structured tool with supplementary modules that probe explanatory models in more depth (Aggarwal, Chen, and Lewis-Fernandez, 2023). The interview is not merely a data collection device. Attention to a person's own idioms of distress, which are the shared, culturally shaped ways in which people express suffering, has been argued to hold therapeutic potential in itself, because being understood in one's own terms is a form of care. The formal purpose of the interview is to enhance the cultural validity of the diagnostic assessment, to help with treatment planning and to support engagement. 8.2 Treat the standardised score as one witness among several A score is evidence. It is not a verdict. Where the score conflicts with the person's own account, with collateral information, with observed functioning or with the clinician's own impression, the conflict must be named in the report rather than resolved silently in favour of the number. #Convergent_evidence should be the standard, and the burden of proof should sit with the instrument when its cultural fit is doubtful. 8.3 Work properly with interpreters Interpreter mediated assessment is a skill, and the evidence indicates that most practitioners have never been taught it. The published recommendations are clear and worth following. Use trained professional interpreters rather than family members, and never a child. Brief the interpreter before the session about the purpose of the assessment, about the need for literal rather than smoothed translation, and about the fact that a strange sounding answer may be clinically important and must not be tidied up. Discuss in advance which items cannot be meaningfully translated. Debrief afterwards and record the interpreter's observations. Be aware of the wider politics of language and community, since the choice of interpreter can silence a person completely. Recognise that assessment with an interpreter may never be entirely free of bias, and say so in the report. Note also that responsibility for effective communication rests with the clinician and with the institution that employs them, not with the patient (Nielsen and colleagues, 2024). #Interpreter_mediated_assessment is an organisational obligation, not a personal favour. 8.4 Separate the description of performance from the inference of impairment A person can perform poorly on a task without being impaired. Write reports that say what happened, describe the conditions, and then state the inference separately with its uncertainty attached. The sentence structure matters. There is a large difference between saying that a person is impaired in verbal memory and saying that they scored below the reference range on a word list task administered through an interpreter in their third language, using norms derived from a population with substantially more formal education. 8.5 Use mixed methods in research designs Quantitative bias detection and qualitative meaning making are complements, not rivals. A well designed study of an adapted instrument should include invariance testing, DIF analysis, cognitive interviews and qualitative validation against community judgements. Where the statistics and the interviews disagree, that disagreement is a finding and should be published as one. #Mixed_methods designs are the natural methodological expression of an epistemic justice commitment, because they build in a channel through which the population can contradict the numbers. 8.6 Keep an assessment bias log A small and practical suggestion. Clinical services should keep a record of the occasions on which an instrument could not be used appropriately, of the norms that were missing, of the interpreter arrangements that failed and of the items that patients found meaningless. Over a year this log becomes evidence. Evidence is what changes procurement decisions, and procurement decisions determine which tests sit on the shelf. Digital and Algorithmic Assessment: The Next Frontier of the Same Problem 9.1 The promise Digital assessment offers real benefits for international populations. Remote administration can bring scarce specialists to people who cannot reach them. Adaptive testing can shorten sessions. Automatic translation and speech processing might, in principle, reduce dependence on scarce interpreters, and remote assessment may help ensure access to bilingual clinicians in the future. 9.2 The risk The risk is straightforward. The field may encode existing bias into systems that are faster, cheaper and much harder to challenge. Several technical, ethical, legal, assessment and training issues need to be resolved before such technologies are implemented widely (Nielsen and colleagues, 2024). Three deserve particular attention. #Training_data is the first. A model trained on clinical records from one health system inherits that system's diagnostic patterns, including its historical misdiagnosis of minority patients. The model does not know that these were errors. It learns them as ground truth and reproduces them at scale. #Algorithmic_opacity is the second. When a clinician makes a culturally biased inference, the person can, at least in principle, argue with them. When a system produces a risk score from a model that nobody can inspect, the ground on which to argue disappears. Contesting a score requires access to the reasoning behind it. Removing that access removes the possibility of contest, which is epistemic injustice implemented in software. #Data_extraction is the third. Digital assessment tools are frequently built by collecting data from populations who have no ownership of the resulting product and no share in its value. The community supplies the raw material of the model and then pays a licence fee to use the tool. Governance of this relationship needs to be settled before the tools are built, not after. 9.3 What responsible digital assessment would require The same framework applies. Involve the population in defining the outcome the model is meant to predict. Test for differential performance across groups and publish the results, not merely the overall accuracy. Retain a human channel through which a person can contest a score and have their contestation recorded. Keep the system interpretable enough that a clinician can explain to a patient why it said what it said. And be willing to conclude that a model which performs unequally should not be deployed, even when the average performance looks impressive. An average is a poor guide to justice, because it is precisely the mechanism by which minority error is hidden inside majority success. Ethics, Governance and Training 10.1 Consent as a process Informed consent in cross cultural assessment is often treated as a signature collected once and then filed, rather than as an ongoing conversation, and this bureaucratic model has been criticised as one of the ways in which colonial dynamics persist in global mental health research (ElChaar, Oliver, Brown, and Roberts, 2026). A person who does not understand what the test measures, who will see the result, what the result can be used for and what happens if they decline has not consented in any meaningful sense. #Informed_consent should be revisited during the assessment, not only at its start. 10.2 The right to contest a score Consider adding an explicit right of reply to assessment reports. The person is invited to read the report, or to have it read to them in their own language, and to record their disagreement, which is then attached to the document. This costs almost nothing and it directly addresses testimonial injustice, because it restores the person's status as a witness to their own life. It also produces useful data about where instruments fail. 10.3 Training Cross cultural assessment is still taught, in most programmes, as an optional module or a single lecture. The professional consensus in cross cultural neuropsychology calls for improved clinical training in culturally sensitive assessment and for guidelines on interpreter mediated work (Franzen and colleagues, 2022). For students reading this article, the implication is direct. If your programme does not teach these skills, you will have to acquire them yourself, and you should assume that competence with a manual is not competence with a person. 10.4 Who does the research Finally, the composition of the research workforce matters. A field in which the people who build the instruments are drawn from a narrow set of countries and institutions will keep producing instruments that reflect that narrowness, however careful the individual researchers are. Funding, authorship, editorial boards and doctoral admissions are all part of the #epistemic_justice question, and they are not solved by better statistics. Illustrative Cases The following composite cases are constructed for teaching purposes. They are not reports of specific individuals. 11.1 The screened out student A postgraduate student from South Asia attends a university counselling service after failing to submit work for two months. She reports headaches, difficulty sleeping and a feeling of heaviness. On a standard depression screener she scores below the clinical threshold, because several items ask about self critical thoughts and loss of pleasure in a vocabulary she does not use about herself, and because she is careful about what she writes down while her visa depends on her enrolment. She is offered study skills support and discharged. Six months later she withdraws from her programme. The failure here is not the clinician's compassion, which was present. It is the instrument, the threshold, the norms and the institutional workflow that treated a score as a decision rule. A culturally responsive service would have used the score as one input, asked directly about somatic expression, and given weight to the student's own account of what was happening. 11.2 The misdiagnosed elder An older man who migrated as a labourer thirty years ago is referred for memory assessment. He completed four years of schooling in a rural area and speaks the host language conversationally but not fluently. Testing is conducted with an interpreter who has no experience of neuropsychological work and who quietly corrects and completes his answers. He scores in the impaired range on word list learning, verbal fluency and a task requiring him to copy a complex figure. A diagnosis of dementia is recorded. Every element of the score is contaminated. The norms assume more education. The tasks assume literacy and familiarity with abstract drawing. The interpreter smoothed the data. A culturally responsive assessment would have used a battery designed for low education populations, briefed the interpreter, gathered a detailed history of functioning from the family, and stated in the report that a normative interpretation was not possible. 11.3 The instrument that improved A research team working with a refugee community set out to adapt an anxiety scale. Cognitive interviews revealed that four items were meaningless, that respondents interpreted the frequency options in reverse, and that the community's central experience of anxiety involved fear for family members left behind, which no item captured. The team added a locally developed module, reported it separately, and published the deleted items and the reasons for deletion. The resulting tool identified people whom the original scale had missed, and the community group retained the right to approve any future use of the instrument. This is what #epistemic_justice looks like when it is operationalised rather than merely announced. Limitations and a Research Agenda 12.1 Limitations of this article This is a conceptual and integrative paper, not a systematic review, and its selection of literature is therefore partial. It draws mainly on English language sources, which is an irony that should be stated plainly rather than hidden. It also treats broad categories such as migrants and international students as if they were coherent, when in reality the variation within each group is enormous. Finally, the proposed framework has not been empirically tested as a whole, although each of its components has support in the existing literature. 12.2 What needs to be studied Several questions are open and would make good doctoral projects. Does participatory item development actually improve predictive validity, or only acceptability? The claim is plausible and it is largely untested. How much does the quality, as opposed to the quantity, of education explain apparent cognitive differences between migrant and host populations? This requires better measures of educational quality than the field currently possesses. Can DIF and invariance analyses be combined with cognitive interviewing in a formal way, so that qualitative evidence about item meaning is used to interpret statistical misfit rather than merely to illustrate it? What are the effects on outcomes, not merely on satisfaction, of adding a right of reply to psychological reports? And, urgently, how do current #digital_assessment and algorithmic tools perform across linguistic and cultural groups, and who is checking? 12.3 A note to students The most useful contribution a student can make is often not a new instrument. It is a careful, honest study of an existing instrument in a population where it has never been examined, published together with the items that failed and the reasons why. That paper will be cited, it will be used, and it will prevent harm. Negative and inconvenient findings about widely used tests are among the most valuable and least published contributions in the whole field. Conclusion Psychological testing is a practice of listening. It listens through a highly structured, heavily filtered and statistically processed channel, but it is listening nonetheless, and everything that can go wrong in ordinary listening can go wrong here as well, only more quietly and with more authority. This article has argued that the problems facing international populations in assessment are not merely technical. They are epistemic. When a test is built without the participation of the people it will be used on, their ways of understanding themselves are excluded before the first item is written. When their difficulties with the instrument are recorded as deficits, their testimony is discounted. When their concepts are known but not used, the exclusion becomes a choice. And when the resulting numbers travel into institutions that cannot see the uncertainty behind them, the injustice is laundered into objectivity. The path forward is not the abandonment of measurement. Measurement, done well, is one of the few tools that can make invisible suffering visible and hold institutions to account. The path forward is measurement that is built with people rather than about them, that reports its own limits honestly, that treats a failed item as information rather than as noise, and that leaves open a door through which the person being measured can say, this is not me, and be heard. That is a modest standard. It is also, at present, an unmet one. Meeting it is work that the next generation of psychologists will have to do, and it begins with the unglamorous, unfashionable and entirely necessary decision to ask the people you are measuring what they think you are measuring. Hashtags #Epistemic_Injustice_In_Testing #Culturally_Responsive_Assessment #Cross_Cultural_Psychometrics #Test_Fairness_For_International_Students #Decolonising_Assessment #Measurement_Equivalence #Global_Mental_Health_Assessment #Migrant_And_Refugee_Mental_Health #Participatory_Test_Development #Cultural_Validity #Assessment_Ethics #Psychology_Students #Bias_In_Psychological_Tests #Inclusive_Psychometrics #Justice_In_Measurement References Aggarwal, N. K., Chen, D., and Lewis-Fernandez, R. (2023). Eliciting social stressors, supports, and determinants of health through the DSM-5 cultural formulation interview. Frontiers in Psychiatry, 14, 1148170. https://doi.org/10.3389/fpsyt.2023.1148170 Baarnhielm, S., Rohlof, H., and DeMarinis, V. (2024). Editorial: Clinical implementation of the DSM-5 cultural formulation interview. Frontiers in Psychiatry, 15, 1520122. https://doi.org/10.3389/fpsyt.2024.1520122 Daly, T., and Buedo, P. (2024). Psychiatrists can also experience epistemic injustice: Reconnecting with ethos in mental health. Philosophy of Medicine, 5(1). https://doi.org/10.5195/pom.2024.209 Deffner, D., Rohrer, J. M., and McElreath, R. (2022). A causal framework for cross-cultural generalizability. Advances in Methods and Practices in Psychological Science, 5(3). https://doi.org/10.1177/25152459221106366 Delgado-Alvarez, A., Nielsen, T. R., Delgado-Alonso, C., Valles-Salgado, M., Lopez-Carbonero, J. I., Garcia-Ramos, R., and Matias-Guiu, J. A. (2023). Validation of the European Cross-Cultural Neuropsychological Test Battery (CNTB) for the assessment of mild cognitive impairment due to Alzheimer's disease and Parkinson's disease. Frontiers in Aging Neuroscience, 15, 1134111. https://doi.org/10.3389/fnagi.2023.1134111 ElChaar, S., Oliver, E., Brown, F. L., and Roberts, B. (2026). Decolonising informed consent in global mental health research. PLOS Mental Health. https://doi.org/10.1371/journal.pmen.0000535 Faruk, M. O. (2025). Addressing epistemic injustice in the mental healthcare of Indigenous people in Bangladesh: Implications for global mental health. Cambridge Prisms: Global Mental Health, 12. Fernandez, A. L., Arriondo, G. J., Folmer, M., Vaiman, M., Leite, G. R., and Hardy, D. J. (2022). The Multicultural Neuropsychological Scale (MUNS): Validity, reliability, normative data and cross-cultural evidence. Culture and Brain, 10(2), 167-193. Fischer, R., Karl, J. A., Luczak-Roesch, M., and Hartle, L. (2025). Why we need to rethink measurement invariance: The role of measurement invariance for cross-cultural research. Cross-Cultural Research. https://doi.org/10.1177/10693971241312459 Fisher, J. (2023). Epistemic injustice: The silenced voices. International Journal of Mental Health Nursing, 32(4), 1186-1189. https://doi.org/10.1111/inm.13163 Franzen, S., Watermeyer, T. J., Pomati, S., Papma, J. M., Nielsen, T. R., and colleagues, on behalf of the European Consortium on Cross-Cultural Neuropsychology (2022). Cross-cultural neuropsychological assessment in Europe: Position statement of the European Consortium on Cross-Cultural Neuropsychology (ECCroN). The Clinical Neuropsychologist, 36(3). https://doi.org/10.1080/13854046.2021.1981456 Fricker, M. (2007). Epistemic injustice: Power and the ethics of knowing. Oxford University Press. Ioannou, M., Olsson, S., Wold, A. B., Dellepiane, M., and Steingrimsson, S. (2023). Approaching highly sensitive person as a cultural concept of distress: A case study using the cultural formulation interview in patients with bipolar disorder. Frontiers in Psychiatry, 14, 1148646. https://doi.org/10.3389/fpsyt.2023.1148646 Kidd, I. J., Spencer, L., and Carel, H. (2022). Epistemic injustice in psychiatric research and practice. Philosophical Psychology, 36(3), 503-531. https://doi.org/10.1080/09515089.2022.2156333 Kidd, I. J., Spencer, L., and Harris, E. (2023). Epistemic injustice should matter to psychiatrists. Philosophy of Medicine, 4(1). Kusano, K., Napier, J. L., and Jost, J. T. (2025). The mismeasure of culture: Why measurement invariance is rarely appropriate for comparative research in psychology. Personality and Social Psychology Bulletin. https://doi.org/10.1177/01461672251341402 Leitgob, H., Seddig, D., Asparouhov, T., Behr, D., Davidov, E., De Roover, K., Jak, S., Meitinger, K., Menold, N., Muthen, B., Rudnev, M., Schmidt, P., and van de Schoot, R. (2023). Measurement invariance in the social sciences: Historical development, methodological challenges, state of the art, and future perspectives. Social Science Research, 110, 102805. https://doi.org/10.1016/j.ssresearch.2022.102805 Newbigging, K., Salla, A., Schon, U.-K., and King, C. (2024). Editorial: Addressing epistemic injustice in mental health. Frontiers in Psychiatry, 15, 1382528. https://doi.org/10.3389/fpsyt.2024.1382528 Nielsen, T. R., Franzen, S., Watermeyer, T., Jiang, J., Calia, C., Kjaergaard, D., Bothe, S., and Mukadam, N., on behalf of the European Consortium on Cross-Cultural Neuropsychology (2024). Interpreter-mediated neuropsychological assessment: Clinical considerations and recommendations from the European Consortium on Cross-Cultural Neuropsychology (ECCroN). The Clinical Neuropsychologist, 38(8), 1775-1805. https://doi.org/10.1080/13854046.2024.2335113 Okoroji, C., Mackay, T., Robotham, D., Beckford, D., and Pinfold, V. (2023). Epistemic injustice and mental health research: A pragmatic approach to working with lived experience expertise. Frontiers in Psychiatry, 14, 1114725. https://doi.org/10.3389/fpsyt.2023.1114725 Pantazakos, T., and Arnaud, S. (2024). Determining the scope of epistemic injustice within psychiatry. Philosophical Psychology. https://doi.org/10.1080/09515089.2024.2377291 Wilson, C. J., Bowden, S. C., Byrne, L. K., Joshua, N. R., Marx, W., and Weiss, L. G. (2023). The cross-cultural generalizability of cognitive ability measures: A systematic literature review. Intelligence, 98, 101751. https://doi.org/10.1016/j.intell.2023.101751 Wilson, C. J., Bowden, S. C., Byrne, L. K., Vannier, L.-C., Hernandez, A., and Weiss, L. G. (2023). Cross-national generalizability of WISC-V and CHC broad ability constructs across France, Spain, and the US. Journal of Intelligence, 11(8), 159. https://doi.org/10.3390/jintelligence11080159

Social work is a profession built on the promise of help, yet it operates through relationships that are never equal. The practitioner holds legal authority, controls access to resources, writes the official record, and speaks the language of the institution. The service user arrives with a life history, a set of resources, and a way of speaking that may or may not be recognised as legitimate. This article argues that the sociology of Pierre Bourdieu gives social work a precise and usable vocabulary for describing what happens in that gap. Using the concepts of #habitus, #field, #cultural_capital, #symbolic_power and #symbolic_violence, the paper develops a conceptual framework that treats the social work encounter as a structured space of struggle rather than a neutral meeting of two individuals. The article proceeds through a critical narrative synthesis of recent Bourdieusian scholarship in social work, social care and health, published mainly between 2021 and 2025. It sets out how the three states of cultural capital (embodied, objectified and institutionalised) shape assessment, eligibility decisions, court reports and case closure. It then examines how the professional habitus of the social worker, formed by class background, training and organisational pressure, quietly decides which forms of capital count. From this analysis the paper proposes a five part practice framework: reflexive objectivation, capital mapping, linguistic justice, recognition joined to redistribution, and field level advocacy. Three composite practice scenarios illustrate the framework in child protection, adult social care and youth justice. The article ends by taking seriously the standard criticisms of Bourdieu, especially the charge of determinism, and argues that these criticisms sharpen rather than defeat the framework. The main claim is simple. Social workers cannot escape power, but they can learn to see it, name it, and use it with far greater honesty. Keywords: Bourdieu; habitus; cultural capital; symbolic violence; power; social work practice; reflexivity; welfare state 1. Introduction Every social work intervention begins with an encounter. A worker knocks on a door, opens a file, sits across a table, and asks questions. On the surface this looks like two people talking. Underneath, something more complicated is taking place. One person has been trained, credentialed, employed and mandated by the state. The other has usually been referred, reported, assessed or summoned. One will write a document that carries legal weight. The other will be described in it. Social work has long been aware of this imbalance. #Anti_oppressive_practice, #empowerment models, strengths based approaches and #co_production have all tried to soften it. What has been harder to achieve is a shared analytical language for describing exactly how the imbalance works in the ordinary details of practice: in the way a parent phrases an answer, in the way a worker interprets a silence, in the way an assessment form rewards some kinds of knowledge and ignores others. This is where the work of Pierre Bourdieu becomes useful. Bourdieu spent his career trying to explain how #social_inequality survives without anyone having to enforce it openly. His answer was that inequality is carried inside people, in their bodies, their tastes, their speech, their confidence and their sense of what is possible. He called this internalised structure #habitus. He argued that habitus operates inside structured social spaces, or #fields, and that within any field certain resources are treated as valuable while others are treated as worthless. Those resources he called #capital, and he insisted that capital is not only economic. It is also cultural, social and symbolic. Applied to social work, the implication is uncomfortable. If the social work office is a field, then it has its own rules, its own currency, and its own hierarchy. Service users enter that field carrying resources formed elsewhere, in neighbourhoods and families and schools that value different things. A parent who can speak in the register of a professional meeting, who can produce documents, who understands that the correct response to an allegation is measured concern rather than anger, is already advantaged. A parent who cannot do these things is not simply less skilled. They are holding the wrong currency. Recent scholarship has taken this idea seriously. A critical review of thirty four published works on Bourdieu and social work concluded that his theory can strengthen critical and radical practice, particularly through the concepts of habitus, capital, field and reflexivity, but that the symbolic dimension of his thought has been badly underdeveloped in the profession even though it is central to understanding #oppression (Wolniak and Houston, 2023). Others have argued that Bourdieu remains relevant but largely absent from social work teaching and research, a strange gap given how directly he speaks to stratification, marginalisation and the role of the welfare state (Agotnes, Barua and Spjeldnaes, 2024). This article tries to close part of that gap. It is written for students and early career practitioners who want more than a slogan about power. It offers a framework that can be carried into a home visit. The argument runs in three movements. First, that social work interventions are structured by #cultural_capital in ways that are usually invisible to the people making decisions. Second, that the professional habitus of the social worker is itself a product of class, training and organisational conditioning, and therefore cannot be treated as a neutral instrument of judgement. Third, that once these two things are recognised, a set of concrete practice moves becomes available which can reduce, though never abolish, the #symbolic_violence built into the helping relationship. 2. Aims and Research Questions The article has four aims. To set out Bourdieu's conceptual toolkit clearly enough that a reader with no sociology background can use it. To map how cultural capital operates at each stage of a typical intervention: referral, assessment, planning, review and closure. To examine the habitus of the practitioner as an object of analysis rather than a source of neutral judgement. To translate the analysis into practice principles that can survive contact with a real caseload. Three questions guide the discussion. How do the three states of cultural capital shape who is believed, who is helped, and who is punished within social work interventions? In what ways does the professional habitus of social workers reproduce the very inequalities the profession claims to oppose? What practical strategies can practitioners use to redistribute #symbolic_power without abandoning statutory duties? 3. Methodology This is a conceptual paper supported by a #critical_narrative_synthesis of the literature. It is not a systematic review and does not claim to be exhaustive. The approach had three stages. Stage one: sourcing. Peer reviewed articles and scholarly books were identified that apply Bourdieusian concepts to social work, social care, health care and welfare institutions. Priority was given to work published between 2021 and 2025 so that the analysis speaks to current conditions of practice, including austerity, workforce shortages and the growth of algorithmic decision support. Foundational texts by Bourdieu himself were retained because the concepts cannot be discussed responsibly without them. Stage two: thematic organisation. Sources were grouped around recurring themes: class and poverty; symbolic domination; the neoliberal state; reflexivity; interprofessional relations; and critiques of Bourdieu. These themes echo those identified in the existing review literature (Wolniak and Houston, 2023), which gives some confidence that the map is not idiosyncratic. Stage three: synthesis and application. The themes were then tested against practice. Three composite scenarios were constructed. They are not case studies and contain no real personal data. They are illustrative devices, built from patterns reported across the literature, and their function is pedagogical rather than evidential. The limitations of this method are stated openly in section 13. The main one is that a conceptual synthesis can show that a framework is coherent and plausible. It cannot show that it works. That requires #empirical_research, which section 14 calls for. 4. Theoretical Framework: Bourdieu's Thinking Tools Bourdieu disliked the idea of theory as an abstract system. He described his concepts as thinking tools, meant to be used on real material and adjusted when they failed. The following is a working guide. 4.1 Field A field is a structured social space with its own logic. Law is a field. Medicine is a field. Statutory #child_protection is a field. Each has its own rules about what counts as evidence, what counts as competence, and who is allowed to speak with authority. Two features matter for social work. First, a field is an arena of struggle. Actors within it compete for position and recognition. Second, each field has a #doxa, a set of assumptions so taken for granted that they are never argued about. In child protection, the doxa might include the belief that a parent who becomes angry in a meeting is demonstrating risk rather than distress. Nobody writes this down. Everybody acts on it. The social work field is also nested inside larger fields: the local authority, the welfare state, the national economy. This matters because pressure travels downwards. Analyses of the Care Act 2014 in England show how national economic policy, funding cuts and workforce shortages reshape what social workers are able to value and do at the level of a single assessment (Bark, 2025). 4.2 Habitus Habitus is the part of society that has been deposited inside the person. It is the set of lasting dispositions, tastes, reflexes and expectations formed by a life history. It shapes how a person walks into a room, how they respond to authority, what they find embarrassing, what they consider normal, and what they believe is realistically available to them. Three things are commonly misunderstood. First, habitus is not a personality. It is social structure worn as second nature. Second, habitus is not a prison. It is a set of tendencies, not a script. It generates improvisation within limits. Third, habitus is not only mental. It is physical. It lives in posture, accent, eye contact and tone. This is why Bourdieu called it #embodied_history. Recent work has emphasised how #inequality supplies the very context in which habitus is formed, so that dispositions towards trust, hope and engagement are themselves unequally distributed (Sharlamanov, Jovanoski and Kostovska, 2024). For practice, the key insight is that a parent's apparent hostility, disengagement or fatalism may not be a symptom of pathology. It may be a rational disposition formed by decades of encounters with institutions that did not help. 4.3 Capital Bourdieu extended the idea of capital beyond money. Four forms concern us. #Economic_capital is material: income, savings, housing, transport, the ability to take a day off work to attend a meeting. #Social_capital is the network: who you know, who will vouch for you, who will collect your child from school when you are in hospital. #Cultural_capital is the one this article foregrounds. Bourdieu described it in three states. The embodied state is what lives in the person: vocabulary, accent, confidence, ease with professionals, the ability to argue without appearing aggressive, the ability to appear calm while being accused. The objectified state is what a person possesses in material cultural form: books in the home, a computer, a printer, a tidy folder of letters from the school, photographs of a child's birthday. The institutionalised state is what has been officially certified: qualifications, diplomas, professional registrations, a driving licence, a parenting course completion certificate. #Symbolic_capital is what the other forms turn into once they are recognised as legitimate by the field. It is prestige, credibility, honour. A social worker's professional registration is institutionalised cultural capital that converts, inside the meeting room, into symbolic capital: the right to be believed. Capital is convertible. Money buys tutoring, which produces qualifications, which produce prestige. But conversion runs in one direction more easily than the other, and this asymmetry is one of the engines of #social_reproduction. 4.4 Symbolic Power, Symbolic Violence and Misrecognition This is the part of Bourdieu's thought that social work has used least and needs most. Symbolic power is the power to make categories stick. It is the power to name a situation and have the name accepted. A social worker who writes "mother presented as evasive" is exercising symbolic power. So is one who writes "mother was cautious given her previous experience of services". Both are interpretations. Only one is likely to be challenged. Symbolic violence is what happens when the dominated accept the terms of their own domination as natural and deserved. It is not shouting. It is not coercion. It is the parent who leaves a meeting agreeing that she is a bad mother, using the professional's vocabulary to describe herself. It is the young person who accepts that he is "not the type" who goes to university. #Misrecognition is the mechanism that allows this to work. The arbitrary is experienced as natural. The middle class habitus of the assessment form is experienced not as one culture among many, but as competence itself. Research on youth justice has shown how systemic features of participation processes can produce exactly this effect, with children formally invited to co-produce their own interventions while the terms, language and setting of participation remain entirely defined by professionals (Creaney and Burns, 2024). Studies of parents in child protection have described similar dynamics: cooperation is defined in advance, and any deviation is read as risk. 4.5 Hysteresis A less familiar concept, but a valuable one. #Hysteresis is what happens when a field changes faster than the habitus of the actors within it. The dispositions that once brought success now bring failure. Actors keep playing an old game inside a new one. Bourdieusian analysis of English adult social care has used hysteresis to describe a profession whose collectivist and welfarist habitus is increasingly out of step with an individualised, market oriented, resource constrained field, producing confusion about the value and function of social work itself (Bark, 2025). Practitioners feel this as a sense that they are doing the job wrong when in fact the job has changed underneath them. 4.6 Reflexivity Finally, Bourdieu insisted that the analyst must turn the tools on themselves. He called this participant objectivation. It is not the same as introspection or "reflection on feelings". It is the systematic attempt to identify the social position from which one is looking, and the distortions that position introduces. For social work, this means asking not only "how do I feel about this family?" but "what is it about my training, my class, my caseload and my organisation that makes this family look the way it does to me?" Houston and Swords (2022) have argued that this reflexive move must be joined to an account of the emotional and affective dimensions of habitus, since feelings themselves are socially produced and quietly reframe how risk and need are constructed. 5. The Social Work Encounter as a Field of Struggle Consider a standard child and family #assessment. The field has a physical shape: an office, or a living room temporarily converted into an office. It has a time limit. It has a form. It has a set of legitimate speech acts. Questions may be asked by one party and must be answered by the other. Notes may be taken by one party. The record produced belongs to the institution. Within this field, the worker's capital is dense and recognised. Their qualification is institutionalised cultural capital. Their employment is economic capital. Their access to colleagues, legal advice and multi agency panels is social capital. Their right to be believed is symbolic capital. The parent's capital, meanwhile, may be substantial in other fields and worth very little here. Deep knowledge of how to survive on a low income, a wide network of neighbours who provide childcare, extraordinary skill in de-escalating a child's distress, the ability to negotiate with a landlord: none of these appear on the form. This is the central insight. #Poverty is not only a lack of money. It is also a systematic mismatch between the capital a person holds and the capital a field rewards. The poverty aware paradigm developed in critical social work makes a related argument, insisting that poverty is a violation of rights that people actively resist, and that professional knowledge must be built through close relationships with the people affected rather than about them (Krumer-Nevo, 2020). Evaluations of poverty aware programmes implemented across social service departments in Israel found meaningful shifts in both service user experience and practitioner orientation when this stance was adopted at scale (Timor-Shlevin, Saar-Heiman and Krumer-Nevo, 2023), and organisational case study work has shown that such change depends on structures, not only on individual goodwill (Saar-Heiman, Nahari and Krumer-Nevo, 2023). The same asymmetry appears in health care. The concept of cultural health capital was developed to explain why patients with equivalent conditions receive unequal treatment. Certain communication styles, forms of knowledge and ways of presenting oneself are rewarded by clinicians, and these are unevenly distributed (Shim, 2010). Qualitative research with socially disadvantaged cardiac patients has since shown how the same mechanism plays out in practice, with patients who cannot mobilise the expected cultural resources experiencing worse communication and less responsive care (Rasmussen, Guise and Overgaard, 2023). Social work operates on the same terrain. 6. Cultural Capital in the Assessment Process #Assessment is where cultural capital does its heaviest work. It is worth walking through the three states in turn. 6.1 Embodied Cultural Capital: The Body That Answers Questions The embodied state is the hardest to see because it is read as personality. A parent who maintains steady eye contact, speaks in complete sentences, uses the word "concerns" rather than "problems", pauses before answering, and expresses regret without either defensiveness or collapse, is performing a specific cultural competence. That performance is not a moral achievement. It is a resource, acquired through a life in which such performances were modelled and rewarded. A parent who interrupts, who becomes tearful and then angry, who says "you people always do this", who does not know that the correct move is to accept partial responsibility, is not necessarily less loving or less safe. They are holding the wrong currency. The problem is that #assessment_frameworks encourage practitioners to read the second parent's behaviour as evidence about the child's safety. Under caseload pressure, embodied cultural capital becomes a proxy for capacity to change. Nobody decides to do this. It happens because the field's doxa makes it feel like ordinary professional judgement. Three specific mechanisms deserve naming. #Linguistic_capital. Bourdieu treated language as the most direct carrier of symbolic power. The vocabulary of social work is a technical register: threshold, escalation, protective factor, contingency plan. Parents who cannot speak it are structurally disadvantaged in meetings that are conducted in it, and they know they are. Emotional register. Institutions reward a particular emotional style: contained, cooperative, mildly anxious. Distress that exceeds that band is coded as instability. Distress that falls below it is coded as lack of insight. The band itself is a cultural artefact. Bodily comportment. Punctuality, dress, the state of the home, whether one stands or sits, whether one offers the worker a drink. These are read constantly and recorded rarely. 6.2 Objectified Cultural Capital: The Home as Evidence Home visits convert domestic objects into evidence. Books signal aspiration. A computer signals engagement with education. A clean kitchen signals order. A folder of school letters signals organisation. None of these are, in themselves, measures of a child's safety or a parent's love. They are, however, cheap to observe and easy to write down. The trap is that objectified cultural capital tracks #economic_capital almost perfectly. Books cost money. A working printer costs money. Space to keep a house tidy costs money. When a practitioner records the absence of these things, they are frequently recording poverty and calling it #neglect. This is not an argument that material conditions are irrelevant to children. It is an argument that practitioners must be able to distinguish between three different statements: this family lacks resources; this family lacks capacity; this family lacks care. These are routinely collapsed, and the collapse falls hardest on the poorest. 6.3 Institutionalised Cultural Capital: The Certificate as Currency Institutionalised cultural capital is the state's own recognition. A parenting programme certificate, a completed anger management course, a recovery programme discharge letter. Two consequences follow. First, #compliance becomes convertible. Attending the course generates capital even when the course changed nothing. Refusing the course destroys capital even when the parent is already doing the thing the course teaches. The certificate is legible to the field; the underlying change is not. Second, the state distributes this capital unevenly. Courses have waiting lists, are held during working hours, require transport and childcare. The parent who cannot attend is not necessarily unwilling. They are unable to convert their economic and temporal constraints into the currency the field demands. The record will nonetheless read: "did not engage". The phrase #did_not_engage is arguably the single most consequential piece of symbolic violence in contemporary social work. It transfers a structural failure into a personal deficit, and it does so in a form that will be read aloud in court. 7. The Habitus of the Practitioner If service users bring habitus into the room, so do social workers. Three layers are worth separating. 7.1 Class Habitus Social work draws heavily on people from working class and minoritised backgrounds, and it also professionalises them. Qualification, registration and employment do not erase an original habitus, but they overlay it with a professional one. The result is often a divided or #cleft_habitus, in which the practitioner recognises the service user's world intimately and is simultaneously required to categorise it. This can be a resource. It can also be a source of pain, and of overcompensation. A practitioner who has escaped poverty may be less patient with those who have not, precisely because their own escape is fragile and hard won. 7.2 Professional Habitus #Professional_habitus is installed through training, which teaches a way of seeing. Students learn to notice risk, to document defensively, to think in categories that map onto legal thresholds. This is necessary. It is also a form of cultural conditioning that narrows perception. The literature is clear that this professional habitus is now formed under conditions of severe pressure. A systematic synthesis of thirty nine qualitative studies found that #bureaucracy and #managerialism have restructured practice around paperwork, outputs and compliance, systematically devaluing relationships and reducing the time available for direct work (Pascoe, Waterhouse-Bradley and McGinn, 2023). Mixed methods research in Northern Ireland reinforced this, showing that although policy documents celebrate relationship based practice, workers spend increasing time on administrative tasks at the expense of contact with the people they serve, and that the commitment to relationships is often little more than rhetoric (Pascoe, 2025). The effect on habitus is direct. A worker whose day is structured by deadlines and thresholds will develop dispositions suited to deadlines and thresholds. Speed becomes a virtue. Ambiguity becomes a threat. The parent who requires forty minutes of patient listening becomes, in the quiet arithmetic of the caseload, a problem. 7.3 Organisational Habitus and the Two Hands of the State Bourdieu described the modern state as having two hands, a #right_hand and a #left_hand_of_the_state. The right hand is the treasury, the ministries of finance, the enforcers of budgetary discipline. The left hand is composed of teachers, nurses, social workers, the people who deal with the consequences of what the right hand does. Social workers, he suggested, are caught in an almost impossible squeeze, fighting simultaneously against the demoralisation of those they try to help and against the slowness and complexity of the bureaucracy that employs them (Agotnes, Barua and Spjeldnaes, 2024). This squeeze is not a personal failing. It is the structural position of the profession. Analyses of the professional identity of social workers in mental health services find repeated evidence of misalignment between the values workers hold and the expectations placed on them by their organisations (Bark, Dixon and Laing, 2023). Scholarship on the changing shape of welfare provision has similarly warned about the privatising of public issues, whereby collective problems are redefined as individual concerns and social work is recruited into managing the fallout (Roets, Kessl and Lorenz, 2023; Moth, 2023). Recognising this changes the meaning of practitioner #burnout. It is not simply exhaustion. It is hysteresis: the pain of a habitus formed for one game being made to play another. 8. Symbolic Violence in Everyday Practice: Four Mechanisms It helps to be specific. Four recurring mechanisms translate structural inequality into interpersonal harm. 8.1 The Imposition of Legitimate Language Meetings are conducted in professional register. Parents are expected to respond in it. Those who cannot are described as unclear, evasive or lacking insight. The imposition is invisible because the register is experienced by professionals as simply clear. Research on #epistemic_injustice in social work has made this argument with force. When the knowledge of people with lived experience is systematically discounted in favour of professional knowledge, an injustice is done not only to their interests but to their standing as knowers (Ali et al., 2024). Related work has traced how hermeneutical injustice arises when neither the professional nor the service user has the shared conceptual resources to make the service user's experience intelligible, and has argued that a stance of #cultural_humility offers a practical response (Anka, 2024). At a broader level, calls for #epistemic_justice in social work research point to the way dominant Western modes of inquiry have crowded out other ways of knowing (Maleku, 2025). 8.2 The Naturalisation of the Threshold #Thresholds are administrative constructions. They vary between authorities, change with budgets, and are reinterpreted under pressure. Yet in practice they are spoken about as though they described a fact in the world. The child "meets threshold" or does not, as though the threshold were a temperature. Naturalising the threshold performs two functions. It relieves the practitioner of the burden of a discretionary decision, and it presents a resource driven judgement as an objective finding. The family experiences the outcome as a statement about their worth. 8.3 The Commodification of Strength #Strengths_based_practice is, in principle, an emancipatory idea. In a resource constrained field it can become something else. Evidence given to policy inquiries in England has described social workers feeling obliged to emphasise deficits in order to get people over eligibility criteria, while resource allocation systems simultaneously discount support packages for people who have social connections, effectively converting a person's relationships into a reason to give them less (Bark, 2025). Similar tensions have been documented in the space between #personalisation policy and practice, where workers are left to absorb the contradiction (Southall, Lonbay and Brandon, 2021). The symbolic violence here is subtle. The person is told they are being empowered. They experience a reduction in support. And because the language of empowerment is used, they may struggle even to name what has happened to them. 8.4 The Recording of Deficit #Case_records are not passive. They are instruments of symbolic power that travel through the system, are read by strangers, and are cited in court. Every noun chosen in a record is a small act of categorisation. The asymmetry is stark. The parent's speech is summarised, interpreted and paraphrased. The professional's speech is recorded as fact. The parent rarely reads the record until it is too late to change it, and often lacks the linguistic capital to contest its framing when they do. 9. The Organisational Field: Managerialism, Metrics and Discretion None of the above can be fixed by good intentions alone, because none of it originates in the individual worker. The field of statutory social work in many high income countries has been reshaped by three decades of #new_public_management: risk management, standardisation, fragmentation, audit and accountability. The consistent finding across the literature is that these reforms increase paperwork, narrow #discretion and squeeze out #relationship_based_practice (Pascoe, Waterhouse-Bradley and McGinn, 2023; Pascoe, 2025). Social workers remain #street_level_bureaucrats who exercise real discretion, but they do so within tightening constraints and with diminishing time (Lipsky, 2010). Bourdieu's contribution here is to explain why practitioners so often comply with systems they dislike. It is not cowardice. It is #illusio: the investment in the game that makes the game feel worth playing. To be a good social worker within the field as it currently exists is to become skilled at the field's rules. Those rules include the timely completion of forms. Over time, the skills required to survive the field become the dispositions of the habitus, and the worker who once entered the profession to sit with people finds themselves genuinely anxious about a missed deadline and only vaguely uneasy about a missed conversation. The literature on disruptive social work responds to this by asking what forms of dissent remain available to practitioners who wish to challenge the field rather than adapt to it (Feldman, 2022). This is the right question, and it is a field level question, not a personal one. 10. A Bourdieusian Practice Framework: Five Moves The rest of this article is constructive. If the analysis above is right, what follows for practice? Five moves are proposed. They are ordered from the individual to the structural, and they are cumulative rather than alternative. Move One: #Reflexive_Objectivation Not reflection. Objectivation. The practitioner asks a specific set of questions before and after significant encounters. What is my own class trajectory, and how does it shape what I find normal in a home? What behaviours make me feel respected, and am I confusing respect with safety? What did my training teach me to notice, and therefore what does it teach me to miss? What organisational pressures are acting on me today, and how might they be shaping what I am about to write? Whose account of this family will be recorded as fact, and whose as claim? These questions are uncomfortable by design. Houston and Swords (2022) argue that this reflexive work must extend to the affective layer of habitus, because socially induced feelings quietly reshape how risk and need are framed. Supervision is the obvious place for it, but only if supervision is protected from becoming a case management audit. Move Two: #Capital_Mapping Instead of asking what a family lacks, the worker maps what they hold, in all four currencies, and then asks which of those holdings the field currently refuses to recognise. A simple structure: Economic: income, debt, housing security, transport, food. Social: who helps, who could help, who has been driven away, who is missing. Cultural (embodied): languages spoken, practical skills, survival knowledge, ways of managing a child's distress that the parent has developed and never had named. Cultural (objectified): what the home contains and what it means to the family rather than to the worker. Cultural (institutionalised): qualifications, certificates, and crucially, what has been attempted and blocked. Symbolic: where is this person respected? In their building, their faith community, their family? What credibility do they hold that the service does not see? Work on families of disabled children has shown very clearly how differential access to economic, social and cultural resources determines who can successfully navigate welfare bureaucracies, even in generous welfare states, so that structural inequality is reproduced inside the very systems designed to reduce it. Capital mapping makes this visible at the level of a single case. The purpose is not to produce a nicer document. It is to change the direction of the assessment gaze. The question shifts from "what is wrong with this family" to "what is this family holding, and why does our system not count it". Move Three: #Linguistic_Justice Symbolic power travels through language, so language is where it can first be redistributed. Practical steps: Say plainly, at the start, what the meeting is for, who will read the record, and what the possible outcomes are. Uncertainty is itself a form of domination. Translate technical terms out loud, every time, without waiting to be asked. People rarely ask, because asking exposes them. Offer the person the chance to correct the record before it is finalised, and record their disagreement in their own words when they disagree. Avoid the vocabulary of deficit where a descriptive alternative exists. "Did not attend three appointments" is a fact. "Did not engage" is a verdict. Ask people what words they use for their own situation, and then use those words. This is the practical face of epistemic justice. It treats the service user as a knower, not only as an object of knowledge (Ali et al., 2024; Anka, 2024). Move Four: #Recognition Joined to #Redistribution Recognition without material change is a trap. Telling a parent that her knowledge is valued while doing nothing about her rent arrears is a more sophisticated form of the same injury. The poverty aware paradigm makes this the centre of its argument, insisting that relational affirmation must be joined to the active realisation of rights and the provision of material support (Krumer-Nevo, 2020; Timor-Shlevin, Saar-Heiman and Krumer-Nevo, 2023). In Bourdieusian terms, this is an argument about capital conversion. The worker uses their own institutional capital to convert into economic and social capital for the service user: making the benefits claim, writing the housing letter, attending the school meeting as an ally rather than an assessor, spending the professional credibility they hold rather than hoarding it. This is not charity. It is the deliberate transfer of symbolic capital from the person who has it to the person who does not. Move Five: #Field_Level_Advocacy Individual practice cannot repair a field. If eligibility thresholds are producing symbolic violence, the threshold must be contested. If a resource allocation algorithm penalises people for having friends, the algorithm must be challenged. This is where Bourdieu's own political commitments matter. He argued for the role of the critical intellectual, and he wrote directly about the suffering produced by the state's abandonment of its left hand. Contemporary work has taken up this call, arguing for a social work that disrupts hegemonic narratives and refuses the privatisation of public problems (Roets, Kessl and Lorenz, 2023; Feldman, 2022). In practice this means: aggregating cases into evidence, using professional bodies and unions, publishing what is happening, insisting that the record of a systemic failure is written down as a systemic failure. It also means practical #solidarity with #service_user_movements, since a group that recognises its shared position is in a far stronger structural position than a set of isolated individuals. 11. Application: Three Composite Scenarios The following are constructed illustrations, not real cases. 11.1 Child Protection: The Mother Who "Did Not Engage" A single mother of two is referred after repeated school absences. She misses two assessment appointments. At the third she arrives late, speaks quickly, contradicts herself about dates, and becomes angry when asked about the children's father. The draft assessment records: presented as hostile, minimised concerns, did not engage. A Bourdieusian reading changes the picture. Field: the meeting takes place at 10 a.m. in an office forty minutes away by two buses. Economic capital: she works a zero hours shift pattern; attending costs a shift. Embodied cultural capital: she has no practice in the professional register and no template for how one is meant to sound when accused. Habitus: her own childhood involved care proceedings. Her anger is a rational disposition formed by experience, not a symptom. Symbolic violence: the record converts a structural barrier into a moral judgement, and she will not have the linguistic resources to contest it in court. Practice response. Capital mapping identifies that she has organised informal childcare across three neighbours, that she has kept both children fed through a benefits sanction, and that she has taught her older child to manage the younger one's asthma. None of this appears on the form. Linguistic justice means her disagreement is recorded verbatim. Redistribution means the worker spends an afternoon resolving the sanction. Field level advocacy means the team records, formally, that appointment scheduling is generating false negatives about engagement in low income households. 11.2 Adult Social Care: The Man With Too Many Friends An older man with mobility problems is assessed under strengths based principles. He has a large church network. The resource allocation system reduces his indicative personal budget on the basis of available informal support. He is told he is being empowered. He receives less. This is the commodification of strength, and it has been documented in policy evidence and in Bourdieusian analysis of the Care Act (Bark, 2025). Practice response. The worker names the conversion explicitly in the record: the person's social capital is being used to offset the authority's economic constraint. They challenge the indicative figure. They document the challenge. They ensure the man understands what is being proposed and in what language he may object. If the pattern repeats, it is escalated as a systemic issue rather than absorbed case by case. 11.3 Youth Justice: Participation Without Power A sixteen year old is invited to a multi agency panel to "have his say". The meeting uses risk terminology he has never been taught. He is asked to speak last, after seven professionals. He says "yeah, whatever", and this is recorded as a lack of motivation. This is precisely the structure described in research on participatory practices in youth justice, where children are formally included in processes whose language, timing and terms remain wholly professional, so that the invitation to participate can itself become a vehicle of symbolic violence (Creaney and Burns, 2024). Practice response. The panel is restructured: he speaks first, in his own vocabulary, with an advocate he chose, having been given the papers in advance in a form he can read. The professional language is translated aloud. His stated priorities are written into the plan in his own words, and where they are overruled, the reason is recorded and explained to him directly. 12. Implications for Social Work Education If habitus is formed, it can be formed differently, and #social_work_education is where that reforming begins. Teach the concepts early and concretely. Habitus, field, capital and symbolic violence should not appear once in a sociology module and never again. They should be used as tools in practice teaching, applied to real assessments and real case notes. Make the student's own habitus an object of study. Students should be asked, systematically, to locate their own class trajectory, their linguistic capital and their assumptions about what a good home looks like. This is uncomfortable and must be handled with care, but the alternative is a workforce that mistakes its own culture for competence. Teach record writing as an exercise of power. Every social work student should analyse real case records for how symbolic power operates in them: which speaker is quoted, which is summarised, which verbs are used, where a structural fact has been converted into a personal deficit. Confront the epistemic hierarchy. Cultural humility, as an educational stance, involves accepting that professional knowledge is one knowledge among several, and that unlearning is a legitimate part of learning (Anka, 2024). People with lived experience should teach, not merely appear as case material (Ali et al., 2024). Prepare students for hysteresis. Students should be told plainly what they are entering: a field shaped by austerity, high turnover and bureaucratic load, in which their values will collide with their conditions. Naming this in advance converts a private sense of failure into a shared structural analysis, which is the first condition of collective action. 13. Critiques and Limitations of the Framework A serious framework must survive its own criticisms. The charge of determinism. The most common objection is #determinism: that habitus explains everything and therefore permits nothing. If dispositions are structurally produced, where is agency? Bourdieu's own answer is that habitus generates rather than dictates, that it is a feel for the game rather than a rulebook, and that crises and field changes open space for improvisation. But the objection has force, and social workers should be wary of using habitus as a new language for writing people off. If "her habitus prevents change" simply replaces "she lacks capacity", nothing has been gained and something has been lost. The problem of measurement. Habitus is notoriously difficult to operationalise. It is easy to invoke and hard to demonstrate, and the risk of circular reasoning is real: behaviour is explained by habitus, and habitus is inferred from behaviour. The neglect of race, gender and disability, and the need for #intersectionality. Bourdieu's primary axis is class. Applying his framework without intersectional supplementation risks reproducing exactly the erasures that social work is trying to overcome. Cultural capital operates racially: which accents, which names, which religious practices are recognised as legitimate is never a purely class question. Practitioners using this framework must import an intersectional analysis rather than assuming that class does all the work. The relative neglect of the symbolic in social work's own uptake of Bourdieu. The existing review literature notes that social work has borrowed habitus and capital fairly freely while largely ignoring symbolic power, symbolic violence and misrecognition, which is where the theory's critical force actually lies (Wolniak and Houston, 2023). This article has tried to correct that, but it means much of the applied literature it draws on is thinner than one would want. The accessibility problem. Bourdieu's writing is difficult, his output is enormous, and his philosophical assumptions do not sit comfortably with social work's dominant traditions. This has been identified as one reason for his continued marginality in the profession (Agotnes, Barua and Spjeldnaes, 2024). Any serious attempt to embed the framework must therefore be an act of translation, which is part of what this article has attempted. The limits of the present study. This is a conceptual synthesis. It draws on a purposive rather than exhaustive body of literature, it privileges Anglophone and European sources, and it does not test its own framework against data. The composite scenarios illustrate; they do not evidence. 14. Directions for Future Research Four lines of #future_research follow. Documentary analysis of case records. A systematic linguistic analysis of assessments and court reports, coded for the conversion of structural facts into personal deficits, would provide hard evidence for a claim that is currently mostly theoretical. Comparative work on thresholds. If thresholds are field artefacts rather than objective measures, then similar cases should be treated differently across authorities with different budgets. This is testable. Evaluation of the five move framework. A practice development study in a single team, with pre and post measures of service user experience, would establish whether capital mapping and linguistic justice change anything that people actually feel. Intersectional extension. Empirical work is needed on how cultural capital is racialised and gendered in welfare encounters, building on the existing health literature on cultural health capital (Shim, 2010; Rasmussen, Guise and Overgaard, 2023). 15. Conclusion Social work cannot be practised without power. The profession holds legal authority, controls resources, and writes the official version of other people's lives. Pretending otherwise is not humility; it is a way of avoiding responsibility for what power does. Bourdieu's contribution is to make the workings of that power describable. He shows that the encounter between a worker and a service user is not a meeting of two individuals but a meeting of two histories inside a structured field with its own currency. He explains why the parent who cannot perform middle class composure is penalised, why the certificate matters more than the change it was supposed to certify, and why the phrase "did not engage" does so much damage. He explains, too, why practitioners who entered the profession to sit with people find themselves anxious about deadlines instead, and why that anxiety is not a personal weakness but the mark of a habitus caught in a changing field. The framework offered here does not promise to abolish this. Nothing can. Symbolic violence is not a bug in the system that a better form will fix. It is the ordinary way that unequal fields reproduce themselves, and social work is an unequal field. What the framework offers is more modest and more useful. It offers a way of seeing clearly enough to act. Reflexive objectivation lets the worker locate themselves. Capital mapping lets them see what the family holds rather than only what it lacks. Linguistic justice returns some of the power of naming. Recognition joined to redistribution turns respect into rent money. Field level advocacy insists that what is structural be recorded as structural. The #habitus_of_care is not a gentler disposition. It is a more honest one: a professional way of being that knows what it is doing, knows what it costs, and refuses to call an act of power an act of kindness. That refusal is where a more just #social_work_practice begins. Hashtags #Bourdieu_in_social_work #The_Habitus_of_Care #Cultural_Capital_And_Care #Power_Dynamics_In_Social_Work #Symbolic_Violence_In_Practice #Reflexive_Practice #Critical_Social_Work #Poverty_Aware_Practice #Capital_Mapping #Linguistic_Justice #Field_Habitus_Capital #Anti_Oppressive_Practice #Social_Work_Education #Welfare_State_Sociology #Epistemic_Justice_In_Social_Work References Agotnes, G., Barua, P. and Spjeldnaes, I. O. (2024) Social work and Pierre Bourdieu: relevance for and in a Norwegian welfare state context. Nordic Social Work Research. https://doi.org/10.1080/2156857X.2024.2368177 Ali, N., Felstead, K., Mohamed, O., McIvor, A., Wilkes, J. and Watters, D. (2024) Why is it important to ensure the voice and influence of people with lived experience in social work research and practice? The British Journal of Social Work, 54(4), pp. 1391-1401. https://doi.org/10.1093/bjsw/bcae101 Anka, A. (2024) Using the concept of epistemic injustice and cultural humility for understanding why and how social work curricular might be decolonized. Social Work Education, 43(9), pp. 2880-2896. https://doi.org/10.1080/02615479.2023.2299245 Bark, H. (2025) Social work, personalisation, and the Care Act 2014: a Bourdieusian analysis. The British Journal of Social Work, 55(3), pp. 1161-1177. https://doi.org/10.1093/bjsw/bcae180 Bark, H., Dixon, J. and Laing, J. (2023) The professional identity of social workers in mental health services: a scoping review. International Journal of Environmental Research and Public Health, 20(11), 5947. https://doi.org/10.3390/ijerph20115947 Bourdieu, P. (1977) Outline of a Theory of Practice. Cambridge: Cambridge University Press. Bourdieu, P. (1984) Distinction: A Social Critique of the Judgement of Taste. Cambridge, MA: Harvard University Press. Bourdieu, P. (1986) The forms of capital, in Richardson, J. G. (ed.) Handbook of Theory and Research for the Sociology of Education. New York: Greenwood Press, pp. 241-258. Bourdieu, P. (1990) The Logic of Practice. Stanford, CA: Stanford University Press. Bourdieu, P. (1991) Language and Symbolic Power. Cambridge: Polity Press. Bourdieu, P. (2000) Pascalian Meditations. Cambridge: Polity Press. Bourdieu, P. and Passeron, J. C. (1977) Reproduction in Education, Society and Culture. London: Sage. Bourdieu, P. and Wacquant, L. (1992) An Invitation to Reflexive Sociology. Chicago: University of Chicago Press. Creaney, S. and Burns, S. (2024) Freedom from symbolic violence? Facilitators and barriers to participatory practices in youth justice. Youth Justice. https://doi.org/10.1177/14732254231156844 Emirbayer, M. and Williams, E. M. (2005) Bourdieu and social work. Social Service Review, 79(4), pp. 689-724. Feldman, G. (2022) Disruptive social work: forms, possibilities and tensions. The British Journal of Social Work, 52(2), pp. 759-775. https://doi.org/10.1093/bjsw/bcab045 Fricker, M. (2007) Epistemic Injustice: Power and the Ethics of Knowing. Oxford: Oxford University Press. Garrett, P. M. (2007) Making social work more Bourdieusian: why the social professions should critically engage with the work of Pierre Bourdieu. European Journal of Social Work, 10(2), pp. 225-243. Houston, S. and Swords, C. (2022) Responding to the weight of the world: unveiling the feeling Bourdieu in social work. The British Journal of Social Work, 52(4), pp. 1934-1951. https://doi.org/10.1093/bjsw/bcab161 Krumer-Nevo, M. (2016) Poverty-aware social work: a paradigm for social work practice with people in poverty. The British Journal of Social Work, 46(6), pp. 1793-1808. https://doi.org/10.1093/bjsw/bcv118 Krumer-Nevo, M. (2020) Radical Hope: Poverty-Aware Practice for Social Work. Bristol: Policy Press. Lipsky, M. (2010) Street-Level Bureaucracy: Dilemmas of the Individual in Public Services, 30th anniversary edition. New York: Russell Sage Foundation. Maleku, A. (2025) Toward epistemic justice in social work research: (re)imagining global knowledge production. The British Journal of Social Work, 55(2), pp. 613-620. https://doi.org/10.1093/bjsw/bcaf052 Moth, R. (2023) Understanding Mental Distress: Knowledge, Practice and Neoliberal Reform in Community Mental Health Services. Bristol: Policy Press. Pascoe, K. M. (2025) The impact of bureaucracy and managerialism on relationship-based practise: a mixed methods study of frontline social work in Northern Ireland. Social Policy and Administration, 59(3), pp. 438-453. https://doi.org/10.1111/spol.13068 Pascoe, K. M., Waterhouse-Bradley, B. and McGinn, T. (2023) Social workers' experiences of bureaucracy: a systematic synthesis of qualitative studies. The British Journal of Social Work, 53(1), pp. 513-533. https://doi.org/10.1093/bjsw/bcac106 Rasmussen, A. N., Guise, A. and Overgaard, C. (2023) Understanding social inequalities in cardiac treatment through the lens of cultural health capital: a study of Danish socially disadvantaged ischemic heart patients' lived experiences of healthcare interactions. Social Theory and Health, 21(1), pp. 17-32. https://doi.org/10.1057/s41285-021-00173-1 Roets, G., Kessl, F. and Lorenz, W. (2023) New charity economy and social work: reclaiming the social dimension of public life in the context of changing welfare rationales. Social Work and Society, 21(1), pp. 1-21. Saar-Heiman, Y., Nahari, M. and Krumer-Nevo, M. (2023) Critical social work in public social services: poverty-aware organizational practices. Journal of Social Work. https://doi.org/10.1177/14680173221144204 Sharlamanov, K., Jovanoski, A. and Kostovska, M. (2024) Social inequalities as a context for the formation of habitus. Discover Global Society, 2(1). https://doi.org/10.1007/s44282-024-00125-w Shim, J. K. (2010) Cultural health capital: a theoretical approach to understanding health care interactions and the dynamics of unequal treatment. Journal of Health and Social Behavior, 51(1), pp. 1-15. Simmonds, B. (2023) Ageing and the Crisis in Health and Social Care: Global and National Perspectives. Bristol: Policy Press. Southall, C., Lonbay, S. P. and Brandon, T. (2021) Social workers' negotiation of the liminal space between personalisation policy and practice. European Journal of Social Work, 24(2), pp. 238-250. https://doi.org/10.1080/13691457.2019.1633624 Timor-Shlevin, S., Saar-Heiman, Y. and Krumer-Nevo, M. (2023) Poverty-aware programs in social service departments in Israel: a rapid evidence review of outcomes for service users and social work practice. International Journal of Environmental Research and Public Health, 20(1), 889. https://doi.org/10.3390/ijerph20010889 Tucker, L., Webber, M. and Jobling, H. (2022) Mapping the matrix: understanding the structure and position of social work in mental health services in England and Wales. The British Journal of Social Work, 52(6), pp. 3210-3229. https://doi.org/10.1093/bjsw/bcab235 Wiegmann, W. L. (2017) Habitus, symbolic violence, and reflexivity: applying Bourdieu's theories to social work. Journal of Sociology and Social Welfare, 44(4), pp. 95-116. Wolniak, M. and Houston, S. (2023) A sociologist in the field of social work: Pierre Bourdieu's theory and its relevance for social work practice. Critical and Radical Social Work, 11(2), pp. 183-198. https://doi.org/10.1332/204986021X16455445960144

Software has changed its role. For most of the digital era, a program waited for a command, carried it out, and stopped. A newer class of systems does not wait. It plans, selects tools, calls other programs, revises its own steps, and reports back only when a goal has been reached or abandoned. Work organized around such systems is described here as agentic workflows, and the systems themselves as autonomous agents. This article asks a psychological question rather than a technical one. When a person works next to an actor that acts on its own, what changes inside the person? This review synthesizes empirical and theoretical work published mainly between 2021 and 2025 across human factors, information systems, organizational behavior, cognitive psychology, and human computer interaction. Seven domains of behavioral adaptation are identified: the reallocation of attention and the decay of vigilance; the drift of trust away from calibrated levels toward either over reliance or algorithm aversion; cognitive offloading and the slow reshaping of metacognition; the redistribution of effort and its uneven effect on productivity; the renegotiation of professional identity and self efficacy; the diffusion of accountability and the emergence of a responsibility gap; and altered social behavior toward machines that are treated, partly, as machine teammates. A five stage model of adaptation is proposed: contact, calibration, delegation, drift, and restructuring. The article argues that behavioral adaptation is not a side effect of agentic systems but their main effect, because the human being is the part of the system that is hardest to redesign and easiest to change without noticing. Practical guidance is offered for students, educators, and managers, along with a measurement agenda for future research. Keywords Agentic workflows; autonomous agents; human agent interaction; trust calibration; cognitive offloading; automation bias; delegation; professional identity; future of work; AI literacy. 1. Introduction A student in 2015 who used a computer to write an essay used it as a surface. The machine held the words, checked the spelling, and did nothing else. A student in 2025 may instead describe a goal, watch a system search sources, draft sections, revise itself after checking its own output, and return a finished piece. The first student was using a tool. The second student is supervising an actor. The distance between those two situations is short in years and very long in psychology. This article is about that distance. Its subject is behavioral adaptation: the set of durable changes in perception, effort, judgment, motivation, and self concept that appear in people who work regularly inside agentic workflows. The claim advanced here is simple to state and difficult to absorb. The most important thing that autonomous systems change is not the task. It is the person doing the task. The claim rests on an old observation from human factors research. Whenever a machine takes over part of a job, the human part of the job does not simply shrink. It changes shape. The operator stops doing and starts watching. Watching is a different cognitive activity from doing, with different failure modes, a different learning curve, and different emotional weight. Decades of work on aviation autopilots, industrial control rooms, and clinical decision support established that removing manual effort often increases monitoring load, degrades situation awareness, and produces errors of a kind that never existed before the machine arrived (Endsley, 2023; O'Neill, McNeese, Barron, & Schelble, 2022). The current generation of language based agents extends this pattern from cockpits and control rooms into essays, code, spreadsheets, legal memos, and lesson plans. It brings the psychology of the supervisor to people who never trained as supervisors. Three features of contemporary agentic systems make the psychological question urgent rather than merely interesting. The first is initiative. A conventional tool has no goals of its own and takes no steps that were not requested. An agentic system is given an objective and permitted to choose the intermediate steps. This shifts the human contribution upward, from execution to specification and review, a shift that Baird and Maruping (2021) describe as delegation to an agentic artifact rather than mere use of an artifact. delegation is a social act. It carries expectations, disappointment, blame, and relief. People do not delegate to hammers. The second is opacity. The steps taken by a modern system are frequently unavailable, partially available, or available in a form that is too long or too technical to check. Lebovitz, Lifshitz-Assaf, and Levina (2022) show that professionals confronted with opaque outputs do not respond in a uniform way. Some engage deeply and search for verification, others disengage and accept, and the choice depends less on the quality of the system than on how the professional understands their own expertise. Opacity therefore does not simply reduce trust. It sorts people into different psychological strategies. The third is fluency. Language based agents produce output that is smooth, confident, well organized, and grammatically clean, regardless of whether it is correct. Human beings have spent their entire evolutionary and cultural history using fluency as a proxy for competence. That proxy has now been decoupled from truth at scale. The result is a persistent mismatch between the confidence a system displays and the confidence it deserves, and human judgment is not naturally equipped to hold that gap open (Steyvers & Kumar, 2024). Taken together, initiative, opacity, and fluency create conditions in which ordinary human heuristics misfire in orderly and predictable ways. Understanding those misfires is the purpose of this article. 1.1 Aim and contribution The aim is integrative rather than experimental. The literature on human interaction with intelligent systems is large but scattered. Human factors researchers study vigilance and #automation_bias. Information systems researchers study #delegation and role change. Organizational scholars study control, resistance, and #professional_identity. Cognitive psychologists study #cognitive_offloading and memory. Human computer interaction researchers study #explainability and reliance behavior. These groups rarely read each other, and each has captured a piece of the same animal. This article contributes in four ways. It offers a working definition of #agentic_workflows that is psychological rather than technical. It organizes the evidence into seven domains of #behavioral_adaptation. It proposes a five stage process model that connects the domains over time. It sets out a measurement agenda, since one reason the field is fragmented is that it has no shared instruments. 1.2 Structure Section 2 defines the object of study. Section 3 assembles the theoretical foundations. Section 4 describes the review method. Section 5 presents the seven domains of #behavioral_adaptation. Section 6 introduces the stage model. Section 7 examines moderators. Section 8 discusses implications for #students, educators, and organizations. Section 9 proposes a research agenda, Section 10 states limitations, and Section 11 concludes. 2. Defining Agentic Workflows and Autonomous Non-Human Actors 2.1 From tool to actor A definition based on architecture will age badly. Model sizes, memory systems, and tool calling protocols change every few months. A definition based on the human experience of the system will last longer, because human cognition changes slowly. An #autonomous_agent, for the purposes of this article, is a non-human system that satisfies four conditions from the perspective of the person working with it. It pursues a goal rather than executing an instruction. The person supplies an end state, and the system determines the path. Even when the path is later revealed, it was not authored by the person. It takes actions in the world without a confirmation step for every action. It may search, write files, send messages, run code, or call other systems. Some actions are gated; many are not. It persists across time. It maintains state, revisits earlier steps, retries after failure, and continues while the person is not looking. It produces output whose internal justification is not fully inspectable at reasonable cost. The person can check the result but usually cannot cheaply check the reasoning. An #agentic_workflow is any organized sequence of work in which one or more such systems occupy positions that were previously occupied by people or by fully deterministic software, and in which the human role includes at least one of: specifying goals, approving intermediate actions, verifying outputs, or absorbing the consequences of failure. Notice that this definition says nothing about intelligence, consciousness, or reasoning. It describes a relationship. That is deliberate. The psychological effects described in this article do not depend on whether the system truly reasons. They depend on whether the person behaves as though it might. 2.2 Degrees of autonomy Autonomy is not binary. It is useful to distinguish five practical levels, because #behavioral_adaptation differs sharply across them. At level one, the system suggests and the person disposes. Each suggestion is inspected before acceptance. This is the classic decision support arrangement, and most of the experimental literature on #reliance was built here. At level two, the system drafts a complete artifact and the person edits. The unit of review is now the whole product, not the individual step. Verification cost rises and, crucially, the person's sense of ownership begins to weaken. At level three, the system executes multi step plans and reports back at the end. The person sees a summary rather than a trace. #oversight becomes sampling rather than checking. At level four, the system operates continuously against standing objectives and only escalates exceptions. The person is now an exception handler, a role known from process control to be prone to #complacency and slow response. At level five, systems coordinate with other systems, and the human is present mainly for accountability and periodic audit. This level is emerging rather than common, but its psychology is already visible in fully automated pipelines. The important pattern is that human cognitive involvement does not decline smoothly with autonomy. It declines, then spikes violently at the moment of failure. This is the well documented irony of automation, and it survives intact in the agentic era. The person is least practiced and least attentive precisely when they are most needed. 2.3 Why the workflow is the right unit of analysis Much research studies single interactions. A participant sees an AI recommendation and accepts or rejects it. This design is clean and has produced valuable findings, but it cannot capture #behavioral_adaptation, because adaptation is a function of repetition. Habits form over hundreds of interactions, not one. Skills decay over months. #professional_identity is renegotiated over years. The #agentic_workflow, defined as a repeated cycle of specification, execution, review, and consequence, is therefore the correct unit. It contains the feedback that shapes behavior. A person who is never told that the system was wrong will not learn to distrust it. A person who is punished for the system's error but not rewarded for catching it will learn to check less, not more, because checking is costly and its benefits are invisible when it succeeds. This point deserves emphasis. Most #behavioral_adaptation is rational given the incentives the person actually faces. It looks irrational only when we forget to look at the #feedback_loops. 3. Theoretical Foundations 3.1 Trust and calibrated reliance The dominant framework in this area treats trust as an attitude and reliance as a behavior, with performance depending on the fit between them. Good outcomes require #trust_calibration: the person should rely on the system when it is likely to be right and override it when it is likely to be wrong. Miscalibration takes two forms. #over_reliance means accepting outputs that should have been rejected. Under reliance, often called #algorithm_aversion, means rejecting outputs that should have been accepted. Meta analytic work confirms that trust in automated and intelligent systems is shaped by system performance, transparency, task characteristics, and individual differences, but the effect sizes are moderate and the variance across studies is large (Kaplan, Kessler, Brill, & Hancock, 2023; Hancock, Kessler, Kaplan, Brill, & Szalma, 2021). Trust is not a fixed trait applied to a system. It is a moving estimate that people update badly. The reason people update badly is instructive. Calibration requires knowing when the system fails, and the failures of a fluent agent are often silent. A wrong answer that reads well produces no error signal. The person's estimate of reliability is therefore built from the subset of failures they happened to catch, which systematically overstates reliability. This is a structural, not a personal, defect. A second complication is that trust operates at the wrong level of granularity. People form a global impression of a system as competent or incompetent, when in fact competence varies enormously by task. Dell'Acqua and colleagues (2023) describe this as a jagged frontier: the system is superhuman on one task and confidently useless on a neighboring task that looks identical to the human. A person carrying a single trust score cannot navigate a jagged frontier. They will over trust on the far side and under trust on the near side, and their average performance will look mediocre even though the tool is excellent. 3.2 Delegation and agency Baird and Maruping (2021) reframe interaction with agentic systems as a transfer of rights and responsibilities rather than an act of usage. Their theoretical framework treats #delegation as bidirectional: humans delegate to artifacts, and artifacts increasingly delegate back to humans, for instance by escalating an exception or requesting a decision. This reframing matters psychologically because #delegation triggers social cognition. Research on human delegation shows that people delegate less than is optimal when they feel responsible for outcomes, delegate more than is optimal when the delegate is perceived as high status, and experience discomfort when the delegate outperforms them. All three patterns appear in human agent settings. Fügener, Grahl, Gupta, and Ketter (2022) demonstrate that people are poor at knowing which tasks to hand over, because effective #delegation requires accurate knowledge of both one's own ability and the system's ability, and most people have neither. Their earlier work raises the further possibility that heavy collaboration with a capable system makes individuals more similar to one another in judgment, reducing the diversity that made human groups valuable in the first place (Fügener, Grahl, Gupta, & Ketter, 2021). 3.3 Cognitive load, offloading, and metacognition Human beings routinely export cognitive work to the environment. We use notes, calendars, and other people. The general principle, sometimes described as #cognitive_offloading, is that we offload when the perceived cost of internal processing exceeds the perceived cost of external retrieval. Agentic systems reduce the external cost of an enormous range of cognitive operations to near zero. The predictable consequence is that offloading expands. The less obvious consequence is that offloading changes what is practiced, and what is not practiced changes. Tankelevitch and colleagues (2024) make the sharpest theoretical contribution here. They argue that generative systems shift the human burden from object level cognition to #metacognition. The person no longer performs the task, but must now judge whether the task was performed correctly, decide how to decompose a goal, estimate their own confidence, and estimate the system's confidence. These are metacognitive operations, and they are exactly the operations that human beings are worst at and least trained in. The load does not disappear. It moves to the weakest joint. Empirical support is accumulating. In a large survey of knowledge workers, Lee and colleagues (2025) found that higher confidence in a generative system was associated with less #critical_thinking effort, while higher confidence in one's own expertise was associated with more. The tool did not remove thinking; it redistributed it according to who felt sure of what. 3.4 Social response and anthropomorphism People respond socially to machines that display social cues, even when they know the machine is not a person. #anthropomorphism has been reviewed extensively in recent years, with evidence that human like cues raise perceived warmth and competence, increase willingness to disclose, and increase both trust and disappointment (Li & Suh, 2022; Blut, Wang, Wunderlich, & Brock, 2021). Language is the strongest social cue available. An agent that writes in the first person, apologizes, hedges, and expresses uncertainty invites a social response by construction. The result is that many people relate to #autonomous_agents through a mixed model. They know it is software. They also thank it, argue with it, feel judged by it, and feel guilty about overriding it. The mixed model is not confusion; it is the normal operation of a mind built for a social world. This produces measurable behavior. People are more polite to systems that appear polite. They accept advice more readily from systems that express appropriate uncertainty. And they experience a specific discomfort, sometimes described as a form of #social_presence, when the system appears to observe their performance (Tong, Jia, Luo, & Fang, 2021). 3.5 Identity, self-efficacy, and status Work is a source of self concept. When a system performs the part of the job that carried prestige, the person's account of who they are becomes unstable. Strich, Mayer, and Fiedler (2021) document this directly among employees whose decisions were substituted by an intelligent system: some rebuilt their #professional_identity around new tasks, while others experienced identity threat and disengagement. #self_efficacy, the belief that one can execute a task successfully, is affected in a subtler way. Efficacy is built primarily through mastery experience. If the system performs the task, the mastery experience belongs to the system, not the person. Over time, output rises while the internal sense of capability stagnates or falls. This dissociation between performance and felt competence is one of the least studied and most important effects of #agentic_workflows, and it is especially relevant for #students, whose entire purpose is to accumulate mastery experiences. 3.6 Moral psychology Finally, agentic systems change moral behavior. Kobis, Bonnefon, and Rahwan (2021) argue that machines can influence human ethics through several channels, including acting as advisors whose suggestions provide moral cover. Krugel, Ostermaier, and Uhl (2023) show experimentally that inconsistent moral advice from a language model shifts users' judgments even when users believe they were unaffected. Bonnefon, Rahwan, and Shariff (2024) review the wider field and note that people apply different moral standards to machine actors than to human ones, which complicates any assignment of blame in a mixed team. The result is a #responsibility_gap. Nobody chose the action. The person specified a goal. The system chose a path. When the path is harmful, moral intuitions have nowhere obvious to land, and the psychological pressure resolves in the easiest direction, which is usually toward #moral_disengagement. 4. Method 4.1 Approach This article is an integrative narrative review. The goal is theory building rather than effect size estimation, because the underlying studies are too heterogeneous in design, task, population, and system for meaningful pooling. Where quantitative synthesis exists in the literature, such as the meta analysis by Vaccaro, Almaatouq, and Malone (2024), it is reported as evidence rather than reproduced. 4.2 Sources and inclusion Literature was assembled from human factors, information systems, organizational behavior, cognitive psychology, human computer interaction, and economics of technology. Priority was given to work published from 2021 onward, since the behavior of interest depends on system capabilities that did not exist earlier. A small number of older, foundational works are cited where a concept has no adequate recent replacement. Studies were included if they reported empirical evidence about human behavior, cognition, attitudes, or identity in the presence of a system with at least some capacity for independent action or generation, or if they offered a theoretical framework directly addressing such interaction. Purely technical papers about model performance were excluded, as were studies of fully manual work and studies of simple recommender interfaces where the human retained complete step by step control. 4.3 Synthesis Findings were coded into behavioral domains through iterative grouping. A domain was retained when it was supported by evidence from at least two independent research traditions, a criterion adopted to guard against importing the local enthusiasms of any single field. Seven domains met this test. They are presented in Section 5, followed by a process model that arranges them in time. 4.4 Limitations of the method A narrative review reflects the judgment of its author. The selection is not exhaustive, publication bias in the underlying literature is likely, and the field is moving faster than its own peer review. Findings from 2023 may describe systems that no longer exist. The response to this problem, adopted throughout, is to emphasize mechanisms that depend on stable features of human cognition rather than on transient features of particular products. 5. Seven Domains of Behavioral Adaptation 5.1 Attention reallocation and the decay of vigilance The first thing that changes is where people look. In manual work, #attention follows action. The person attends to what they are doing because doing requires it. In #agentic_workflows, action is elsewhere, and attention must be sustained without the support of activity. This is a monitoring task, and human beings are known to be poor monitors. Sustained #vigilance degrades within tens of minutes, and it degrades faster when the monitored process is reliable, because reliability removes the events that would otherwise recapture attention. This produces the central paradox of #oversight. The better the system, the worse the human watches it. A system that fails one time in twenty trains the watcher to watch. A system that fails one time in two hundred trains the watcher to stop. Practically, three shifts in #attention have been observed. Attention narrows to the output and away from the process. Because the process is long and opaque, and the output is short and legible, people check what is cheap to check. Errors that are visible only in the process therefore become invisible in practice. Attention shifts from content to surface. Fluent output invites surface reading. Readers report having read a document when they have in fact skimmed for coherence markers, which agentic output always supplies. Attention becomes episodic. Rather than continuous engagement, people check in. #situation_awareness, defined as knowing what is going on well enough to act, therefore falls, and Endsley (2023) argues that restoring it requires deliberate design of transparency rather than more human effort. The consequence is not that people become lazy. It is that the environment stops paying them for attention. #error_detection is unrewarded when errors are rare and unpunished when they are missed, and behavior follows payoff. 5.2 Trust drift: over-reliance and aversion The second domain is the movement of #trust away from a calibrated level. #over_reliance is now well documented. Bucinca, Malaya, and Gajos (2021) showed that people accept incorrect AI recommendations at high rates and, importantly, that giving them explanations does not fix the problem, because explanations are also consumed heuristically. What helped was a cognitive forcing function, an interruption that required the person to commit to their own judgment before seeing the system's. The effective intervention was friction. That finding has been reproduced and refined. Vasconcelos and colleagues (2023) argue that reliance follows a cost benefit logic: people over rely when the cost of thinking exceeds the perceived benefit, and explanations reduce over reliance only when they make verification cheaper than deferral. Schemmer and colleagues (2023) similarly find that explanation quality matters less than whether the explanation supports appropriate #reliance. Klingbeil, Grutzner, and Schreck (2024) show that over reliance persists even when participants are explicitly warned that the system is fallible, and that it carries measurable performance costs. The mirror image, #algorithm_aversion, remains real but is narrower than early accounts suggested. Mahmud, Islam, Ahmed, and Smolander (2022) synthesize the literature and conclude that aversion is strongest in subjective or moral domains, when errors are visible, and when the person feels personally accountable. Berger, Adam, Ruhr, and Benlian (2021) show that aversion softens when the system demonstrably learns from its mistakes, which suggests that people are not rejecting machines as such but rejecting machines that appear incapable of correction. The two failures share a cause. Both reflect a global attitude applied to a system whose competence is local. #trust_calibration fails not because people trust too much or too little on average, but because they trust with insufficient resolution. An important asymmetry deserves attention. Over reliance is quiet. Its costs appear later, elsewhere, and are attributed to something else. Aversion is loud. Its costs appear immediately as wasted time and refused help. Organizations therefore see and punish aversion while remaining blind to over reliance, and rational employees respond by drifting toward acceptance. The bias in the system is not neutral. It pushes in one direction. 5.3 Cognitive offloading, skill change, and metacognition The third domain concerns what happens to the mind that is no longer practicing. The evidence for immediate #cognitive_offloading is unambiguous. Noy and Zhang (2023) found that access to a generative system reduced time spent on professional writing tasks by roughly forty percent while raising quality ratings, with the largest gains among the initially weakest writers. Peng, Kalliamvakou, Cihon, and Demirer (2023) reported that developers using a code completion agent finished a defined task more than fifty percent faster. Brynjolfsson, Li, and Raymond (2025) found substantial productivity gains among customer support agents, again concentrated among the least experienced. These are real gains and should not be minimized. But they describe output, not learning, and the two are not the same. In education, the conditions that produce good performance during practice are often not the conditions that produce durable learning. Effort, retrieval, and productive difficulty build memory; smoothness does not. A tool that removes difficulty removes the thing that was doing the teaching. Direct evidence on longer term #skill_decay in agentic settings is still thin, which is itself a finding worth stating plainly. What exists is suggestive. Lee and colleagues (2025) report that self assessed use of #critical_thinking falls as confidence in the system rises. Studies of #creativity find that individual output improves while collective variety shrinks: Doshi and Hauser (2024) show that generative assistance made individual stories more novel while making the overall set of stories more similar to each other, and Anderson, Shah, and Kreminski (2024) find comparable #homogenization in group ideation. The person gets better and the population gets narrower. There is a further quiet effect on #memory. When retrieval is delegated, encoding weakens. People who know that information will remain accessible remember its location rather than its content. Applied to professional knowledge, this means that practitioners may retain a map of what the system can do while losing the substance the system does. The map is useful right up to the moment the system is wrong, at which point the substance is what was needed. The metacognitive picture is therefore the following. Object level work goes down. Metacognitive work goes up. Metacognitive skill does not automatically follow, because nobody teaches it. The gap between the demand and the skill is where most agentic failures live. 5.4 Effort redistribution and the uneven shape of productivity The fourth domain concerns effort. Effort is not simply reduced by #agentic_workflows. It is moved, and the move is uneven. Dell'Acqua and colleagues (2023) provide the clearest demonstration. Consultants using a capable system performed markedly better on tasks inside the system's competence and markedly worse on a task just outside it, because they accepted plausible but wrong output. Average performance concealed two opposite effects. This is the jagged frontier again, and it establishes that #productivity effects cannot be summarized by a single number. Vaccaro, Almaatouq, and Malone (2024) reach a related conclusion through meta analysis. Across many studies, human machine combinations did not reliably outperform the better of the two alone. Synergy occurred under identifiable conditions, notably when the human held information the system lacked, and it failed when the human's contribution was merely to approve. #hybrid_intelligence is possible but it is not automatic, and assuming it is automatic is how organizations lose the benefit. Where does the effort actually go? Field observation and the studies above suggest four places. It goes into specification, since a poorly stated goal produces confident irrelevance. It goes into #verification, which is often harder than production, because checking an unfamiliar solution requires reconstructing the reasoning that produced it. It goes into integration, since agent output arrives context free and must be reconciled with a situation the agent does not know about. And it goes into repair, since failures in long chains are discovered late and are expensive. The uncomfortable implication is that the saved effort and the added effort fall on different people. The person who prompts saves. The person who verifies pays. When these are the same individual, the workflow is self correcting. When they are different individuals, or when verification is nominally assigned to someone with no time, the workflow silently converts a productivity gain into a hidden liability. 5.5 Identity, status, and motivation The fifth domain is the self. Raisch and Krakowski (2021) describe the automation augmentation paradox: attempts to augment human work reliably create pressure to automate it, and automation removes the human capabilities that augmentation depended on. This paradox is experienced by individuals as ambiguity about what they are for. The evidence on identity threat is consistent across settings. Strich, Mayer, and Fiedler (2021) found that when a system took over the core decision that defined a role, some employees reconstructed their identity around adjacent tasks such as relationship management and exception handling, while others experienced the change as a loss and withdrew. Which path an individual took depended heavily on whether the organization offered a credible new story about what their expertise was now for. Anthony, Bechky, and Fayard (2023) extend this to the system level, arguing that collaboration with intelligent systems changes not only individual roles but the occupational structures that make expertise legible to others. Expertise is a social claim. If the claim can no longer be demonstrated, it decays regardless of whether the underlying skill remains. #motivation follows identity. Work that is experienced as authored is motivating. Work that is experienced as reviewed is less so. The task of correcting a machine's draft carries little of the intrinsic satisfaction of producing one, even when the output is better and the time is shorter. Several observers have noted a specific dissatisfaction reported by skilled workers who are more productive and less content, and this pattern is worth taking seriously rather than dismissing as nostalgia. Meaning is a genuine input to sustained performance, not a decoration on top of it. For #students the identity question is sharper still, because they have no prior identity to defend. A student who has never written an unassisted essay has no baseline sense of their own capacity. Their #self_efficacy is not threatened; it is simply never built. This is a different and possibly more serious problem than #deskilling, because deskilling implies a skill that existed. 5.6 Responsibility, blame, and moral drift The sixth domain concerns who is answerable. Autonomous action creates a gap between the person who set the goal and the action that caused the harm. Legally, the gap is usually closed by assigning liability to the human. Psychologically, it is not closed at all, and psychology governs behavior. Three effects follow. Blame is diffused, since responsibility spreads thinly across the specifier, the reviewer, the vendor, and the system, and thin responsibility is weak responsibility. #moral_disengagement becomes available, since a person who did not perform an act finds it easier to distance themselves from its consequences, and Kobis and colleagues (2021) argue that machine advisors can serve precisely this function by supplying justification. And moral judgment itself shifts, since Krugel and colleagues (2023) show that exposure to machine generated moral advice moves human moral positions even when people deny being influenced, which means the influence is not something people can simply decide to resist. Bonnefon, Rahwan, and Shariff (2024) note that people hold machines to different standards than humans, often demanding higher reliability while simultaneously excusing machine errors as accidents rather than faults. These two attitudes coexist without difficulty, and the combination allows an organization to demand perfection from a system in principle while forgiving it in practice. The result is that #accountability tends to become ceremonial. A named human signs off. That human did not do the work, cannot verify the work in the time available, and knows that signing is expected. The signature transfers legal risk without transferring cognitive control, which is the worst possible arrangement, since it satisfies #governance requirements while providing no actual #oversight. 5.7 Social behavior toward non-human actors The seventh domain is the social relationship itself. People form relationships with conversational systems. Skjuve, Folstad, Fostervold, and Brandtzaeg (2021) documented the development of perceived closeness with a chatbot companion over time, following a trajectory that resembles human relationship development, and this occurred despite full user awareness that the partner was software. In work settings the relationship is less intimate but no less real. People report frustration, gratitude, and something close to embarrassment. They apologize to agents. They soften their instructions. They hesitate to override an agent that has been right several times, in a manner that resembles deference to a competent colleague rather than adjustment of a tool. Sundar and Lee (2022) argue that this requires rethinking the basic models of communication, since a message source that is neither human nor a simple channel does not fit existing categories. The practical consequence for #agentic_workflows is that social dynamics familiar from human teams start to appear. There is deference to perceived seniority. There is reluctance to contradict in front of others. There is a tendency to accept a plan because rejecting it feels rude, an impulse that has no rational basis and considerable behavioral force. At the same time, and without contradiction, people can be dismissive and even abusive toward the same systems, particularly after failure. The relationship is unstable because the category is unstable. A #machine_teammate is treated as a colleague when it succeeds and a tool when it fails, and this asymmetry protects the human's self concept at the cost of accurate #mental_models. 6. An Integrative Model: Five Stages of Adaptation The seven domains are not independent, and they do not appear simultaneously. The following stage model organizes them over time. It is offered as a theoretical proposition to be tested, not as an established finding. 6.1 Stage one: contact The person encounters the system. Behavior is dominated by curiosity, testing, and exaggerated attention. Verification is near total, because the person does not yet know what the system can do. Performance gains are modest, since the cost of checking cancels the benefit of automation. Emotionally, the stage is marked by novelty and mild threat. The dominant psychological process is #mental_models formation, and this is where the trouble begins, because the models people form at this stage are built from a small and unrepresentative sample of interactions. 6.2 Stage two: calibration The person begins to form beliefs about reliability. Verification becomes selective. If the system has performed well on the sampled tasks, #trust rises quickly; a small number of visible successes is sufficient, because human trust updating is asymmetric and generous when the stakes have not yet been felt. Calibration is where most interventions are aimed and where most interventions fail, because the information required for good calibration is precisely the information that agentic systems do not naturally provide. The person needs to know the boundary of competence. They observe only outcomes. 6.3 Stage three: delegation Trust reaches a level at which the person hands over categories of work rather than individual tasks. Effort is redistributed. #productivity rises, often sharply. This is the stage that organizations observe, celebrate, and measure. It is also the stage at which the person's practical skill stops being exercised. Nothing bad has happened yet. Nothing bad is visible yet. The system is working and the person is getting more done. Every incentive points toward more #delegation. 6.4 Stage four: drift Drift is the central concept of this model. It is the slow, unmonitored, and individually rational movement of behavior away from safe #oversight. Drift has four components that reinforce one another. #vigilance falls because errors are rare. #verification thins because checking is costly and unrewarded. #skill_decay proceeds because unused capacities weaken, which makes verification harder, which thins it further. And #accountability becomes ceremonial, which removes the last incentive to resist the first three. Drift is not a failure of character. Each individual step is sensible. A person who checks every output of a system that is right ninety nine times in a hundred is wasting their employer's money and their own life. The trouble is that the ninety nine successes do not warn them about the one failure, and by the time it arrives their capacity to catch it has quietly gone. Drift is invisible from inside. It is measurable only from outside, and only if someone chooses to measure it, which is why it usually is not. 6.5 Stage five: restructuring Eventually something forces the system into the open. A serious failure, an audit, a regulatory demand, a new hire who asks an obvious question. At this point the workflow is restructured, and one of three outcomes follows. In the first, the human role is rebuilt around genuine complementarity. The person is given the tasks where they add real information, is trained in #verification and #metacognition, and is measured on #error_detection rather than throughput. This is the outcome that #hybrid_intelligence research recommends and it is rare, because it costs money and slows things down. In the second, the human role is formalized as ceremonial. Sign off procedures are added without adding time or capability. Compliance improves and safety does not. This is the most common outcome, because it is cheap. In the third, the human is removed. If the human contribution was in fact only approval, and approval was not adding anything, then removing it is honest. This outcome is less common than feared and more defensible than admitted. 6.6 The model as a warning The value of the stage model is that it locates the danger correctly. The danger is not stage one, where people are anxious about a new technology and researchers write about #algorithm_aversion. The danger is stage four, where everyone is satisfied, output is high, and the human competence that the whole arrangement depends on is being quietly consumed. 7. Moderators Adaptation is not uniform. Five classes of moderators appear consistently. 7.1 Individual differences Domain #expertise is the most powerful moderator, and it works in two directions. Experts verify better because they can detect errors that novices cannot see. Experts also sometimes verify less, because they are confident and because the system flatters their existing beliefs. Lebovitz and colleagues (2022) show that professionals with strong tacit knowledge were the most likely to engage critically, but only when they had a way to test the system's output against something they trusted. Need for cognition, tolerance of ambiguity, and metacognitive accuracy all predict better calibration, though the effects are modest. General #AI_literacy predicts appropriate #reliance more reliably than raw technical skill, which suggests that knowing how a system fails matters more than knowing how it works. Age and experience interact in a way that deserves attention. Novices gain the most in output and lose the most in learning. This is the central asymmetry facing education. 7.2 Task characteristics Tasks with clear ground truth support calibration, because errors are visible. Tasks with delayed or contested feedback prevent calibration entirely and drive behavior toward whatever is socially rewarded. Tasks with high stakes increase verification, but only up to the point where verification becomes impossible, after which they increase ceremonial approval instead. Tasks with subjective quality criteria produce the strongest #algorithm_aversion and also the strongest #homogenization when aversion is overcome. 7.3 System characteristics #transparency helps, but not in the way people expect. Zerilli, Bhatt, and Weller (2022) argue that transparency modulates trust rather than simply increasing it, and can reduce trust appropriately when the system's limits become visible. Explanations that make verification cheaper improve outcomes; explanations that merely add persuasive text make things worse, because they add fluency without adding information. Expressions of #uncertainty by the system are among the most promising design levers, provided the uncertainty is well calibrated. A system that says it is unsure gives the human a reason to look, which is the only mechanism that reliably reverses drift. 7.4 Organizational context The organization decides what is rewarded, and behavior follows. If throughput is measured and error catching is not, drift is guaranteed. If #psychological_safety is low, people will not report that the system was wrong, because reporting implies that they either used it badly or failed to catch it. Silence about failure is the single most effective way to prevent calibration. Kellogg, Valentine, and Christin (2020) analyze algorithmic control at work and show that systems which evaluate as well as assist produce defensive behavior rather than learning. Rahman (2021) demonstrates that opaque evaluation leads workers to change their behavior to satisfy an imagined algorithm rather than the actual task, an effect that generalizes readily to any workflow in which the agent's judgment is also a record. 7.5 Culture Cross cultural evidence remains thin, which is a real weakness in the field. What exists suggests differences in the acceptability of machine authority and in the willingness to override, but the studies are few and the constructs are not equivalent across settings. This is stated here as a gap rather than a finding. 8. Discussion 8.1 What this means for students Students occupy the most exposed position in this transition, and they are usually addressed with rules rather than with understanding. The following implications follow from the evidence rather than from anxiety. The purpose of an assignment is not the artifact. It is the change in the student. If a student produces an excellent essay and is unchanged, the assignment has failed, no matter how good the essay is. This is not a moral claim about cheating. It is a claim about what education is. Difficulty is not an obstacle to #learning. It is the mechanism of learning. Removing it feels like progress and is not. The productive response is not to refuse the tools, which is neither possible nor sensible, but to be deliberate about which difficulties are preserved. A workable practical rule is to use agents for tasks whose skill you already possess and do not need to build, and to protect from them the tasks whose skill you are currently building. A student learning to construct an argument should construct arguments. The same student, formatting a bibliography, is learning nothing by hand and should delegate freely. The most valuable and least taught skill in an agentic world is #verification. Being able to look at a plausible answer and determine whether it is correct is a distinct capacity, harder than producing the answer, and almost entirely absent from current curricula. It should be taught explicitly, assessed explicitly, and practiced on material where the errors are real. Finally, students should be told the truth about #self_efficacy. The feeling of competence that comes from a good result produced by a system is not evidence of competence. It is evidence of a good result. The distinction will matter later, in a room where the system is not available or is wrong, and the habit of confusing the two is expensive to unlearn. 8.2 What this means for educators Assessment that measures artifacts will measure the agent. This is not a hypothetical risk; it is the current situation. Assessment must therefore move toward processes, defenses, and live reasoning, which is more expensive and is the price of the change. Educators should also consider teaching #agent_supervision directly. Students will spend their careers supervising non-human actors. They currently receive no instruction in how to do it. Giving them deliberately flawed agent outputs and asking them to find the flaws is one of the few interventions with clear support in the reliance literature, because it restores the error signal that fluent systems suppress. 8.3 What this means for organizations Three recommendations follow from the stage model. Measure drift, not just output. If nobody knows how often the system is wrong and how often that is caught, the organization is flying blind and its human #oversight is decorative. Deliberate insertion of known errors, sometimes used in security and quality assurance, is the most direct available method, and it is uncomfortable precisely because it works. Make #verification a real job with real time and real reward. A reviewer who is measured on throughput will approve. This is not a failure of professionalism. It is a correct response to the incentives that were set. Preserve #expertise deliberately. If a capability is needed only when the system fails, it will not be maintained by ordinary work, because ordinary work no longer uses it. It must be maintained on purpose, through practice that has no immediate productive value, which is exactly the kind of expenditure that efficiency pressure eliminates first. 8.4 Reframing the central problem The debate about autonomous agents is usually conducted as a debate about capability. Will the system be good enough. The evidence reviewed here suggests a different framing. Even a highly capable system produces poor outcomes if the human alongside it drifts, and a moderately capable system produces excellent outcomes if the human retains calibrated reliance and genuine oversight. The binding constraint on the value of agentic workflows is therefore human behavioral adaptation, not machine capability. This is good news, because human systems can be designed, and it is bad news, because designing them requires slowing down. 9. Research Agenda Six priorities follow from the gaps identified in this review. First, longitudinal studies. Almost everything known about reliance comes from single session experiments. Drift is a temporal phenomenon and cannot be observed in an hour. Studies running six months or longer, in real workplaces, with objective error data, are the highest value work available in this field and are almost entirely missing. Second, measurement of drift. There is no accepted instrument for measuring the erosion of verification behavior over time. Candidate measures include verification depth per output, latency before acceptance, override rate conditional on system error, and detection rate for seeded errors. Standardizing these would let separate research groups compare results, which they currently cannot. Third, skill trajectories. The question of whether prolonged agent use degrades underlying competence, and under what conditions, is open, important, and answerable. It requires baseline measurement, controlled exposure, and unassisted retest. The design is not difficult. It has simply not been done at scale. Fourth, metacognitive training. If Tankelevitch and colleagues (2024) are right that the burden has shifted to metacognition, then metacognitive training should improve outcomes in agentic settings. This is a clean, testable prediction and a rare case where a theoretical claim yields an immediate intervention. Fifth, the multi agent case. Nearly all evidence concerns a single human and a single system. Real workflows increasingly involve several agents, several humans, and agents that instruct other agents. Nothing is known about how trust, blame, and attention behave in these settings, and there is no reason to expect the single agent findings to transfer. Sixth, cultural and demographic variation. The field's participant pool is narrow. Claims about human psychology built on that pool should be treated as provisional until tested more widely. 10. Limitations This article is a narrative synthesis and inherits the weaknesses of the form. The selection of sources reflects a single perspective. The seven domain structure is an interpretive scheme and other defensible schemes exist. The stage model is a theoretical proposal supported by convergent evidence rather than a validated process model, and it has not been tested as a whole. The literature itself has three limitations that the review cannot repair. It is dominated by short experiments. It is dominated by Western, educated, English speaking participants. And it studies systems that are replaced faster than they can be studied, which means that every empirical claim about a specific capability carries an expiry date. The stable part, and the part on which this article rests, is the human being. The tendency to trust fluency, to stop watching what rarely fails, to offload what is costly, to protect the self concept, and to diffuse blame across a group are not features of 2025. They will still be here when the current systems are museum pieces. 11. Conclusion Working with an autonomous agent is not an upgraded form of using a tool. It is a relationship with a non-human actor that has initiative, that cannot be fully inspected, and that speaks fluently whether or not it is right. Human psychology responds to that situation with a set of adaptations that are individually reasonable and collectively dangerous. People stop watching what rarely fails. They form a single global estimate of trust for a system whose competence is local and jagged. They offload cognitive work and, over time, lose the capacity that would let them check it. They redistribute effort toward specification and verification without being given the time or training to do either well. They renegotiate who they are around a diminished task. They allow accountability to become a signature. And they treat the system as a colleague when it succeeds and a tool when it fails. None of this is inevitable, and none of it is a reason to reject agentic workflows, which deliver real and substantial value. It is a reason to design the human side of the workflow with as much care as the machine side, and currently the human side is not designed at all. It is left to drift, and drift is not a neutral process. It has a direction, and the direction is toward less oversight, less skill, and less responsibility, arriving at exactly the moment when all three are needed. The most important recommendation in this article is therefore also the simplest. Watch the human. That is where the system will fail. References Anderson, B. R., Shah, J. H., & Kreminski, M. (2024). Homogenization effects of large language models on human creative ideation. In Proceedings of the 16th Conference on Creativity and Cognition (pp. 413 to 425). Association for Computing Machinery. Anthony, C., Bechky, B. A., & Fayard, A. L. (2023). Collaborating with artificial intelligence: Taking a system view to explore the future of work. Organization Science, 34(5), 1672 to 1694. Baird, A., & Maruping, L. M. (2021). The next generation of research on IS use: A theoretical framework of delegation to and from agentic IS artifacts. MIS Quarterly, 45(1), 315 to 341. Bansal, G., Wu, T., Zhou, J., Fok, R., Nushi, B., Kamar, E., Ribeiro, M. T., & Weld, D. S. (2021). Does the whole exceed its parts? The effect of AI explanations on complementary team performance. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems (pp. 1 to 16). Association for Computing Machinery. Berger, B., Adam, M., Ruhr, A., & Benlian, A. (2021). Watch me improve: Algorithm aversion and demonstrating the ability to learn. Business and Information Systems Engineering, 63(1), 55 to 68. Blut, M., Wang, C., Wunderlich, N. V., & Brock, C. (2021). Understanding anthropomorphism in service provision: A meta analysis of physical robots, chatbots, and other AI. Journal of the Academy of Marketing Science, 49(4), 632 to 658. Bonnefon, J. F., Rahwan, I., & Shariff, A. (2024). The moral psychology of artificial intelligence. Annual Review of Psychology, 75, 653 to 675. Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889 to 942. Bucinca, Z., Malaya, M. B., & Gajos, K. Z. (2021). To trust or to think: Cognitive forcing functions can reduce overreliance on AI in AI assisted decision making. Proceedings of the ACM on Human Computer Interaction, 5(CSCW1), 1 to 21. Cameron, L. D., & Rahman, H. (2022). Expanding the locus of resistance: Understanding the co constitution of control and resistance in the gig economy. Organization Science, 33(1), 38 to 58. Chen, V., Liao, Q. V., Wortman Vaughan, J., & Bansal, G. (2023). Understanding the role of human intuition on reliance in human AI decision making. Proceedings of the ACM on Human Computer Interaction, 7(CSCW2), 1 to 32. Chugunova, M., & Sele, D. (2022). We and it: An interdisciplinary review of the experimental evidence on how humans interact with machines. Journal of Behavioral and Experimental Economics, 99, 101897. Dell'Acqua, F., McFowland, E., Mollick, E., Lifshitz Assaf, H., Kellogg, K., Rajendran, S., Krayer, L., Candelon, F., & Lakhani, K. R. (2023). Navigating the jagged technological frontier: Field experimental evidence of the effects of AI on knowledge worker productivity and quality. Harvard Business School Technology and Operations Management Unit Working Paper 24-013. Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. Endsley, M. R. (2023). Supporting human AI teams: Transparency, explainability, and situation awareness. Computers in Human Behavior, 140, 107574. Fugener, A., Grahl, J., Gupta, A., & Ketter, W. (2021). Will humans in the loop become borgs? Merits and pitfalls of working with AI. MIS Quarterly, 45(3), 1527 to 1556. Fugener, A., Grahl, J., Gupta, A., & Ketter, W. (2022). Cognitive challenges in human artificial intelligence collaboration: Investigating the path toward productive delegation. Information Systems Research, 33(2), 678 to 696. Hancock, P. A., Kessler, T. T., Kaplan, A. D., Brill, J. C., & Szalma, J. L. (2021). Evolving trust in robots: Specification through sequential and comparative meta analyses. Human Factors, 63(7), 1196 to 1229. Hemmer, P., Schemmer, M., Vossing, M., & Kuhl, N. (2021). Human AI complementarity in hybrid intelligence systems: A structured literature review. In Proceedings of the 25th Pacific Asia Conference on Information Systems. Kaplan, A. D., Kessler, T. T., Brill, J. C., & Hancock, P. A. (2023). Trust in artificial intelligence: Meta analytic findings. Human Factors, 65(2), 337 to 359. Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366 to 410. Kim, S. S. Y., Watkins, E. A., Russakovsky, O., Fong, R., & Monroy Hernandez, A. (2023). Humans, AI, and context: Understanding end users trust in a real world computer vision application. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (pp. 77 to 88). Klingbeil, A., Grutzner, C., & Schreck, P. (2024). Trust and reliance on AI: An experimental study on the extent and costs of overreliance on AI. Computers in Human Behavior, 160, 108352. Kobis, N., Bonnefon, J. F., & Rahwan, I. (2021). Bad machines corrupt good morals. Nature Human Behaviour, 5(6), 679 to 685. Krugel, S., Ostermaier, A., & Uhl, M. (2023). ChatGPT's inconsistent moral advice influences users judgment. Scientific Reports, 13, 4569. Lai, V., Chen, C., Smith Renner, A., Liao, Q. V., & Tan, C. (2023). Towards a science of human AI decision making: An overview of design space in empirical human subject studies. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency (pp. 1369 to 1385). Lebovitz, S., Levina, N., & Lifshitz Assaf, H. (2021). Is AI ground truth really true? The dangers of training and evaluating AI tools based on experts know what. MIS Quarterly, 45(3), 1501 to 1526. Lebovitz, S., Lifshitz Assaf, H., & Levina, N. (2022). To engage or not to engage with AI for critical judgments: How professionals deal with opacity when using AI for medical diagnosis. Organization Science, 33(1), 126 to 148. Lee, H. P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery. Li, M., & Suh, A. (2022). Anthropomorphism in AI enabled technology: A literature review. Electronic Markets, 32(4), 2245 to 2275. Ma, S., Lei, Y., Wang, X., Zheng, C., Shi, C., Yin, M., & Ma, X. (2023). Who should I trust: AI or myself? Leveraging human and AI correctness likelihood to promote appropriate reliance in AI assisted decision making. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (pp. 1 to 19). Mahmud, H., Islam, A. K. M. N., Ahmed, S. I., & Smolander, K. (2022). What influences algorithmic decision making? A systematic literature review on algorithm aversion. Technological Forecasting and Social Change, 175, 121390. Mollick, E. (2024). Co intelligence: Living and working with AI. Portfolio. Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187 to 192. O'Neill, T., McNeese, N., Barron, A., & Schelble, B. (2022). Human autonomy teaming: A review and analysis of the empirical literature. Human Factors, 64(5), 904 to 938. Passi, S., & Vorvoreanu, M. (2022). Overreliance on AI: Literature review. Microsoft Research Technical Report MSR-TR-2022-12. Peng, S., Kalliamvakou, E., Cihon, P., & Demirer, M. (2023). The impact of AI on developer productivity: Evidence from GitHub Copilot. Working paper. Rahman, H. A. (2021). The invisible cage: Workers reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945 to 988. Raisch, S., & Krakowski, S. (2021). Artificial intelligence and management: The automation augmentation paradox. Academy of Management Review, 46(1), 192 to 210. Rastogi, C., Zhang, Y., Wei, D., Varshney, K. R., Dhurandhar, A., & Tomsett, R. (2022). Deciding fast and slow: The role of cognitive biases in AI assisted decision making. Proceedings of the ACM on Human Computer Interaction, 6(CSCW1), 1 to 22. Schemmer, M., Kuhl, N., Benz, C., Bartos, A., & Satzger, G. (2023). Appropriate reliance on AI advice: Conceptualization and the effect of explanations. In Proceedings of the 28th International Conference on Intelligent User Interfaces (pp. 410 to 422). Shneiderman, B. (2022). Human centered AI. Oxford University Press. Skjuve, M., Folstad, A., Fostervold, K. I., & Brandtzaeg, P. B. (2021). My chatbot companion: A study of human chatbot relationships. International Journal of Human Computer Studies, 149, 102601. Steyvers, M., & Kumar, A. (2024). Three challenges for AI assisted decision making. Perspectives on Psychological Science, 19(5), 722 to 734. Strich, F., Mayer, A. S., & Fiedler, M. (2021). What do I do in a world of artificial intelligence? Investigating the impact of substitutive decision making AI systems on employees professional role identity. Journal of the Association for Information Systems, 22(2), 304 to 324. Sturm, T., Gerlach, J. P., Pumplun, L., Mesbah, N., Peters, F., Tauchert, C., Nan, N., & Buxmann, P. (2021). Coordinating human and machine learning for effective organizational learning. MIS Quarterly, 45(3), 1581 to 1602. Sundar, S. S., & Lee, E. J. (2022). Rethinking communication in the era of artificial intelligence. Human Communication Research, 48(3), 379 to 385. Tankelevitch, L., Kewenig, V., Simkute, A., Scott, A. E., Sarkar, A., Sellen, A., & Rintel, S. (2024). The metacognitive demands and opportunities of generative AI. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems (pp. 1 to 24). Tong, S., Jia, N., Luo, X., & Fang, Z. (2021). The Janus face of artificial intelligence feedback: Deployment versus disclosure effects on employee performance. Strategic Management Journal, 42(9), 1600 to 1631. Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta analysis. Nature Human Behaviour, 8(12), 2293 to 2303. Vasconcelos, H., Jorke, M., Grunde McLaughlin, M., Gerstenberg, T., Bernstein, M. S., & Krishna, R. (2023). Explanations can reduce overreliance on AI systems during decision making. Proceedings of the ACM on Human Computer Interaction, 7(CSCW1), 1 to 38. Vicente, L., & Matute, H. (2023). Humans inherit artificial intelligence biases. Scientific Reports, 13, 15737. Wang, X., & Yin, M. (2021). Are explanations helpful? A comparative study of the effects of explanations in AI assisted decision making. In Proceedings of the 26th International Conference on Intelligent User Interfaces (pp. 318 to 328). Xu, W., Dainoff, M. J., Ge, L., & Gao, Z. (2023). Transitioning to human interaction with AI systems: New challenges and opportunities for HCI professionals to enable human centered AI. International Journal of Human Computer Interaction, 39(3), 494 to 518. Zerilli, J., Bhatt, U., & Weller, A. (2022). How transparency modulates trust in artificial intelligence. Patterns, 3(4), 100455.

Diplomacy is often described as a game of interests, but it is carried out by tired human beings in rooms that stay lit long past midnight. This article brings together two bodies of knowledge that rarely speak to each other: the sociology of the #diplomatic_field, built on Pierre Bourdieu's concept of habitus, and the psychology of fatigue, recovery, and #resilience. The aim is to explain why negotiators behave the way they do when they are exhausted, and why some ministries produce durable professionals while others burn them out. The article proceeds as a structured conceptual review. It synthesises recent work on habitus in International Relations, recent evidence on cognitive fatigue and sleep loss in negotiation, and current thinking on resilience as flexible self-regulation rather than as a fixed personality trait. From this synthesis it proposes a three-level framework, the Habitus-Fatigue-Adaptation model, which treats fatigue not as a private medical problem but as a socially distributed condition shaped by rank, class, gender, delegation size, and institutional design. The article argues that the diplomatic habitus, the deep set of dispositions that make a diplomat look calm, patient, and competent, is itself a partial cause of #negotiation_fatigue, because it rewards the concealment of strain. It closes with testable propositions, practical recommendations for foreign ministries, and an agenda for future empirical research on #diplomatic_wellbeing. Keywords: habitus; diplomacy; negotiation fatigue; resilience; self-regulation; foreign service; multilateral negotiation; occupational psychology 1. Introduction At around three in the morning, in the final hours of a multilateral conference, a delegate rises to accept a compromise text that she would have rejected on the first day. Nothing in the legal substance has changed. What has changed is her. She has been awake for twenty hours, she has read six versions of the same paragraph, and she has run out of the mental fuel required to fight one more round. Observers will later describe the outcome as a diplomatic success or a diplomatic failure. Almost no one will describe it as what it partly was: a decision made by a #depleted mind inside a system designed, whether by accident or by design, to deplete minds. This article treats that scene as a serious object of study. It asks a simple question with complicated roots. What happens to the professional self of a diplomat when the demands of the job exceed the resources available to meet them, and how do the unwritten rules of the profession shape both the exhaustion and the response to it? The question matters for three reasons. First, it matters practically. Modern #diplomacy is not a leisurely craft. Negotiators in trade, climate, arms control, and humanitarian access work under time pressure, in foreign languages, across time zones, and increasingly under public scrutiny in real time. The endgame of a major conference is routinely conducted in a state that occupational health researchers would classify as unsafe in almost any other high-stakes profession, from aviation to surgery. A scoping review of the mental health of #diplomatic_personnel found that the professional group has been studied surprisingly little, despite a job profile that combines repeated relocation, family disruption, security threats, and long hours (Brooks et al., 2023). Second, it matters theoretically. International Relations has spent two decades importing the sociology of Pierre Bourdieu to explain how diplomats acquire a shared sense of the game. The concept of #habitus, meaning the durable dispositions people carry from their upbringing and training, has been used to explain why diplomats from very different states nevertheless recognise one another as members of the same profession. Yet the literature has stayed mostly at the level of status, taste, and #symbolic_capital. It has said very little about the body that carries the habitus, and almost nothing about what happens to that body when it is starved of sleep. As Nair (2024) argues, IR has often used habitus as a convenient label rather than as a serious account of how dispositions are produced. A psychology of the tired diplomat is one way to take the production of habitus seriously. Third, it matters normatively. If exhaustion is unequally distributed, then so is the capacity to hold a position under pressure. A delegation of forty from a large state can rotate its negotiators through the night. A delegation of two from a small island state cannot. Fatigue, in this light, becomes a quiet form of #power. It is not neutral. It advantages those who can afford to spread the load. 1.1 The gap in the literature Three separate conversations are currently taking place, and they do not hear each other. The first is the sociological conversation in International Relations about the #diplomatic_field. It has produced excellent work on how class, education, language, and gender determine who is recognised as a competent diplomat (Standfield, 2022; Huju, 2023; Nair, 2024). This conversation is strong on structure and weak on the mind. The second is the psychological conversation about #cognitive_fatigue, self-control, and recovery. It has produced strong experimental and neuroscientific evidence about what long periods of mental effort do to decision making (Wiehler et al., 2022), and about the conditions under which people recover (Sonnentag, Cheng, & Parker, 2022). This conversation is strong on the mind and almost silent on the political field in which minds operate. The third is the applied conversation inside foreign ministries and international organisations about wellbeing, staff retention, and #burnout. It is largely practical, often internal, and rarely connected to theory at all. This article argues that the three conversations need each other. Without the sociology, the psychology of fatigue treats the negotiator as a generic laboratory participant. Without the psychology, the sociology of diplomacy treats the diplomat as a body-less bearer of capital. Without theory, the practical conversation about wellbeing produces resilience workshops that may not work, and in some cases appear to be associated with worse outcomes when introduced too late (Gudmundsdottir et al., 2023). 1.2 Aims and contribution The article has four aims. To explain, in accessible language, what habitus means and why it is useful for understanding diplomats. To review current psychological evidence on fatigue, sleep loss, and #self_regulation, including evidence that complicates the popular story. To propose an integrated framework, the Habitus-Fatigue-Adaptation (HFA) model, that connects the two. To derive testable propositions and practical implications for #foreign_ministries. The contribution is conceptual rather than empirical. No new data are collected. The value lies in the integration, in the honest treatment of contested evidence, and in the reframing of #negotiation_fatigue as a structural rather than an individual problem. 1.3 Structure of the article Section 2 reviews the literature in three parts: habitus and the diplomatic field, the psychology of fatigue, and the modern science of resilience. Section 3 sets out the HFA framework and its propositions. Section 4 describes the method used to select and synthesise the literature. Section 5 presents the analysis, organised around five sources of strain in diplomatic work. Section 6 discusses implications for theory, institutions, and training. Section 7 states limitations. Section 8 concludes with a research agenda. 2. Literature Review 2.1 Habitus, field, and capital: the sociological toolkit Pierre Bourdieu offered a way out of an old argument in social science. One side said that people are shaped by structures. The other said that people act freely. Bourdieu said that both are partly right, and that the link between them is the body. His answer was #habitus. Habitus is the set of dispositions a person acquires through upbringing, schooling, and professional training. It includes taste, accent, posture, timing, humour, and a sense of what is realistic to want. It is not a rulebook that people consult. It is closer to a feel for the game, in the way that an experienced footballer does not calculate where to run but simply finds himself there. Habitus is durable, it is largely unconscious, and it is transposable, meaning it travels with the person from one situation to another. Habitus operates inside a #field. A field is a structured arena of competition with its own stakes and its own currency. The literary field, the legal field, and the diplomatic field each have their own rules about what counts as a good move. The currency of a field is #capital, which Bourdieu divided into economic capital, cultural capital such as education and refinement, social capital such as networks, and #symbolic_capital, which is the recognition and prestige that others grant you. The value of these three tools together is that they explain reproduction. People who possess the habitus that a field rewards find the field easy, and their ease is read by others as talent. People who do not possess it must work harder to produce the same effect, and their effort is often visible, which itself counts against them. This is why inequality in professions tends to persist even when formal barriers are removed. 2.2 The diplomatic field and the diplomatic habitus Diplomacy is an unusually clear case of a Bourdieusian field. It has an entrance exam, a hierarchy of posts, an internal language, and a shared aesthetic of behaviour. It also has a highly specific ideal of the competent professional: measured, unflappable, discreet, fluent, and capable of disagreeing without appearing to disagree. Recent scholarship has traced how this ideal is produced and who is excluded by it. Huju (2023) shows how Indian diplomats have historically been recruited into an elite cosmopolitan style of belonging in order to be taken seriously in a diplomatic order shaped by Western norms, and how ambivalent that inheritance feels from the inside. Standfield (2022) shows through the career of a British diplomat and United Nations official that recognition of #diplomatic_competence is shaped by gender, race, and class, and that the same action can be read as virtuosic when performed by one body and as inappropriate when performed by another. Nair (2024) reviews the wider use of habitus in International Relations and argues that scholars must pay closer attention to how habitus is actually produced, through family, schooling, language, and class, rather than using the word as a synonym for practice. Three features of the diplomatic habitus matter for the argument of this article. Composure as capital. In most professions, visible strain is tolerated. In diplomacy, it is costly. A negotiator who looks tired is assumed to be weak, and weakness invites pressure. Composure therefore functions as a form of #symbolic_capital that must be maintained at all times, including in the twentieth hour of a session. Maintaining a calm exterior while feeling something else is what organisational psychologists call surface acting, and it is not free. It requires continuous #emotional_labour. Availability as virtue. The diplomatic field rewards presence. To be in the room is to matter. Delegates therefore compete to remain in rooms, and leaving the room to sleep can be read as a lack of commitment or, worse, as a signal that one's government does not really care about the issue. The result is a professional culture in which #self_neglect is not merely tolerated but quietly admired. Endurance as merit. Long hours are frequently recounted in the profession as badges of honour. Stories of all-night sessions are told with pride. This is an ordinary feature of elite professions, but in diplomacy it has an unusual consequence, because the decisions taken during those hours are binding on states and sometimes on millions of people. Taken together, these features mean that the diplomatic habitus systematically hides the very condition this article is about. #Fatigue exists, but the field trains its members not to display it, and the field's assessment of #competence penalises those who do. 2.3 The psychology of fatigue: a contested science It would be convenient to say that psychology has a settled account of what happens to a tired negotiator. It does not. What it has is a lively and instructive argument, and the argument itself is useful. The strength model and its crisis. For two decades the dominant idea was ego depletion. It held that #self_control draws on a limited resource, so that after a first act of restraint a person has less capacity for a second one. The model was intuitive and it generated an enormous literature. Then came the replication crisis. Large multi-laboratory replication projects failed to find the effect reliably, and later attempts to test the phenomenon under optimal conditions produced mixed results. Englert and Bertrams (2021) reviewed one such multi-site preregistered test and concluded, pointedly, that the field still has neither strong evidence for nor strong evidence against depletion, because the manipulations used may not be doing what researchers assume. Miyake and Carruth (2023) examined a popular moderator, the belief that willpower is unlimited, and found that the supporting evidence is weak at best. The lesson is not that fatigue is a myth. The lesson is that the metaphor of a fuel tank draining is probably wrong, and that researchers should be cautious about any framework built on it. This is important for our purposes because a great deal of practitioner literature on #decision_fatigue in negotiation, leadership, and law rests on the fuel tank picture. Newer accounts: cost, opportunity, and metabolism. More recent work replaces the fuel tank with two better ideas. The first is that #mental_effort is costly rather than finite. Effort feels unpleasant, and the brain tracks the value of continuing to spend it against the value of doing something else. Under this account, a tired negotiator has not run out of anything. She has reached a point at which her system rates further effort as poor value, and she becomes more willing to take the easy option. The second is metabolic. Wiehler and colleagues (2022) had participants perform either high-demand or low-demand cognitive control tasks across roughly a working day, while measuring brain chemistry with magnetic resonance spectroscopy and tracking economic choices. In the high-demand group only, they observed a shift in preference towards options that were immediate and required little effort, a reduction in pupil dilation during deliberation, and an accumulation of glutamate in the lateral prefrontal cortex, a region central to cognitive control. Their interpretation is that a hard day of mental work leaves a by-product in the control system that makes further control more expensive to mobilise. Translate that into a negotiation. A negotiator at hour eighteen is not simply "tired" in a vague sense. She is a person whose control system has become expensive to run, who now finds immediate and low-cost options more attractive than delayed and high-cost ones, and who is therefore more likely to accept a text now rather than hold out for a better text later. This is the psychological mechanism behind what practitioners call #closing_pressure, and it explains why endgames are so often scheduled at night. Sleep loss: the surprising evidence. If the story ended there, we would expect a straightforward finding: sleep-deprived negotiators reach worse agreements. The evidence is more interesting than that. Hausser, Halfmann, and Huffmeier (2022) set out the theoretical case for expecting sleep deprivation to damage negotiation, noting that international summits, government formation talks, and debt negotiations are routinely conducted by people who have not slept. Sleep loss is known to impair working memory, reduce the ability to take advice, flatten mood regulation, and narrow attention, all of which look fatal for the discovery of joint gains. Then Halfmann, Huffmeier, Faber, and Hausser (2025) actually ran the experiments. Across three experiments involving sleep-deprived and rested dyads, plus a Bayesian meta-analysis across studies, they found no evidence that sleep deprivation reduced the joint economic quality of agreements. Laypeople expected it would. It did not, at least in their integrative bargaining task. In a further component of the research they interviewed a set of elected politicians, including senior office holders, about how they cope with negotiating while sleep deprived, which pointed towards active #compensation strategies. This finding should be handled carefully rather than dismissed or overstated. It does not prove that all-night negotiation is harmless. It suggests, at minimum, three things. First, that experienced negotiators, and possibly even inexperienced ones in structured tasks, can compensate for a bad night by drawing on well-learned routines. Second, that laboratory tasks lasting a couple of hours may not capture the multi-week grind of a real conference. Third, and most importantly for this article, that fatigue may express itself not in the quality of a deal but in other places entirely: in relationships, in emotional regulation, in the willingness to make a concession that is politically expensive at home, in health, and in whether the negotiator is still in the profession five years later. The gap between what a laboratory can measure and what a career actually costs is precisely where the sociology of the field becomes necessary. 2.4 Resilience: from trait to flexible self-regulation The word #resilience has been damaged by overuse. In institutional language it often means "the ability of staff to absorb whatever we do to them". It is worth recovering the scientific meaning, because the science has changed in ways that matter. The older popular model treated resilience as a trait, a stable personal quality that some people have and others lack. It generated questionnaires, and the questionnaires generated training courses. Bonanno (2021) calls the central problem the #resilience_paradox. Decades of research show that the most common response to a potentially traumatic event is a stable trajectory of healthy functioning. Yet when researchers try to predict who will follow that trajectory, the predictors turn out to be individually weak. No single variable, and often no combination of variables, identifies in advance who will be resilient. Popular resilience questionnaires simply ignore this by measuring a handful of presumed key traits. His proposed solution is that resilience is not a possession but a process, and specifically a process of flexible self-regulation. Bonanno and colleagues describe this as a flexibility sequence with three linked steps: reading the context accurately, having a wide repertoire of possible coping responses available, and monitoring feedback in order to change the response when it is not working (Bonanno, Chen, & Galatzer-Levy, 2023; Bonanno, Chen, Bagrodia, & Galatzer-Levy, 2024). The trait question, "is this person strong?", is replaced by the process question, "can this person read the situation, choose from a range of responses, and switch when the chosen response fails?" Bonanno (2024) has summarised the position in three axioms: resilience is common, there is no single resilience type, and resilience must be understood as an outcome that emerges from a process rather than as a stable characteristic. The relevance for diplomacy is direct and slightly uncomfortable. The diplomatic habitus rewards a narrow repertoire. It teaches composure, patience, indirect speech, and the suppression of visible emotion. These are genuine skills, and they work well in most diplomatic situations, which is exactly why they become automatic. But a narrow, automatic repertoire is the opposite of #regulatory_flexibility. A negotiator whose only tool is composure will keep using composure in a situation that calls for a walk-out, a display of anger, an honest admission of exhaustion, or a request for a recess. The very dispositions that make a diplomat effective in the ordinary case may reduce the flexibility that resilience actually requires in the extreme case. 2.5 Recovery: the neglected variable If fatigue is the accumulation of cost, then recovery is the process that clears it. Occupational health psychology has developed this into a mature field with a clear central concept. The core concept is psychological detachment, meaning genuine mental disengagement from work during non-work time. Not merely being away from the desk, but not thinking about the desk. Steed and colleagues (2021) synthesised the evidence on employee recovery and found detachment to be a central mechanism linking non-work time to wellbeing and performance outcomes. Sonnentag, Cheng, and Parker (2022) reviewed the field and pointed to what is sometimes called the #recovery_paradox: the people who most need to detach, because their jobs are most demanding, are precisely the people who find it hardest to do so. High demands produce rumination, and rumination blocks recovery. Karabinski and colleagues (2021) conducted a meta-analysis of interventions designed to improve detachment and found that such interventions can work, which is encouraging, but the effects depend heavily on design and on whether the surrounding job demands change at all. Diplomacy may be one of the worst possible occupations for detachment. The work follows the diplomat home because the home is often inside the mission compound, the social life is often official, the spouse's life is often organised around the posting, and the crisis does not respect the local time zone. The study by Gudmundsdottir and colleagues (2023) of foreign service spouses during the pandemic found that around a third reported moderate personal burnout, with a small further group reporting high or severe burnout, and that resilience was inversely related to burnout. Notably, knowing that evacuation support existed was associated with greater resilience, yet only about a quarter of respondents knew that it did. That single finding says a great deal: the resource existed, but the knowledge of it did not, and it is the knowledge, not the paperwork, that does the psychological work. The same study reported something that should give every training department pause. Attendance at stress management or resilience workshops was associated with higher burnout, which the authors reasonably interpret as a timing effect. People sign up for such workshops when they are already struggling. The implication is that #resilience_training delivered as a cure is close to useless, and possibly counterproductive, whereas the same content delivered as preparation, before the crisis, may work. 3. Conceptual Framework: The Habitus-Fatigue-Adaptation (HFA) Model The literature reviewed above suggests that neither sociology nor psychology alone can explain the behaviour of the tired diplomat. What follows is an attempt to join them. The Habitus-Fatigue-Adaptation model has three levels. Each level has its own logic, and the levels interact. 3.1 Level one: the dispositional level (habitus) At the first level sits the diplomat's habitus: the internalised dispositions acquired through class background, education, language training, diplomatic academy, and years of posting. This level supplies the negotiator's default settings. It determines what feels natural, what feels rude, what feels like giving up, and what feels like winning. Two properties of the habitus are load-bearing in this model. The first is #automaticity. Because habitus operates below conscious deliberation, it is cheap to run. A diplomat does not have to spend effort deciding to be polite. Politeness is already installed. This is a genuine resource, and it is the most plausible explanation for the compensation effects observed in the sleep deprivation studies. Well-learned routines survive fatigue better than effortful reasoning does. An exhausted diplomat can still be charming, because charm has become automatic. The second is #concealment. The same habitus that supplies free politeness also supplies a duty to hide strain. This is not free. It is precisely the kind of controlled, effortful suppression that Wiehler and colleagues associate with expensive prefrontal control. The habitus therefore both saves effort and spends it, and the balance between the two shifts as the hours pass. Proposition 1. The diplomatic habitus reduces the visible effects of fatigue on interpersonal performance while increasing the hidden physiological and emotional costs of fatigue. 3.2 Level two: the situational level (fatigue) At the second level sits the accumulation of strain within a given negotiation episode. Drawing on the cost and metabolic accounts rather than the discredited fuel tank metaphor, fatigue is modelled here not as an empty tank but as a rising price. As the episode continues, the price of exerting cognitive control rises, and the negotiator's system increasingly favours options that are immediate, simple, and low in effort. This produces a set of predictable behavioural drifts, which practitioners will recognise: A drift from #integrative_bargaining to distributive behaviour. Searching for joint gains requires holding several issues in mind at once. Splitting the difference does not. As control becomes expensive, negotiators simplify. A drift towards the status quo text. Reading a new proposal properly is expensive. Accepting a familiar one is cheap. A drift towards reactive rather than proactive moves. Fatigue shortens the horizon of planning. A drift towards immediate closure. This is the direct behavioural signature reported by Wiehler and colleagues: a preference shift towards short-delay, low-effort options. Proposition 2. As a negotiation episode lengthens, negotiators will show an increasing preference for immediate and low-effort agreement options, independent of the substantive merits of those options. Proposition 3. This drift will be weaker for tasks that are highly routinised within the negotiator's habitus, and stronger for tasks that require novel reasoning, unfamiliar technical content, or a language in which the negotiator is not fully fluent. Proposition 3 has an uncomfortable corollary. Negotiators working in a second or third language are effectively paying an extra tax on every hour of the negotiation, because language processing that is automatic for a native speaker remains partly controlled for them. #Linguistic_capital, in Bourdieu's sense, converts directly into stamina. 3.3 Level three: the adaptive level (resilience as flexibility) At the third level sits the response. Following Bonanno, resilience is treated here not as a trait but as a sequence: read the context, select from a repertoire, monitor and switch. The model predicts that diplomatic training produces excellent context sensitivity, since reading a room is the core diplomatic skill, but a restricted repertoire, since the field punishes most of the alternatives to composure. The result is a professional who is exceptionally good at knowing what is happening and unusually poor at doing anything other than absorbing it. Proposition 4. Diplomats will score high on context sensitivity and low on repertoire breadth relative to other high-demand professions, and this profile will predict elevated exhaustion despite high observed performance. Proposition 5. Institutional permission is the main determinant of repertoire breadth. Where an organisation explicitly authorises a behaviour, such as calling a recess, rotating the chair, or declaring that a delegation will not negotiate after midnight, individual negotiators will use it. Where it does not, they will not, regardless of how much resilience training they have received. Proposition 5 is the practical heart of the model. It relocates resilience from the individual to the institution. It also explains the otherwise puzzling finding that resilience workshops sometimes correlate with worse outcomes: a workshop grants no permission. It teaches an individual to endure a structure while leaving the structure untouched. 3.4 The interaction: fatigue as a distributed condition The three levels combine into a claim that is central to this article. Fatigue in diplomacy is not evenly distributed, and its distribution follows the distribution of capital. Consider two delegations at the same conference. The first has thirty officials, a dedicated legal team, a rotating night shift, an embassy in the host city, and a minister who arrives rested. The second has three officials, no legal support, no rotation, and a head of delegation who is also the ambassador, the note-taker, and the press officer. Both delegations sit at the same table. Both are formally sovereign equals. But the second delegation is negotiating with an additional, invisible opponent, and that opponent is time. This is not a metaphor. It is a resource asymmetry with measurable behavioural consequences, and it operates precisely through the mechanisms described in Proposition 2. Proposition 6. Delegation size, rotation capacity, and local institutional support are stronger predictors of endgame concession behaviour than the individual psychological characteristics of the negotiators involved. If Proposition 6 is even partly correct, then the design of multilateral processes, in particular the widespread practice of pushing decisions into a final sleepless night, functions as a structural advantage for large and wealthy states. #Procedural_fatigue becomes a form of soft coercion that no one needs to intend. 3.5 Summary of the framework Level Core concept Main mechanism Key question Dispositional Habitus Automatic routines lower effort costs, but composure norms raise hidden costs What has become second nature, and what must still be performed? Situational Fatigue as rising cost Control becomes expensive, shifting preferences towards immediate and low-effort options What is the price of thinking clearly at this hour? Adaptive Flexible self-regulation Context sensitivity, repertoire breadth, feedback monitoring What responses are actually available and permitted? 4. Method This article is a structured conceptual review, sometimes called an integrative or theory-building review. It does not report new empirical data. The purpose of this section is therefore to make the process of source selection and synthesis transparent, so that the reasoning can be checked. 4.1 Approach The review followed four steps. Step one: domain mapping. Three literatures were identified as relevant: the Bourdieusian sociology of diplomacy within International Relations; the experimental and neuroscientific psychology of cognitive fatigue, self-control, and sleep loss; and the occupational health psychology of recovery, burnout, and resilience. Step two: source selection. Within each domain, priority was given to recent peer-reviewed work, particularly review articles in high-quality outlets, meta-analyses, and empirical studies conducted with the relevant professional populations. Where a foundational concept was required, such as habitus or field, the concept is explained directly rather than through layers of secondary commentary. Step three: identification of tensions. Rather than presenting each literature as settled, the review deliberately sought out contested findings, in particular the replication problems surrounding ego depletion and the null result on sleep deprivation and joint negotiation outcomes. A synthesis that ignored these tensions would be more persuasive and less true. Step four: theory construction. The HFA model was built by identifying the points at which one literature answers a question that another literature cannot. The model was then expressed as six falsifiable propositions. 4.2 Boundaries and inclusion logic The review is bounded in three ways. It concerns professional state diplomats and negotiators working in bilateral and multilateral settings, including officials seconded to international organisations. It does not attempt to cover the psychology of heads of state, of humanitarian field workers, or of private commercial negotiators, although evidence from adjacent professions is used where the underlying mechanism is plausibly the same. It concerns fatigue arising from cognitive and emotional demand, together with sleep loss and insufficient recovery. It does not cover trauma resulting from violent incidents, which has its own substantial literature. It concerns #occupational_psychology rather than clinical diagnosis. Where clinical terms such as burnout appear, they are used as they are used in the cited studies. 4.3 A note on evidence quality Two honest cautions are necessary. First, the direct evidence base on diplomats is thin. Brooks and colleagues (2023) make this point plainly in their scoping review: relative to other high-risk occupational groups, the mental health of diplomatic personnel has been little studied, and the studies that exist are heterogeneous in method and population. Much of the psychological evidence used here is therefore transferred from adjacent groups, and transfer always carries risk. Second, some of the most widely repeated ideas in the practitioner literature on #decision_fatigue rest on a model that has not survived replication well. This article deliberately avoids building on that model, and readers who encounter confident claims about willpower being a fuel that runs out should treat them with caution. 5. Analysis: Five Sources of Strain in the Diplomatic Field This section applies the framework to the working life of a diplomat. It organises the analysis around five sources of strain. They are separated for clarity, but in practice they arrive together. 5.1 Temporal strain: the architecture of the endgame Multilateral negotiation has a characteristic shape. The early days are procedural. The middle days produce brackets, which are the disputed pieces of text. The final hours produce decisions. This shape is not accidental. Delegations withhold concessions until the last possible moment, because a concession offered early is spent, whereas a concession offered at the point of collapse buys something. The consequence is that some of the most consequential decisions in international politics are systematically taken at the point of the participants' lowest cognitive capacity. The psychological literature explains what this does. According to the metabolic and cost accounts, control becomes expensive after a long period of demand, and preference shifts towards options that are immediate and cheap (Wiehler et al., 2022). The endgame is thus a machine for producing agreement, though not necessarily for producing good agreement. It produces closure. Three qualifications keep this argument honest. First, closure is not worthless. In many cases a mediocre agreement is genuinely better than no agreement, and the exhaustion that drives parties to close is doing real diplomatic work. Practitioners know this and use it. #Exhaustion_as_tactic is an old technique, and the chairs of major conferences use scheduling as an instrument of pressure quite deliberately. Second, the experimental evidence does not straightforwardly show that tired negotiators reach worse deals. Halfmann and colleagues (2025) found no reduction in joint economic outcomes among sleep-deprived dyads, contrary both to their own theoretical predictions and to lay expectations. Any argument that all-night sessions produce bad treaties must reckon with this. Third, and reconciling the first two, the harm of temporal strain may not appear in the deal at all. It may appear in what the negotiator gives away that is not captured by a joint economic outcome: political capital at home, precision in the drafting, the relationship with a counterpart who will be needed next year, or the negotiator's own health. Analytic claim. The endgame does not necessarily produce worse agreements. It produces agreements whose costs are displaced away from the negotiating table and onto the negotiator, the domestic political system, and the implementation phase. 5.2 Cognitive strain: complexity, technicality, and language Modern negotiation is technical. A climate negotiator must hold in mind the legal architecture of a framework convention, the physical science of emissions accounting, and the domestic political limits of thirty other states. A trade negotiator must track tariff schedules, rules of origin, and dispute settlement mechanics. This is not diplomacy in the nineteenth century sense of persuasion between gentlemen. It is expert work under time pressure. Boothby, Cooney, and Schweitzer (2023) note in their review of negotiation research that the field has moved decisively towards complexity, finding that effects once described as simple and universal turn out to depend on context, relationship, and identity. This is a mature scientific position, and it maps well onto diplomatic reality, where nothing is ever a clean two-party distributive bargain. Three features of cognitive strain in diplomacy deserve attention. Working memory load. Integrative bargaining, the search for trades that make both sides better off, requires holding multiple issues and multiple priorities in mind at once. This is expensive. When control becomes costly, negotiators simplify, and simplification tends to collapse a multi-issue problem into a single-issue contest. The likely result is not a failure to agree, but a worse structure of agreement. The language tax. English dominates most multilateral settings. For a negotiator whose first language is not English, every hour of the negotiation carries an additional processing cost, and that cost compounds. This is a clear example of #cultural_capital converting into a material advantage. A native speaker at hour eighteen is tired. A non-native speaker at hour eighteen is tired and still translating. The drafting trap. Legal texts reward precision, but precision is exactly the capacity that fatigue erodes first. Ambiguity introduced at four in the morning becomes a decade of dispute. This is one of the few places where the cost of #negotiation_fatigue eventually becomes visible in the public record, although by then it is usually attributed to legal complexity rather than to sleep. 5.3 Emotional strain: the labour of composure The diplomat's face is a professional instrument. It must show interest without enthusiasm, disagreement without hostility, and disappointment without despair. This performance is continuous, and it is work. Organisational psychology distinguishes between surface acting, in which the felt emotion and the displayed emotion diverge, and deep acting, in which the person genuinely reshapes the felt emotion. The first is reliably associated with strain. The second is less so, but is harder to sustain. The diplomatic habitus is, in part, a machine for converting surface acting into deep acting. This is what training and long socialisation achieve. A junior officer performs calm and finds it exhausting. A senior officer with thirty years in the service is calm, and finds it cheap. Habitus therefore protects against #emotional_labour costs across a career, which is one plausible reason experienced negotiators cope better with brutal schedules. The protection is incomplete, and it fails in three situations. The first is moral strain. When a diplomat must defend a national position that she personally believes to be wrong, or must maintain warm relations with a counterpart whose government she considers responsible for grave harm, the gap between the felt and the displayed cannot be closed by socialisation. It must be held open by effort. This is among the most under-researched sources of strain in the profession. The second is emotion inside the negotiation itself. Boothby and colleagues (2023) summarise a substantial literature showing that expressed emotions such as anger function as strategic information, with effects that depend heavily on context, on whether the anger is directed at the offer or at the person, and on group membership. A negotiator therefore cannot simply switch emotion off. She must manage it, calibrate it, and sometimes perform it. That is a second layer of control demand stacked on top of the substantive one. The third is the recognition problem. Standfield (2022) shows that the same behaviour is not read the same way when performed by different bodies. A raised voice from a senior ambassador of a powerful state may be read as conviction. The same from a junior delegate of a small state may be read as loss of control. This means that #emotional_regulation demands are unequally distributed. Those who are least securely recognised as competent must work hardest to appear composed, and therefore pay the highest emotional labour cost for the same diplomatic output. Fatigue, again, follows the contours of power. 5.4 Institutional strain: mobility, family, and the impossibility of detachment Diplomatic careers are built on movement. A posting cycle of three to four years means a lifetime of packing, of children changing schools, of a partner setting aside a career, and of friendships that reset every few years. The evidence base here, while thin, is consistent. Brooks and colleagues (2023) found that the literature on diplomatic personnel points to repeated relocation, difficulties with career satisfaction on repatriation, and an interaction between organisational identification and burnout. Their review also reports that younger cohorts appear less willing than their predecessors to tolerate the traditional costs of the career, and more likely to leave if their expectations are not met. That is a warning signal for foreign ministries: the implicit bargain that once held the profession together, in which the officer accepted personal cost in exchange for prestige and a guaranteed career, is weakening. The study of foreign service spouses by Gudmundsdottir and colleagues (2023) adds the family dimension. Around a third of their 421 respondents reported moderate personal burnout during the pandemic, with a smaller group reporting high or severe burnout. Resilience was inversely correlated with burnout, as expected. More useful is what predicted resilience: knowing that evacuation support existed. Only about a quarter of respondents knew that it did. Support that is not known does not support. The deeper point concerns #psychological_detachment. Sonnentag, Cheng, and Parker (2022) identify detachment as the core recovery experience, and describe the recovery paradox: those with the highest demands find it hardest to detach. Diplomatic life is arguably the extreme case. The residence may be inside the compound. The social calendar is official. The partner's identity is defined by the posting. The crisis arrives at a time set by another hemisphere. There is, in the strict sense, no time that is reliably not work. Steed and colleagues (2021) and Karabinski and colleagues (2021) both indicate that detachment matters and that interventions can improve it. Both also indicate that effects depend on the surrounding demands. An intervention that teaches an officer to switch off her phone, inside an institution that expects her to answer it, is a trap dressed as a benefit. Analytic claim. In diplomacy, the recovery deficit is structural rather than behavioural. It cannot be solved at the level of individual habits, because the boundary between work and life has been dissolved by the design of the career itself. 5.5 Structural strain: fatigue as an unequal resource The final source of strain is the one that habitus theory is best equipped to expose. It is the fact that endurance is not only psychological. It is purchased. Large delegations can rotate. Small ones cannot. Wealthy ministries can send legal advisers, scientific advisers, and interpreters. Poorer ones send one person who must be all three. States with an embassy in the host city can send their negotiators somewhere quiet to sleep. States without one house their delegations far away and add two hours of travel to a twenty hour day. Every one of these differences translates, through the mechanisms described in Section 3, into a difference in endgame behaviour. The delegate who has slept is not morally superior to the delegate who has not. She is simply operating a control system that is cheaper to run, and she will therefore find it easier to hold a line at three in the morning. The theoretical consequence is significant. It means that #procedural_design is a form of power. The decision to run a conference to an artificial deadline, to schedule the crucial session at night, to circulate a new text at midnight for adoption at dawn, is not a neutral administrative choice. It systematically advantages the delegations with the deepest benches. In Bourdieu's terms, it is a mechanism through which the possession of capital converts into symbolic victory while appearing to be nothing more than the ordinary hardship of the profession, borne equally by all. The most effective forms of #power are the ones that look like the weather. 6. Discussion 6.1 Implications for theory The first implication is for International Relations. The practice turn has taught the discipline to look at what diplomats do rather than only at what states declare. But practice has been studied mostly as competence, style, and status. This article argues that practice also has a metabolic cost, and that the cost is not incidental to the politics. It shapes who concedes and when. A practice theory that cannot say anything about the third hour past midnight is incomplete. The second implication is for the psychology of negotiation. Most of what is known experimentally about negotiation comes from short laboratory tasks with student participants and a clean payoff structure. Boothby, Cooney, and Schweitzer (2023) explicitly call for more field research and for scholarship that offers practical guidance. Diplomacy is an extraordinary field site: high stakes, long duration, repeated interaction, professional participants, and enormous variation in institutional support. It is also, admittedly, very difficult to access. The null result reported by Halfmann and colleagues (2025) is a good illustration of why access matters. A two-hour laboratory bargaining task after one bad night may simply not be the same phenomenon as the fourth week of a conference. The third implication is for resilience research. Bonanno's reframing of #resilience as flexible self-regulation invites a specific empirical question in professional settings: what is the size of a professional's response repertoire, and who decides what goes into it? The diplomatic case suggests that repertoire size may be largely institutional rather than personal. If so, resilience research in organisations should measure permissions, not just personalities. 6.2 Implications for foreign ministries Six practical implications follow from the framework. They are ordered from the most structural to the most individual, deliberately, because the structural ones matter more. One: redesign the endgame. If decisions taken at four in the morning are systematically biased towards immediate, low-effort options, then any organisation that cares about the quality of those decisions has a reason to change when they are taken. Rules that suspend formal adoption of new text after a stated hour, or that require a minimum interval between the circulation of a text and its adoption, are cheap to write and would materially change the psychology of the room. They would also, not coincidentally, reduce the structural advantage held by large delegations. Two: make rotation a resource question, not a virtue question. Delegations without depth cannot rotate. International organisations and donor states that genuinely wish to support the participation of smaller states could treat delegation capacity, including the ability to staff a night shift, as a legitimate object of support, in the same way that travel funding already is. Three: grant explicit permission. Following Proposition 5, the single most effective intervention available to a ministry is probably the cheapest: to state clearly, in writing and from the top of the hierarchy, that certain behaviours are permitted. That a delegate may request a recess. That a head of mission may hand over. That declining an evening reception is not a career risk. Permission expands the repertoire. Training without permission does not. Four: make support visible. The finding by Gudmundsdottir and colleagues (2023) that knowledge of evacuation support predicted resilience, while only a minority knew of it, generalises. A benefit that staff do not know about produces no psychological protection whatsoever. #Wellbeing_policy should be audited not by what exists but by what is known to exist. Five: move #resilience_training upstream. The same study found workshop attendance associated with higher burnout, most plausibly because people attend when already in trouble. Preparation is not the same as repair. Training that is delivered before the first hardship posting, and that is treated as part of professional formation rather than as a remedial service, is more likely to help and much less likely to stigmatise. Six: teach flexibility, not toughness. If resilience is a sequence of reading, selecting, and switching, then it can be taught as a skill. Diplomats already excel at the reading step. What they need is a wider set of permitted responses and honest practice in switching between them, including the practice of saying, out loud, that they are too tired to take a decision safely. That sentence is currently close to unsayable in the profession. Making it sayable would be a more significant reform than any wellbeing app. 6.3 A note on what resilience should not mean There is a version of the resilience agenda that serves institutions rather than people. In that version, resilience means the capacity of staff to keep absorbing a workload that the organisation has no intention of reducing. It individualises a structural problem, and it turns exhaustion into a personal failing. The scientific literature does not support that version. #Resilience, in the sense used by Bonanno and colleagues, is about flexible adaptation in an environment, not about the passive endurance of an environment. If a ministry adopts the word but not the meaning, it will get worse outcomes and better public relations. 6.4 The paradox at the centre The argument of this article ends in a paradox worth stating plainly. The diplomatic habitus is a magnificent adaptive achievement. It allows a person to remain courteous under insult, precise under pressure, and functional at hours when most professionals would be useless. It is, in the language of the psychology reviewed here, a vast store of automatised routines that keep working when controlled processing has become expensive. This is why diplomats do not collapse. It is why they can, as the sleep studies suggest, compensate. But the same habitus enforces the concealment of strain, narrows the repertoire of permitted responses, and defines endurance as merit. It therefore hides the very condition it helps people to survive, and in hiding it, prevents institutions from addressing it. The diplomat's greatest professional asset is also the mechanism by which the profession fails to see its own #occupational_health problem. This is not a moral criticism of diplomats. It is a description of how fields work. Bourdieu's point was always that the rules which feel most natural are the ones doing the most political work. 7. Limitations This article has clear limits, and stating them is part of the argument. It is conceptual. The six propositions are derived from existing literature, not tested. They are offered as hypotheses, and several of them may be wrong. It relies on transfer. The core psychological evidence comes from laboratory studies, from general working populations, and from adjacent professions. Diplomats may differ. Indeed, the framework itself predicts that they differ, because their habitus is unusually well adapted to exactly the conditions being studied. That prediction cuts both ways: it may mean that the fatigue effects described here are weaker among senior diplomats than the general literature would suggest. It treats contested evidence as contested, which weakens the rhetorical force of the argument. The ego depletion literature is in poor shape. The one direct experimental test of sleep deprivation in negotiation produced a null result on deal quality. An honest synthesis has to live with that, and readers should be sceptical of any account of #diplomatic_fatigue, including this one, that claims more certainty than the evidence allows. It is also, unavoidably, shaped by the availability of research. Almost all of the occupational evidence cited here comes from European and North American services and from their spouses. The experience of diplomats from small states, from post-colonial services, and from ministries without any occupational health provision at all is precisely the experience that the structural argument in Section 5.5 identifies as most important, and it is almost entirely unstudied. 8. Conclusion and Research Agenda Diplomacy is done by bodies. The body gets tired, and when it does, the mind that has to weigh a hundred pages of bracketed text at four in the morning begins to prefer whatever is quickest. This article has argued that this is not a footnote to international politics but a mechanism within it. Three claims summarise the argument. First, the diplomatic habitus both protects against fatigue and hides it. It automates the behaviours that would otherwise cost effort, which is why diplomats appear composed at hours when they should not be. But it also makes the display of strain professionally costly, which is why the profession has so little honest information about its own condition. Second, fatigue in negotiation should be modelled as a rising cost of control rather than as a fuel tank running dry. The consequence is a predictable drift towards immediate, simple, low-effort options as sessions lengthen. Whether that drift produces objectively worse agreements is genuinely uncertain, and the best available experiment says it may not. But the drift itself is well supported, and its costs may simply be displaced elsewhere. Third, resilience is not a personal quality that some diplomats have and others lack. It is a process of flexible adaptation, and the range of adaptations available to any individual is set mainly by the institution. This relocates the wellbeing problem from the officer to the ministry, and from the ministry to the design of international processes themselves. A serious research agenda follows. Field measurement. Diary and experience-sampling studies with delegations during actual conferences, measuring sleep, subjective fatigue, and behavioural markers of concession. This is difficult but not impossible, and it is the single greatest gap in the literature. Delegation asymmetry. Comparative study of endgame concession behaviour as a function of delegation size, rotation capacity, and local support, directly testing Proposition 6. The language tax. Experimental work on negotiation performance under fatigue in a first versus a second language. This would test Proposition 3 and would speak directly to questions of equity in #multilateral_negotiation. Repertoire and permission. Survey and interview work mapping which coping behaviours diplomats believe are permitted, comparing across ministries, and relating repertoire breadth to burnout. This would test Propositions 4 and 5. The unstudied majority. Occupational health research on diplomatic services outside Europe and North America, without which the structural argument advanced here remains a hypothesis about global inequality rather than a demonstration of it. The negotiating room at four in the morning is one of the few places where the machinery of the international system becomes visible on a human scale. Someone is about to accept something. Understanding why is not a question of psychology or of sociology alone. It is a question of what the profession has made of the person, what the hour has made of the mind, and what the institution has permitted either of them to do about it. References Bonanno, G. A. (2021). The end of trauma: How the new science of resilience is changing how we think about PTSD. New York: Basic Books. Bonanno, G. A. (2021). The resilience paradox. European Journal of Psychotraumatology, 12(1), 1942642. doi:10.1080/20008198.2021.1942642 Bonanno, G. A. (2024). The three axioms of resilience. Journal of Traumatic Stress, 37(3), 355-360. doi:10.1002/jts.23071 Bonanno, G. A., Chen, S., Bagrodia, R., and Galatzer-Levy, I. R. (2024). Resilience and disaster: Flexible adaptation in the face of uncertain threat. Annual Review of Psychology, 75, 573-599. doi:10.1146/annurev-psych-011123-024224 Bonanno, G. A., Chen, S., and Galatzer-Levy, I. R. (2023). Resilience to potential trauma and adversity through regulatory flexibility. Nature Reviews Psychology, 2(11), 663-675. doi:10.1038/s44159-023-00233-5 Boothby, E. J., Cooney, G., and Schweitzer, M. E. (2023). Embracing complexity: A review of negotiation research. Annual Review of Psychology, 74, 299-332. doi:10.1146/annurev-psych-033020-014116 Brooks, S. K., Patel, S. S., and Greenberg, N. (2023). Mental health of diplomatic personnel: A scoping review. Occupational Medicine, 73(3), 155-160. doi:10.1093/occmed/kqad032 Clemencon, R. (2023). 30 years of international climate negotiations: Are they still our best hope? The Journal of Environment and Development, 32(2), 103-127. doi:10.1177/10704965231163908 Elfenbein, H. A. (2021). Individual differences in negotiation: A relational process model. Organizational Psychology Review, 11(1), 73-93. doi:10.1177/2041386620986615 Elfenbein, H. A. (2023). Emotion in organizations: Theory and research. Annual Review of Psychology, 74, 489-517. doi:10.1146/annurev-psych-032420-030312 Englert, C., and Bertrams, A. (2021). Again, no evidence for or against the existence of ego depletion: Opinion on "A multi-site preregistered paradigmatic test of the ego depletion effect". Frontiers in Human Neuroscience, 15, 658890. doi:10.3389/fnhum.2021.658890 Gudmundsdottir, S., Larsen, K., Woods Nelson, M., Devine Mildorf, J., and Molek-Winiarska, D. (2023). Burnout and resilience in foreign service spouses during the pandemic, and the role of organizational support. Sustainability, 15(3), 2435. doi:10.3390/su15032435 Halfmann, E., Huffmeier, J., Faber, N. S., and Hausser, J. A. (2025). Sleep deprivation and negotiation. Collabra: Psychology, 11(1), 128507. doi:10.1525/collabra.128507 Hausser, J. A., Halfmann, E., and Huffmeier, J. (2022). Negotiating through the night: How sleep deprivation can affect negotiation processes and outcomes. Negotiation and Conflict Management Research, 16(2), 189-210. doi:10.34891/2022.575 Huju, K. (2023). The cosmopolitan standard of civilization: A reflexive sociology of elite belonging among Indian diplomats. European Journal of International Relations, 29(3), 698-722. doi:10.1177/13540661231170731 Karabinski, T., Haun, V. C., Nubold, A., Wendsche, J., and Wegge, J. (2021). Interventions for improving psychological detachment from work: A meta-analysis. Journal of Occupational Health Psychology, 26(3), 224-242. doi:10.1037/ocp0000280 Miyake, A., and Carruth, N. P. (2023). On the robustness and replicability of the moderating influence of willpower mindset on the ego-depletion effect: Existing evidence is weak at best. Frontiers in Psychology, 14, 1208299. doi:10.3389/fpsyg.2023.1208299 Nair, D. (2024). Using Bourdieu's habitus in International Relations. International Studies Quarterly, 68(2), sqae007. doi:10.1093/isq/sqae007 Sonnentag, S., Cheng, B. H., and Parker, S. L. (2022). Recovery from work: Advancing the field toward the future. Annual Review of Organizational Psychology and Organizational Behavior, 9, 33-60. doi:10.1146/annurev-orgpsych-012420-091355 Standfield, C. (2022). Who gets to be a virtuoso? Diplomatic competence through an intersectional lens. The Hague Journal of Diplomacy, 17(3), 371-401. doi:10.1163/1871191X-bja10112 Steed, L. B., Swider, B. W., Keem, S., and Liu, J. T. (2021). Leaving work at work: A meta-analysis on employee recovery from work. Journal of Management, 47(4), 867-897. doi:10.1177/0149206319864153 Wiehler, A., Branzoli, F., Adanyeguh, I., Mochel, F., and Pessiglione, M. (2022). A neuro-metabolic account of why daylong cognitive work alters the control of economic decisions. Current Biology, 32(16), 3564-3575. doi:10.1016/j.cub.2022.07.010 #Diplomatic_Habitus #Negotiation_Fatigue #Resilience_In_Diplomacy #International_Relations_Theory #Bourdieu_And_IR #Cognitive_Fatigue #Multilateral_Diplomacy #Foreign_Service_Wellbeing #Psychology_Of_Negotiation #Flexible_Self_Regulation #Diplomatic_Practice_Turn #Burnout_In_Foreign_Ministries #Sleep_And_Decision_Making #Symbolic_Power_In_Diplomacy #Global_Governance_Psychology

Institutional co-investments bring together pension funds, sovereign wealth funds, insurance companies, family offices, development finance institutions and private equity sponsors around a single asset or platform. The sums are large, the information is incomplete, the timelines are short, and the outcome depends on a negotiated agreement rather than on a market price. This is exactly the setting in which #psychological_biases stop being an academic curiosity and start moving real capital. This article reviews and synthesises recent behavioural finance, judgment and decision-making, and negotiation research in order to build an integrated framework of how bias operates inside #strategic_financial_negotiations for #institutional_co_investments. Using an integrative conceptual review method, the paper maps ten well-documented biases (#anchoring, #overconfidence, #loss_aversion, #framing, #confirmation_bias, #herding, #escalation_of_commitment, the #fixed_pie_bias, the #self_serving_bias and #status_quo_bias) onto a five stage negotiation cycle that runs from origination to post-closing governance. It argues that co-investment is unusually exposed to bias because three amplifiers act at the same time: information asymmetry between sponsor and co-investor, compressed decision windows created by allocation scarcity, and group dynamics inside investment committees. The paper also distinguishes systematic bias from #noise, or unwanted variability in judgments that should agree, and shows why the two require different remedies. It closes with a practical debiasing architecture built around structured protocols, independent estimates, pre-commitment to walk-away terms, adversarial review and outcome-independent evaluation of decision quality. The contribution is conceptual and integrative rather than empirical, and the framework is offered as a testable agenda for future research. Keywords: behavioural finance; negotiation; co-investment; cognitive bias; institutional investors; decision-making; private capital; debiasing 1. Introduction A large infrastructure fund invites three pension plans to co-invest alongside it in a toll road platform. The sponsor circulates a teaser, sets a two week window for indications of interest, and names a headline valuation. Within that window each pension plan must form a view on traffic forecasts stretching twenty five years into the future, agree governance rights, and decide how much capital to commit. Nobody in the room has full information. Everybody knows that if they hesitate, the allocation moves to somebody else. By the time the term sheet is signed, most of the important judgments have already been made, and most of them were made quickly. This is the everyday texture of #strategic_financial_negotiations in institutional markets, and it is a long way from the frictionless rational agent of classical finance. Standard theory assumes that investors update beliefs correctly, weigh probabilities consistently, and negotiate towards efficient outcomes. Behavioural research has spent five decades documenting that human judgment departs from that ideal in patterned, predictable ways, and that professional training and financial incentives reduce these departures far less than most practitioners expect (Kahneman, Sibony and Sunstein, 2021; Hachicha, Argoubi and Guesmi, 2024). Two literatures speak to the problem, but they rarely speak to each other. The first is #behavioral_finance, which has produced a rich catalogue of biases affecting investment decisions and has recently been consolidated through systematic and bibliometric reviews (Jain, Walia, Singh and Jain, 2021; Maheshwari, Samantaray and Jena, 2023). The second is negotiation research, which studies how two or more parties with partly conflicting interests reach agreement, and which has its own catalogue of distortions such as the #fixed_pie_bias and the #self_serving_bias. Behavioural finance tends to model a single decision maker facing a market. Negotiation research tends to model two individuals facing each other. #institutional_co_investments sit awkwardly between the two. They involve a market price that is not observable, a counterparty who holds most of the information, a committee that must approve the outcome, and a relationship that will continue for a decade after the deal closes. The gap matters because the stakes are unusually asymmetric. In a liquid public market a biased judgment produces a small loss and can be corrected tomorrow. In a co-investment the position is illiquid, concentrated, held for years, and priced through a negotiation rather than by a market. A single distorted #valuation judgment, or a single failure to walk away, can absorb a meaningful share of an institution's annual risk budget. Evidence from corporate acquisitions, which share the same structural features of negotiated price, scarce information and reputational stakes, indicates that behavioural distortions at the top of the organisation translate directly into worse deal outcomes (Twardawski and Kind, 2023; Brahma, Boateng and Ahmad, 2023). 1.1 Research questions This article addresses four questions. First, which #psychological_biases have the strongest theoretical and empirical claim to influence decisions in negotiated institutional transactions? Second, how do those biases distribute across the stages of a co-investment negotiation, from origination through #due_diligence, price setting, documentation and post-closing #governance? Third, what features of the co-investment setting amplify or dampen bias relative to other financial decisions? Fourth, which #debiasing interventions have credible support, and how can they be embedded in institutional process rather than left to individual willpower? 1.2 Contribution The paper makes three contributions. It integrates two literatures that have developed separately, producing a stage by stage map of bias in negotiated transactions. It identifies three amplifiers specific to co-investment, namely #information_asymmetry, allocation scarcity and committee dynamics, and explains why they interact rather than simply add. It separates systematic bias from #noise and argues that institutions currently over-invest in awareness training, which addresses neither problem well, and under-invest in process design, which addresses both. The paper is deliberately conceptual. It does not report new empirical data. It synthesises published findings, most of them from the last five years, and converts them into propositions that can be tested. Readers looking for a definitive causal estimate of how much money bias costs a co-investment programme will not find it here, because that number has not been credibly estimated in the public literature. What they will find is a framework for asking the question properly. 1.3 Structure Section 2 reviews the theoretical foundations and the empirical literature. Section 3 develops the conceptual framework and a set of propositions. Section 4 sets out the review method. Section 5 analyses each bias in the co-investment context and maps it to the negotiation cycle. Section 6 discusses amplifiers, moderators and the distinction between bias and noise. Section 7 sets out practical implications for institutions and students. Section 8 states limitations and a future research agenda. Section 9 concludes. 2. Theoretical Background and Literature Review 2.1 From rational choice to bounded rationality Classical finance rests on the assumption that decision makers maximise expected utility using all available information. The assumption was never meant as a description of psychology. It was meant as a workable approximation, and for many purposes it still is. The difficulty is that it approximates worst precisely where co-investment lives, namely in decisions that are rare, complex, hard to reverse, and made under time pressure with incomplete data. The alternative tradition begins with #bounded_rationality, the recognition that human beings have limited attention, limited memory and limited computational capacity, and therefore rely on simplifying rules. Those rules, or heuristics, are usually efficient. They become #psychological_biases when they produce errors that are systematic rather than random, meaning that the errors point in a consistent direction and do not cancel out across repeated decisions. The modern statement of this idea is the #dual_process account of cognition. One mode of thinking is fast, automatic, associative and effortless. The other is slow, controlled, rule based and effortful. The fast mode is not a defect. It is what allows an experienced investor to sense within minutes that a management team is evasive. The problem is that the fast mode is also confident, silent about its own limits, and difficult to switch off. Under time pressure, which is the normal condition of a competitive co-investment process, the fast mode dominates and the slow mode is reduced to producing a justification after the fact. 2.2 Prospect theory and reference dependence #prospect_theory is the second pillar. Its core insight is that people evaluate outcomes not in terms of final wealth states but in terms of gains and losses relative to a #reference_point, that losses loom larger than equivalent gains, and that people are risk averse in the domain of gains but risk seeking in the domain of losses. Diminishing sensitivity means that the difference between a loss of ten million and a loss of twenty million feels smaller than the difference between zero and ten million. Three consequences follow for negotiation. First, whichever number becomes the reference point will shape the whole discussion, which is why the opening figure in a term sheet is so powerful. Second, a party that has framed itself as protecting an existing position will resist concessions more stubbornly than a party that has framed itself as capturing an opportunity, even when the economics are identical. Third, a party that believes it is already in the loss domain, for example a fund manager whose vintage is underperforming, will accept more risk to get back to par, which is a mechanism through which #loss_aversion produces risk seeking rather than caution. Recent empirical work continues to support reference dependent behaviour in professional as well as retail settings, and it links prospect theoretic reasoning to herding and heuristic shortcuts in emerging and developed markets alike (Goyal, Gupta and Yadav, 2023; Jain, Walia, Singla, Singh, Sood and Grima, 2023). 2.3 The catalogue of biases in financial decision-making The behavioural finance literature has grown quickly and has recently been consolidated. Bibliometric and systematic reviews show a field that expanded sharply after 2010 and that has moved from documenting individual anomalies to examining their interaction, measurement and management (Jain, Walia, Singh and Jain, 2021; Maheshwari, Samantaray and Jena, 2023; Hachicha, Argoubi and Guesmi, 2024). Scale development work has attempted to measure biases as latent constructs rather than infer them from trading records, which is a necessary step if the field is to move beyond correlational studies (Jain, Walia, Kaur and Singh, 2022). Several findings from this recent literature are directly relevant here. #overconfidence remains the most consistently documented distortion among finance professionals. It appears in three forms that are often confused: overestimation of one's own ability or of an asset's prospects, overplacement relative to peers, and overprecision, meaning excessive narrowness of confidence intervals. Overprecision is the most dangerous in #valuation work because it produces forecast ranges that are too tight, which in turn makes downside scenarios look implausible. Research links overconfidence to worse investment performance and to greater risk taking, with risk propensity acting as a channel rather than a confound (Ul Abdin, Qureshi, Iqbal and Sultana, 2022; Naveed and Taib, 2021). Recognition based heuristics, meaning the tendency to prefer the familiar name, the familiar sponsor or the familiar geography, have been shown to shape both the decision and the eventual performance (Ahmad, Wu and Abbass, 2023). In co-investment this appears as a preference for sponsors with whom the institution has an existing relationship, which is not irrational, but which becomes a bias when relationship familiarity substitutes for underwriting. Heuristic shortcuts operate partly through #risk_perception. When investors rely on availability or representativeness, they do not simply misjudge probabilities directly. They form a distorted sense of how risky the situation is, and then act rationally on that distorted sense (Jain and colleagues, 2023; Kumar, Kasilingam and Rajamohan, 2024). This matters for intervention design, because it means that correcting the perception may be more effective than lecturing about the bias. Behavioural factors have also been shown to affect not just decisions but realised performance in market settings, which addresses the common objection that biases are interesting in the laboratory and harmless in practice (Vukovic and Pivac, 2024). Financial literacy and professional experience reduce some biases and leave others untouched, and in certain cases experience increases #overconfidence rather than reducing it (Suresh, 2024). 2.4 Bias at the top of the organisation Because co-investment decisions are made by senior people and ratified by committees, the literature on managerial and board level bias is directly applicable. Studies of #mergers_and_acquisitions provide the closest available analogue to a negotiated, high-stakes, irreversible institutional transaction. Overconfident chief executives acquire more, pay more, and destroy more value on average. The effect is not confined to the individual at the top. Board level overconfidence, measured as an aggregate property of the group rather than of any one director, has been shown to influence acquisition behaviour and outcomes, which suggests that the usual remedy of adding oversight does not work if the oversight body shares the bias (Twardawski and Kind, 2023; Brahma, Boateng and Ahmad, 2023). Overconfidence also shapes financing choices, including the decision to issue debt and the terms accepted (Ge, Jamil and Yu, 2024), and it interacts with firm circumstances so that its effects differ sharply between healthy and distressed organisations (Kowalzick, Ahrens, Lauterbach and Tang, 2024). Importantly, the literature does not treat overconfidence as purely destructive. Confidence supports decisiveness, commitment and the willingness to act on incomplete information, all of which a deal team needs. Recent reviews frame managerial overconfidence as both a promoter of and an obstacle to organisational resilience, depending on context and on the governance structures that surround it (Kunz and Sonnenholzner, 2023). It also encourages entrepreneurial behaviour in emerging market firms, which is not a bad thing (Gu, 2023). The useful question is therefore not how to eliminate confidence but how to calibrate it. 2.5 Bias in negotiation research Negotiation research contributes a second set of distortions that behavioural finance largely ignores. The #fixed_pie_bias is the belief that the parties' interests are directly opposed, so that any gain for one side is a loss for the other. In practice most negotiations contain issues that the parties value differently, which creates room for trades that make both sides better off. Negotiators who assume a fixed pie stop looking for those trades and settle for distributive bargaining over a single number, usually price. The #self_serving_bias is the tendency to interpret ambiguous evidence in a way that favours one's own interest, while sincerely believing the interpretation to be objective. It is the reason two experienced professionals can read the same due diligence report and reach opposite conclusions about what is fair, with neither of them lying. #anchoring in negotiation is the disproportionate influence of the first number placed on the table. It survives warning, it survives expertise, it survives incentives to be accurate, and it operates even when the anchor is obviously arbitrary. It is arguably the single most robust finding in the negotiation literature. The #toughness_bias and the #reactive_devaluation effect describe the tendency to assume the other side is more adversarial than it is, and to value a concession less simply because it came from the counterparty. Negotiation research also emphasises structural constructs that behavioural finance rarely uses. The #BATNA, or best alternative to a negotiated agreement, determines the real bargaining power of each side. The #reservation_price is the point beyond which walking away beats agreeing. The #ZOPA is the range of deals that both sides prefer to no deal. Bias does its damage largely by corrupting the estimation of these three quantities. An overconfident negotiator overestimates the BATNA. A negotiator in the grip of deal momentum quietly slides the reservation price. A negotiator with a fixed pie mindset never discovers how wide the ZOPA really is. 2.6 Institutional co-investment as a decision environment A co-investment is an investment made directly into an asset alongside a sponsor, usually a private equity, infrastructure or real assets manager, rather than exclusively through the sponsor's commingled fund. The economics are attractive for the investor, since fees are typically reduced or waived on the co-invested portion, and control and information rights can be negotiated. The economics are attractive for the sponsor, since co-investment expands available capital for larger deals, preserves fund concentration limits, and strengthens relationships with #limited_partners. Five features define the decision environment. Capital scale. Individual tickets frequently run into the hundreds of millions, which means a single decision can dominate the annual risk budget of a mid-sized allocator. Illiquidity and irreversibility. There is no straightforward exit. A mistake compounds for years rather than being corrected in the next trading session. Structural information asymmetry. The sponsor has usually spent months on the asset. The co-investor receives a curated data room and a short window. The imbalance is not a market failure to be corrected but a permanent feature of the relationship. Time compression. Allocations are scarce and are often offered on a first come basis. Deadlines are frequently set by the sponsor, who benefits from speed. Time pressure is the single most reliable way to push a decision maker from deliberate into intuitive thinking. Relational and reputational stakes. The counterparty is not a stranger. Declining a deal may affect access to the next fund. Accepting a bad deal may be less costly to an individual's career than being seen to miss a good one, which is a straightforward #principal_agent problem layered on top of a cognitive one. Recent work on institutional investors also documents that cultural and contextual factors shape how these professionals process information, meaning that bias profiles are not uniform across markets or across teams. Cross-cultural evidence from fund managers indicates systematic differences in how investment information is weighted and how confidently conclusions are drawn. 2.7 The research gap Three gaps emerge from the review. The behavioural finance literature is overwhelmingly focused on individual and retail investors in liquid public markets, where the decision is a trade rather than a negotiation. Reviews of the field acknowledge that institutional investors are comparatively understudied, and that the studies which do exist concentrate on herding (Maheshwari and colleagues, 2023). The negotiation literature is strong on dyadic bargaining and weak on the organisational machinery that surrounds a large financial transaction, in particular investment committees, delegated authority and staged approvals. Neither literature has produced an integrated model of how bias moves through the stages of a negotiated institutional transaction. That is the gap this article addresses. 3. Conceptual Framework and Propositions 3.1 The five stage co-investment negotiation cycle The framework treats a co-investment as a sequence rather than a single decision. Bias enters at different points and, crucially, is carried forward. An error committed at stage one is rarely corrected at stage four. It is defended. Stage 1: Origination and screening. The sponsor circulates the opportunity. The co-investor decides whether to engage. Judgments are made in hours or days, with minimal information, and they set the frame for everything that follows. Stage 2: Due diligence and valuation. The co-investor builds or reviews a model, tests assumptions, and forms a view of intrinsic value and a walk-away price. This is where the analytical work concentrates and, paradoxically, where bias does the most damage because the outputs look objective. Stage 3: Price and terms negotiation. The parties bargain over valuation, governance rights, fees, information rights, exit provisions and downside protections. This is the stage that negotiation research addresses directly. Stage 4: Documentation and closing. The agreed economics are converted into contract. Attention drops, fatigue rises, and concessions are made on issues that seem technical but are not. Stage 5: Post-closing monitoring and follow-on decisions. The investor holds the asset, receives reports, and periodically faces decisions about further capital, restructuring or exit. This is where #escalation_of_commitment lives. 3.2 Three amplifiers The framework proposes that co-investment is not simply another arena in which known biases appear. It is an arena in which three features systematically amplify them. Amplifier one: information asymmetry. When information is scarce, the mind fills gaps with whatever is available, coherent and easy to retrieve. #confirmation_bias and the #availability_heuristic therefore have more room to operate than they would in a data rich environment. The co-investor is not choosing between two well specified models. The co-investor is choosing between a story provided by the sponsor and no story at all. Amplifier two: time compression. Deliberate reasoning requires time. Compressing the window does not merely reduce the amount of analysis. It changes the type of reasoning employed, shifting weight from the slow, checking, doubting mode to the fast, coherent, confident mode. Every bias that depends on intuitive processing becomes stronger under a deadline. Amplifier three: group dynamics. Investment committees are supposed to correct individual error. Frequently they compound it. A committee that hears a proposal from a respected #deal_champion, that lacks independent access to the underlying data, that speaks in sequence rather than in writing, and that knows the sponsor relationship is valuable, will tend towards #groupthink and towards premature convergence. Board level evidence from acquisitions supports the concern that collective bodies can share rather than cancel bias (Twardawski and Kind, 2023). These amplifiers interact. Time compression makes committees more deferential, because there is no time to dissent. Information asymmetry makes the deal champion's narrative harder to challenge, because nobody else has independent facts. The combination is more damaging than the sum. 3.3 Propositions From the framework, the article derives eight propositions. They are offered as testable claims, not as established findings. P1. The sponsor's initial valuation acts as an #anchor that measurably shifts the co-investor's independently derived valuation, and the shift is larger when the co-investor's own prior analysis is less developed. P2. overconfidence in co-investment appears predominantly as overprecision, showing up as unjustifiably narrow forecast ranges and thin downside cases, rather than as overt optimism about the central case. P3. Time compression in the allocation process increases the influence of intuitive processing and therefore increases the magnitude of anchoring, availability and confirmation effects. P4. Co-investors who have publicly committed internal resources to a transaction show greater #escalation_of_commitment at subsequent decision points than those who have not, independently of the transaction's fundamentals. P5. Investment committees that collect written, independent assessments before discussion produce less dispersed and less biased decisions than committees that deliberate first and record afterwards. P6. Pre-commitment to a written #reservation_price before the negotiation begins reduces the incidence of walk away failure, and the effect is larger when the reservation price is disclosed to a party who is not involved in the deal. P7. fixed pie bias causes co-investors to concentrate negotiation effort on headline valuation while conceding governance, information and exit rights that carry substantial economic value. P8. A material share of the variation in co-investment decisions is #noise, meaning unwanted variability between equally qualified professionals evaluating the same opportunity, rather than shared directional bias, and therefore will not respond to bias awareness training. 4. Methodology 4.1 Design This paper uses an integrative conceptual review. The approach is appropriate when the goal is to synthesise findings across separate literatures and generate a framework, rather than to estimate an effect size. It follows three steps: identification of relevant literature; thematic coding of biases and mechanisms; and synthesis into a stage based framework with testable propositions. The paper does not report original empirical data, and no claims of statistical significance are made. Where illustrative deal situations appear, they are stylised composites used to demonstrate a mechanism and are not accounts of specific transactions. 4.2 Literature identification The review drew on peer reviewed articles in behavioural finance, judgment and decision-making, negotiation and corporate finance, together with recent book length syntheses. Priority was given to work published within the last five years, on the reasoning that the field has consolidated rapidly and that recent systematic and bibliometric reviews provide reliable coverage of the older foundational studies (Jain, Walia, Singh and Jain, 2021; Maheshwari, Samantaray and Jena, 2023; Hachicha, Argoubi and Guesmi, 2024). Foundational contributions from earlier decades are referenced conceptually through those reviews rather than cited individually, which keeps the reference list current without losing the intellectual history. Inclusion required that a study address at least one of the following: a documented psychological bias in a financial or negotiation context; the measurement of such a bias; the organisational or governance conditions that moderate it; or an intervention intended to reduce it. Purely descriptive market anomaly studies without a psychological mechanism were excluded. 4.3 Coding and synthesis Each retained bias was coded on four dimensions: its definition; the empirical support for its existence in professional rather than student populations; the stage or stages of the co-investment cycle at which it plausibly operates; and the evidence, if any, for interventions that reduce it. Biases were then mapped onto the five stage cycle described in Section 3. 4.4 Limitations of the method An integrative review inherits the weaknesses of the studies it reviews. Three deserve mention. Much behavioural finance evidence is cross sectional and survey based, which supports association rather than causation. Self reported measures of one's own overconfidence are particularly vulnerable, since the trait being measured interferes with the measurement. Much of the negotiation evidence comes from laboratory studies with student participants and modest stakes. Whether effects of the same magnitude survive when the participants are experienced professionals and the stakes are hundreds of millions is an open empirical question, though the acquisition literature suggests they survive in some form. Private market data is not public. Deal level co-investment outcomes are rarely disclosed, which is precisely why this literature is thin and why the paper stops at propositions rather than tests. 5. Analysis: Bias Across the Negotiation Cycle 5.1 Anchoring #anchoring is the tendency for an initial value to exert disproportionate influence on subsequent numerical judgments, even when the initial value is uninformative. It is the most direct threat in a negotiated transaction, because the entire process begins with a number provided by the party with the strongest interest in it being high. The mechanism has two components. Insufficient adjustment occurs when the negotiator begins at the anchor and adjusts towards a better estimate, but stops too early, usually as soon as the number becomes minimally plausible. Selective accessibility occurs when the anchor causes the mind to retrieve anchor consistent evidence, so that the negotiator genuinely finds more reasons to believe a high number after seeing a high number. In co-investment, the anchor arrives in the teaser or in the sponsor's model. It rarely arrives alone. It arrives wrapped in a coherent narrative, supported by comparable transactions selected by the sponsor, and reinforced by the implicit message that other investors have already accepted it. That combination makes it far more powerful than the arbitrary anchors used in laboratory studies. The damage is not limited to the price paid. The anchor sets the frame within which the analytical work is done. A team that builds its model after seeing the sponsor's model is not producing an independent estimate. It is producing a variation on the sponsor's estimate, and the range of variation is narrower than the team believes. Three mitigations have reasonable support. The first is to complete an independent #valuation before any exposure to the sponsor's number, and to record it in writing. The second is the consider the opposite technique, meaning the deliberate generation of reasons the anchor might be wrong, which works because it counteracts selective accessibility. The third is to construct a #counteranchor grounded in the co-investor's own analysis, so that the discussion has two reference points rather than one. Awareness alone, meaning simply telling a negotiator that anchoring exists, is not an effective mitigation, and the belief that it is may itself be a bias. 5.2 Overconfidence Overconfidence is not one thing. Overestimation is thinking one's abilities or one's asset are better than they are. Overplacement is thinking one is better than others, which produces the finding that most fund managers place themselves above median. Overprecision is thinking one's estimate is more accurate than it is, which produces confidence intervals that are far too narrow. The evidence from professional finance is consistent. Overconfidence increases trading and risk taking, and it is associated with weaker performance, with risk propensity acting as an intervening mechanism (Ul Abdin, Qureshi, Iqbal and Sultana, 2022). Its effects are moderated by how information is acquired, meaning that the same trait produces different outcomes depending on the informational discipline surrounding it (Naveed and Taib, 2021). At the top of the organisation, overconfidence drives acquisitions, shapes financing decisions and interacts with distress (Twardawski and Kind, 2023; Ge, Jamil and Yu, 2024; Kowalzick, Ahrens, Lauterbach and Tang, 2024). In co-investment, overprecision is the form that does the most damage, and it does it quietly. A discounted cash flow model with a base case, an upside case and a downside case appears to represent uncertainty. In practice the three cases are often clustered tightly around the base case, because the modeller generated the downside by adjusting the base case rather than by asking what would have to be true for the investment to lose money. The result is a #synergy_forecast or a demand forecast whose stated range excludes the outcomes that actually occur. Overconfidence also corrupts the estimation of the #BATNA. A negotiator who overestimates the availability of alternative deals will hold out too long. One who overestimates the quality of the current deal will hold out too little. Both errors flow from the same source. The practical response is calibration rather than suppression. Confidence is required to commit capital, and recent reviews are right to treat it as double edged rather than simply harmful (Kunz and Sonnenholzner, 2023). What is required alongside it is a discipline of recording forecasts, revisiting them against outcomes, and tracking the hit rate of one's own stated confidence intervals. Institutions that have never checked whether their eighty percent confidence intervals contain the truth eighty percent of the time have no basis for believing that they do. 5.3 Loss aversion and reference dependence #loss_aversion means that the pain of a loss exceeds the pleasure of an equivalent gain, typically by a factor somewhere near two. In negotiation it produces several distinct effects. Concessions feel like losses. Once a negotiator has stated a position, moving away from it registers as giving something up rather than as buying agreement. This makes positions sticky in a way that pure economic reasoning does not predict. The endowment effect makes parties overvalue what they already hold. A sponsor who has held an asset for four years will value it above what the market will pay, and will interpret a lower bid as an insult rather than as information. Reference point selection determines whether a party sees itself as gaining or losing, and it is therefore itself a negotiating lever. A co-investor who frames the decision relative to the sponsor's asking price sees every concession as a gain. A co-investor who frames it relative to an independently derived intrinsic value sees the same concessions differently. Neither frame is neutral, but only one of them is anchored to the investor's own analysis. Most importantly, a party operating in the loss domain becomes risk seeking. A manager whose fund is behind its benchmark, or a team that has already lost two competitive processes this year, is not simply under pressure. That party's evaluation of risk has changed shape, and it will now accept gambles it would previously have declined. Recognising this in a counterparty is a source of negotiating advantage. Recognising it in oneself is a source of protection. 5.4 Framing #framing is the finding that logically equivalent descriptions of the same situation produce different choices. A deal described as having an eighty percent chance of meeting its return target is received differently from one described as having a twenty percent chance of missing it, even by professionals who understand the equivalence. Framing in co-investment operates through vocabulary, through comparison sets and through defaults. A transaction presented as a platform for growth invites different questions from one presented as a leveraged bet on continued demand. An asset compared to a set of successful precedents looks different from the same asset compared to a set of failures. A term presented as market standard is harder to resist than the identical term presented as the sponsor's proposal. The sponsor controls most of the framing, because the sponsor controls the materials. The co-investor's most valuable defensive habit is therefore reframing, meaning restating the opportunity in the institution's own vocabulary, in its own comparison set, and against its own criteria, before engaging with the sponsor's version. 5.5 Confirmation bias #confirmation_bias is the tendency to seek, notice and weight evidence that supports a conclusion one has already reached, and to discount evidence that does not. It is the most corrosive bias in #due_diligence because due diligence looks like the opposite of it. The sequence is familiar. A senior person forms an early positive view. The team is instructed to build the investment case. The workstreams are scoped around the strengths. Weaknesses surface, and each is individually mitigated or deemed manageable. Nobody lies, nobody is negligent, and the process concludes with a thick document that says yes to a question that was never seriously asked. The problem is compounded by a structural feature of deal teams. Once resources have been committed and the transaction has been socialised internally, the psychological cost of a negative finding rises. Analysts learn quickly which findings are welcome. Recognition based heuristics reinforce this, since the familiarity of the sponsor supplies a prior that the deal is sound and thereby lowers the perceived need for disconfirming search (Ahmad, Wu and Abbass, 2023). The evidence backed remedy is structural, not attitudinal. A #red_team is a group given the explicit task and the explicit incentives to argue against the transaction. A #pre_mortem is an exercise in which the team is told to imagine that the investment has failed badly three years from now and to write the history of that failure. Both work because they convert the search for disconfirming evidence from a disloyal act into an assigned responsibility. A formally appointed #devils_advocate has weaker effects, because participants know the role is theatrical, which is precisely why the assignment must carry real consequences. 5.6 Herding #herding is the tendency to follow the actions of others rather than one's own information, and it is the most studied bias among institutional investors specifically. Herding is not always irrational. If other investors possess information you lack, following them is a sensible inference. It becomes a bias when the following is based on the fact of others' participation rather than on any assessment of what they know, and when it produces cascades in which each participant relies on predecessors who were themselves relying on predecessors. In co-investment, herding operates through the #syndication process. The presence of a well known sovereign wealth fund or a respected pension plan in the syndicate is read as validation. What is rarely asked is whether that investor conducted independent underwriting or simply relied on the sponsor. Frequently the answer is that everybody assumed somebody else had done the work. Herding is also career protective, and this is where it links to the #principal_agent problem. Being wrong alongside every other major institution carries far less professional cost than being wrong alone. The rational individual response therefore produces a collectively poor outcome, and no amount of individual debiasing will fix it, because the individuals are not making a mistake by their own lights. Only a change in how decisions are evaluated, specifically evaluating the quality of the process rather than the conformity of the outcome, can address it. 5.7 Escalation of commitment and sunk cost Escalation of commitment is the tendency to increase investment in a failing course of action in order to justify prior investment. #sunk_cost reasoning is the specific error of allowing unrecoverable past expenditure to influence a forward looking decision. Co-investment is structurally exposed at two points. Within the transaction, the co-investor accumulates costs: advisers, internal hours, and internal reputation. Each increment raises the psychological price of walking away. Late in a process the question quietly shifts from whether this is a good investment to whether the institution can afford to abandon the work already done. It cannot afford to complete a bad investment, but that is not what the question feels like at eleven at night in week nine. After closing, escalation appears in follow on funding. An asset underperforms. The sponsor requests additional capital. The co-investor faces a decision that should be evaluated purely on the forward prospects of the incremental capital, but is instead evaluated against the wish that the original decision be vindicated. Loss aversion supplies the risk seeking, self justification supplies the narrative, and the result is good money after bad. Three defences have support. Separate the decision maker from the decision defender, so that follow on decisions are approved by people who did not approve the original. Pre-commit in writing to abandonment criteria, defined as specific observable conditions under which the institution will stop, rather than vague intentions. Frame the follow on decision as a fresh investment, by asking whether the institution would buy this asset today at this price knowing what it now knows, given no prior position. 5.8 The fixed pie bias The fixed pie bias, sometimes called the mythical fixed pie, is the assumption that the parties' interests are perfectly opposed. It leads negotiators to fight over a single dimension, almost always price, and to treat everything else as detail. In co-investment this is expensive, because the parties genuinely value different things. The sponsor may care intensely about closing speed, about the size of the cheque, about avoiding a competitive process that reveals price, and about protecting its own fee economics. The co-investor may care about downside protection, about information rights, about tag along provisions, about follow on rights, and about board representation. These are not the same currency. That is precisely what makes trade possible. A negotiator who concedes governance rights in order to win a small reduction in valuation has often given away more than they gained, because governance rights determine what happens in the bad states of the world, and the bad states are what determine long run returns. The failure is not one of arithmetic. It is a failure of imagination produced by the assumption that the negotiation is about one number. The remedy is preparation. Before entering the room, the co-investor should list every issue in play, score each on its own importance, and estimate the counterparty's score. The gaps between the two columns are where value is created. This is elementary, it is taught in every negotiation course, and it is skipped constantly under time pressure. 5.9 The self serving bias and fairness The self serving bias causes each party to arrive at a definition of fairness that happens to favour its own position, and to hold that definition sincerely. It is distinct from strategic misrepresentation. The negotiator is not lying. The negotiator has interpreted ambiguous evidence in a self favouring direction and has forgotten that the interpretation was a choice. In a co-investment, both sides can point to defensible comparable transactions, defensible discount rates and defensible growth assumptions that support their preferred valuation. Each believes the other is being unreasonable. The gap is not bridged by more analysis, because more analysis simply gives each side more material to interpret self servingly. The known mitigation is to agree the standard before applying it. If both parties commit, in advance, to a specific method, a specific comparable set and a specific data source, the self serving interpretation has less room to operate. This is why independent valuation experts and pre-agreed adjustment mechanisms are effective, not because the experts are smarter, but because the standard is fixed before the interests attach to it. 5.10 Availability, representativeness and status quo Three further biases deserve shorter treatment. The #availability_heuristic causes probability to be judged by ease of recall. A recent, vivid failure in one sector will suppress capital allocation to that sector beyond what the evidence supports, and a recent success will attract it. Because institutional memory is short and staff turnover is real, availability tends to track the last two or three years rather than the full cycle. #representativeness causes judgment by similarity rather than by base rate. A company that resembles a past success is judged likely to succeed, without reference to how many similar companies failed. It is the engine behind pattern matching narratives in private markets, and it systematically neglects the denominator. #status_quo_bias favours inaction and the existing allocation. It is often mistaken for prudence. In co-investment it appears as the tendency to keep recycling capital with existing sponsors, and as the tendency to accept the sponsor's standard documentation because renegotiating it is effortful. 5.11 Mapping biases to stages The synthesis can be summarised as follows. At origination and screening, the dominant risks are anchoring, the #availability_heuristic, #representativeness and relationship driven recognition heuristics. Judgments here are fast and consequential because they set the frame. At due diligence and valuation, the dominant risks are confirmation bias, overconfidence in the form of overprecision, and anchoring carried forward from stage one. At price and terms negotiation, the dominant risks are anchoring, loss aversion around concessions, the fixed pie bias, the self serving bias and #framing effects. At documentation and closing, the dominant risks are escalation of commitment, #status_quo_bias towards standard terms, and simple decision fatigue. At post closing and follow on, the dominant risks are escalation of commitment, #sunk_cost reasoning, #hindsight_bias in the review of the original decision, and self attribution, meaning the attribution of good outcomes to skill and bad outcomes to circumstances. 6. Discussion 6.1 Bias is not the only problem: the role of noise A central argument of this paper is that institutions consistently misdiagnose their decision problem. They look for bias, meaning a shared, directional error, and they overlook #noise, meaning unwanted variability in judgments that should agree. The distinction is easy to state. If every member of an investment team overvalues the same asset by fifteen percent, that is bias. If the same asset is valued at eight hundred million by one analyst, one billion by a second, and one point three billion by a third, all working from the same materials, that is noise. Bias moves the average. Noise scatters the individual judgments around it. Both are errors, and importantly, noise can be very large even when average bias is zero, which means an institution that audits only its average is measuring the wrong thing (Kahneman, Sibony and Sunstein, 2021). Noise matters in co-investment for a simple reason. The institution does not make the average decision. It makes one decision, taken by whichever team happened to be assigned, in whatever mood, on whatever day, after whatever recent experience. The dispersion is the exposure. The remedies differ. Bias responds to structured challenge, to independent estimates, and to changing the reference point. Noise responds to what has been termed decision hygiene: using structured, mechanical, criterion based judgment procedures; aggregating independent judgments before discussion; sequencing information so that early impressions do not contaminate later ones; and using relative rather than absolute scales, since people are far more consistent when comparing two things than when scoring one thing in isolation. The reason this distinction is worth pressing is that most institutional responses to behavioural error take the form of training, and training is close to useless against noise. Telling a group of professionals about anchoring does not make their independent estimates converge. Only procedure does that. 6.2 Why professional expertise does not solve the problem A standard objection runs as follows. The biases documented in laboratory research were found in inexperienced participants making unfamiliar decisions with trivial stakes. Professionals with decades of experience, large incentives and formal training will not exhibit them. The objection fails for four reasons. First, the evidence does not support it. Bias has been documented in fund managers, corporate executives, boards and analysts, in field settings, with real money (Twardawski and Kind, 2023; Vukovic and Pivac, 2024; Ul Abdin and colleagues, 2022). Second, expertise requires a particular kind of learning environment to develop reliably: rapid, clear, repeated feedback. Co-investment offers the opposite. Outcomes take five to ten years. They are heavily confounded by market conditions. A single professional may participate in only a few dozen major transactions in an entire career. Under those conditions, confidence grows with experience, but accuracy may not, which is exactly the pattern that produces well calibrated confidence in surgeons and poorly calibrated confidence in long horizon investors. Third, some biases are strengthened rather than weakened by expertise. Experience produces richer pattern libraries, which is exactly what #representativeness feeds on. It also produces stronger priors, which is exactly what confirmation bias feeds on. Evidence that financial literacy and experience have mixed and sometimes counterintuitive effects on bias supports this reading (Suresh, 2024). Fourth, incentives frequently point the wrong way. A career concerned professional who herds is behaving rationally given how they will be judged. The bias is not in the individual. It is in the evaluation system. 6.3 Moderators: what makes some teams less exposed Not all institutions are equally vulnerable, and the framework identifies several moderators. Governance architecture. Committees that require written independent assessments before discussion, that rotate the challenge role, that separate origination from approval, and that keep a formal record of the reasoning are structurally more resistant. The gains come from the sequence, not from the seniority of the members. Information discipline. How information is acquired moderates the effect of overconfidence on decisions (Naveed and Taib, 2021). Teams that mandate a defined minimum set of independent sources, and that refuse to rely solely on the sponsor's data room, are less exposed. Compensation and evaluation. If a professional is rewarded for deals completed rather than for capital deployed well, escalation and herding follow logically. If performance is judged only against the outcome, and never against the quality of the process given what was knowable, then #hindsight_bias will punish good decisions that turned out badly and reward bad decisions that got lucky. Cultural context. Bias profiles are not universal. Cross cultural evidence on institutional investors indicates that the way information is weighted, the willingness to express dissent, and the degree of confidence expressed in an estimate all vary systematically across cultural settings. A committee protocol designed for a low power distance culture, in which junior members contradict senior members freely, may fail in a setting where such contradiction is socially costly. This has direct practical consequences for global institutions that apply a single process worldwide, and it is an under-researched area. Time architecture. Institutions that pre-negotiate response windows with sponsors, and that maintain standing underwriting capacity for co-investment rather than assembling teams on demand, remove the single most powerful amplifier of bias, which is haste. 6.4 The negotiation asymmetry problem There is an uncomfortable implication in the analysis. The sponsor and the co-investor are not equally exposed. The sponsor has spent months on the asset, controls the information, sets the timetable, chooses the comparables, provides the anchor and frames the narrative. The co-investor arrives late, with less information and less time. The structure of the interaction places the co-investor in exactly the conditions under which bias is strongest. Two conclusions follow. First, a co-investor who negotiates without addressing this asymmetry is not simply at an informational disadvantage but at a cognitive one, and the two compound. Second, the most valuable interventions are the ones that happen before the sponsor makes contact: standing underwriting capacity, pre-defined sector views, a pre-agreed internal process, and a habit of forming an independent value estimate before opening the sponsor's model. None of this implies that sponsors act in bad faith. They are doing their job, which includes presenting their asset in the best defensible light. The point is that a process designed by one party will tend to favour that party, and that the favour operates through psychology as much as through information. 6.5 What debiasing can and cannot do The literature on debiasing supports a modest and specific claim. Awareness is necessary and insufficient. Motivation and incentives help but do not eliminate the effects. Structural interventions that change what is done, in what order, by whom, and recorded how, are the ones with the most reliable evidence. Interventions with reasonable support include the following. Independent estimation before exposure to the counterparty's number, which addresses anchoring at its source. Considering the opposite, in which the negotiator explicitly generates reasons the current view is wrong, which addresses both anchoring and confirmation. The pre-mortem, which addresses overconfidence and confirmation by legitimising the search for failure modes. Structured, criterion based scoring of opportunities against a fixed set of dimensions, assessed one dimension at a time, which reduces noise and limits the halo effect by which one attractive feature colours the entire assessment. Independent aggregation, in which committee members submit written views before any discussion, which prevents the first speaker from anchoring the room and preserves the information contained in disagreement. Pre-commitment devices, including a written #reservation_price, written abandonment criteria, and a designated person outside the deal team who holds them. Outcome independent review, in which decisions are audited for process quality against what was knowable at the time, rather than against how they turned out. Interventions with weaker support include general bias training, appointing a devil's advocate without real authority, and simply exhorting people to be objective. These are popular because they are cheap and because they feel like action. There is also a cost side that the literature sometimes underplays. Every layer of structured challenge slows the process, and in competitive co-investment processes speed has real option value. An institution that takes six weeks to decide will not be offered the best allocations. The design problem is therefore not to maximise deliberation but to place deliberation where it earns the most, which is at the points where an error is largest and least reversible: the independent valuation, the reservation price, and the walk away decision. 7. Practical Implications 7.1 For institutions The framework converts into a practical architecture with five commitments. Decide the process before the deal arrives. Every element of the co-investment protocol should be written down while no transaction is live. Under deadline, nobody designs a good process. They use whatever exists. Estimate independently before you look. The team should produce and record its own valuation range, its own downside case and its own walk away price before opening the sponsor's model. This single habit addresses anchoring, framing and confirmation simultaneously, and it costs nothing but sequence. Write before you talk. Committee members should submit independent written assessments before discussion begins. The dispersion in those assessments is itself the most useful diagnostic an institution can generate, because it measures the noise that would otherwise be hidden by consensus. Pre-commit and externalise. The reservation price and the abandonment criteria should be written, dated and held by someone who has no stake in completing the transaction. A commitment that only the deal team can revise is not a commitment. Audit decisions, not just outcomes. A decision journal, recording what was believed, with what confidence, on what evidence, allows an institution to learn. Without it, hindsight rewrites the past and the organisation learns nothing except that it was right all along. 7.2 For students and early career professionals Three points are worth internalising. Bias is not a character flaw and it is not a synonym for stupidity. It is a systematic feature of how competent minds work under uncertainty. The professionals most vulnerable to it are frequently the most experienced and the most confident, because both experience and confidence reduce the felt need to check. The most important number in any negotiation is the one you decided before you entered the room. Everything after that is pressure. Learn to notice the moment when the question changes from whether this is a good investment to whether we can still get this done. That moment is the point at which analysis has ended and justification has begun, and it usually goes unremarked. 8. Limitations and Future Research 8.1 Limitations This paper is conceptual. It generates propositions rather than testing them, and the propositions should not be read as findings. The evidence base is uneven. Anchoring, overconfidence and loss aversion are supported by large and replicated literatures. Herding among institutional investors is well studied. Other claims, in particular those about the interaction of amplifiers and about the specific vulnerability of co-investment relative to other private market decisions, rest on plausible extrapolation from adjacent settings such as acquisitions rather than on direct evidence. Much of the underlying evidence uses samples that are not co-investors. Extrapolating from retail investors, students or corporate executives to institutional co-investment professionals requires an assumption of psychological continuity that is reasonable but unproven. Effect sizes are unknown for the setting. The paper cannot say how much a given bias costs a co-investment programme, and it deliberately declines to guess. 8.2 Future research agenda Six directions appear most promising. Field experiments within institutions, in which the order of information provision is varied, would provide a clean test of proposition one. Randomising whether teams see the sponsor's valuation before or after producing their own is administratively feasible and would be highly informative. Noise audits in investment committees, in which multiple teams independently assess the same real opportunity, would establish the scale of the variability problem. Comparable audits in insurance and forensic settings have found dispersion far larger than practitioners predicted. Studies of forecast calibration in private markets, comparing stated confidence intervals in original investment memoranda against realised outcomes, would allow direct measurement of overprecision using data institutions already hold. Research on committee protocol design, comparing written independent aggregation against conventional discussion in real approval settings, would test proposition five. Cross cultural work on institutional bias profiles, extending the emerging evidence that cultural context shapes how professional investors weigh information, is needed before global institutions can sensibly standardise a decision process. Finally, the role of algorithmic and machine assisted valuation deserves attention. Models can reduce noise, because they are perfectly consistent, but they can also encode and then hide bias, because they are trained on past decisions that were themselves biased. Whether machine assistance in private market underwriting reduces total error or merely relocates it is an open and increasingly urgent question. 9. Conclusion Institutional co-investment concentrates every condition under which human judgment is known to fail. The information is asymmetric, the time is short, the stakes are large, the outcome is irreversible, the price is negotiated rather than observed, the counterparty controls the frame, and the decision is ratified by a group that is often more deferential than it believes. This paper has argued three things. First, that the psychological biases documented across behavioural finance and negotiation research operate in this setting with unusual force, and that they arrive in an ordered sequence across the deal cycle rather than at random. Second, that three amplifiers, information asymmetry, time compression and committee dynamics, interact to make co-investment more exposed than either the finance or negotiation literature would predict alone. Third, that institutions systematically misdiagnose the problem by treating it as a matter of individual awareness, when much of the error is noise rather than bias, and when the interventions with the best evidence are structural rather than educational. The uncomfortable conclusion is that a co-investment decision is not primarily a test of analytical skill. Analytical skill is abundant in this industry. It is a test of process. Two teams of equal ability, looking at the same asset with the same data, will reach materially different conclusions, and the difference will be determined less by intelligence than by the order in which they saw things, the deadline they were given, who spoke first in the room, and whether anyone had written down, in advance, the price at which they would walk away. That is a sobering thought. It is also an encouraging one, because process is something an institution can actually change. References Ahmad, M., Wu, Q. and Abbass, Y. (2023). Probing the impact of recognition based heuristic biases on investment decision making and performance. Kybernetes, 52(10), 4229 to 4256. Brahma, S., Boateng, A. and Ahmad, S. (2023). Board overconfidence and M and A performance: evidence from the UK. Review of Quantitative Finance and Accounting, 60(4), 1363 to 1391. Ge, L., Jamil, T. and Yu, J. (2024). CEO overconfidence and the choice of debt issuance. Journal of Banking and Finance, 161, 107104. Goyal, P., Gupta, P. and Yadav, V. (2023). Antecedents to heuristics: decoding the role of herding and prospect theory for Indian millennial investors. Review of Behavioral Finance, 15(1), 79 to 102. Gu, W. (2023). Impact of managers overconfidence on listed firms entrepreneurial behavior in an emerging market. Journal of Business Research, 155, 113453. Hachicha, F., Argoubi, M. and Guesmi, K. (2024). The knowledge domain and emerging trends in behavioral finance: a scientometric analysis. Research in International Business and Finance, 70, 102350. Jain, J., Walia, N., Kaur, M. and Singh, S. (2022). Behavioural biases affecting investors decision making process: a scale development approach. Management Research Review, 45(8), 1079 to 1098. Jain, J., Walia, N., Singh, S. and Jain, E. (2021). Mapping the field of behavioural biases: a literature review using bibliometric analysis. Management Review Quarterly, 72, 823 to 855. Jain, J., Walia, N., Singla, H., Singh, S., Sood, K. and Grima, S. (2023). Heuristic biases as mental shortcuts to investment decision making: a mediation analysis of risk perception. Risks, 11(4), 72. Kahneman, D., Sibony, O. and Sunstein, C. R. (2021). Noise: A Flaw in Human Judgment. New York: Little, Brown Spark. Kowalzick, M., Ahrens, J. P., Lauterbach, J. G. and Tang, Y. (2024). Overconfident CEOs in dire straits: how incumbent and successor CEOs overconfidence affects firm turnaround performance. Journal of Management Studies, 61(5), 1985 to 2032. Kumar, P., Kasilingam, R. and Rajamohan, S. (2024). Risk perception and perceived investor performance nexus: evaluating the mediating effects of heuristics and prospects with gender as a moderator. SAGE Open, 14(2). Kunz, J. and Sonnenholzner, L. (2023). Managerial overconfidence: promoter of or obstacle to organizational resilience? Review of Managerial Science, 17(1), 67 to 128. Maheshwari, H., Samantaray, A. K. and Jena, J. R. (2023). Unravelling behavioural biases in individual and institutional investors investment decision making: intersection of bibliometric and systematic literature review. IIM Kozhikode Society and Management Review. Naveed, F. and Taib, H. M. (2021). Overconfidence bias, self attribution bias and investor decisions: moderating role of information acquisition. Pakistan Journal of Commerce and Social Sciences, 15(2), 354 to 377. Sood, K., Pathak, P., Jain, J. and Gupta, S. (2023). Gauging investors investment decisions in the crypto market through the PRISM of behavioral biases: a fuzzy AHP approach. International Journal of Emerging Markets. Suresh, G. (2024). Impact of financial literacy and behavioural biases on investment decision making. FIIB Business Review, 13(1), 72 to 86. Thaler, R. H. and Sunstein, C. R. (2021). Nudge: The Final Edition. New York: Penguin Books. Trehan, B. and Sinha, A. K. (2021). A study of confirmation bias among online investors in virtual communities. International Journal of Electronic Finance, 10(3), 159 to 179. Twardawski, T. and Kind, A. (2023). Board overconfidence in mergers and acquisitions. Journal of Business Research, 165, 114026. Ul Abdin, S. Z., Qureshi, F., Iqbal, J. and Sultana, S. (2022). Overconfidence bias and investment performance: a mediating effect of risk propensity. Borsa Istanbul Review, 22(4), 780 to 793. Vukovic, M. and Pivac, S. (2024). The impact of behavioral factors on investment decisions and investment performance in Croatian stock market. Managerial Finance, 50(2), 349 to 366. Article Hashtags #psychological_biases_in_finance #strategic_financial_negotiations #institutional_co_investments #behavioral_finance #negotiation_strategy #decision_making_under_uncertainty #cognitive_bias #private_capital #investment_committee #debiasing #anchoring_effect #overconfidence_bias #loss_aversion_theory #due_diligence_process #finance_students

Organisations across health, education, welfare and business are handing paperwork to machines. Notes are written by ambient listening tools, essays are marked by language models, benefit claims are screened by scoring systems, and shift rosters are set by software. The promise behind this shift is simple and attractive: if machines absorb the #administrative_burden, people will have more time and energy for the parts of the work that only people can do, above all the work of understanding another human being. This article asks whether that promise holds. Drawing on an integrative review of recent empirical and conceptual research published between 2021 and 2026, it examines what actually happens to #empathy and #human_connection when routine tasks are transferred to #algorithms. The evidence gives a mixed picture. Automation does reduce documentation time and self reported exhaustion in some settings, and machine generated messages are often rated as warm, responsive and even more compassionate than those written by trained professionals. Yet the freed time is frequently reabsorbed by higher caseloads, the same tools quietly narrow professional #digital_discretion, and people report feeling less heard the moment they learn that a message came from a machine. The article proposes the Empathy Displacement and Reinvestment framework, which argues that a "time dividend" only becomes an "empathy dividend" under three conditions: time must be released, protected, and reinvested into relationships by design rather than by accident. It closes with practical design and #policy_design principles for institutions, professionals and students who will spend their working lives inside these systems. Keywords: empathy; automation; algorithmic administration; human connection; administrative burden; digital discretion; care ethics; future of work 1. Introduction There is an old complaint in almost every caring or public facing profession. The complaint is that the paperwork is eating the work. Doctors say they became doctors to sit with patients, not to type into a screen while a patient is speaking. Teachers say they trained to teach, not to spend Sunday evenings marking sixty near identical scripts. Social workers say the case management system now decides what a home visit is worth. Nurses, benefits officers, admissions officers, customer service agents and human resources staff say versions of the same thing. The relational core of their job, the part that requires attention to a particular person in a particular situation, keeps getting squeezed by the bureaucratic shell around it. Into this frustration arrives a very persuasive argument. If the shell is what crushes the core, then let machines carry the shell. Let an ambient tool listen to the consultation and produce the note. Let a language model draft the feedback. Let a scoring system triage the applications. The professional is then free, at last, to be present. This is the argument made by vendors, by ministries, by hospital boards and by many thoughtful practitioners, and it is not a foolish argument. It is grounded in a real diagnosis of a real problem. This article takes the argument seriously and then tests it. The central question is not whether machines can process forms faster than humans, because they clearly can. The question is what happens to the relationship between a professional and the person in front of them once the forms are gone. Does the released time flow back into presence, patience and attention? Or does it flow somewhere else entirely, into more cases, more throughput, more surveillance, and a slow erosion of the very skills that the automation was supposed to protect? Three tensions run through the analysis. The first is the tension between #efficiency_logic and relational logic. Efficiency logic measures value in units processed per hour. Relational logic measures value in whether a person felt understood. These two logics are not natural enemies, but they compete for the same scarce resource, which is human attention. When an institution automates, it must decide what the saved minutes are for. That decision is rarely made explicitly, and when it is not made explicitly, efficiency logic tends to win by default. The second is the tension between performed empathy and felt empathy. Recent studies show that text generated by large language models is often rated by readers as more #compassion ate, more responsive and more caring than text written by human professionals, including trained crisis responders (Ovsyannikova, de Mello, & Inzlicht, 2025; Ayers et al., 2023). This is an uncomfortable finding for anyone who assumes that warmth is the last human monopoly. It forces a harder question: if the output feels empathic, does it matter that nobody was actually feeling anything? The third is the tension between the individual and the system. #empathy is usually discussed as a personal virtue, something a good doctor or a good teacher has. But empathy is also an institutional product. It requires time, staffing levels, continuity of relationship, and a manageable caseload. Kerasidou, Baeroe, Berger and Caruso Brown (2021) argue that healthcare systems, not just healthcare workers, must be designed to make empathy possible. A kind person in a fifteen minute slot with forty patients waiting will still fail to be empathic. This means that automation should be judged not by whether it makes individuals kinder but by whether it makes institutions capable of kindness. 1.1 Aim and contribution This article makes four contributions. First, it synthesises evidence from four different literatures that rarely talk to each other: clinical informatics research on #clinical_documentation and #ambient_scribes, public administration research on #algorithmic_bureaucracy and #street_level_bureaucracy, organisational research on #algorithmic_management, and psychological research on perceived and machine generated empathy. Second, it distinguishes clearly between three things that are usually confused: the automation of tasks, the automation of decisions, and the automation of relationships. These are different acts with very different consequences for #human_connection. Third, it proposes the Empathy Displacement and Reinvestment framework, a simple model of when automation helps connection and when it harms it. Fourth, it offers concrete, testable principles for institutions and for students entering these professions, on the assumption that the readers of this article will not be designing the systems so much as living inside them. 1.2 Structure Section 2 reviews the literature and defines the key terms. Section 3 sets out the conceptual framework. Section 4 describes the review method. Section 5 presents evidence from #healthcare, #education, welfare administration, the workplace and #customer_service. Section 6 discusses six mechanisms that link automation to changes in empathy. Section 7 offers design and policy principles. Section 8 sets out limits and a research agenda. Section 9 concludes. 2. Literature Review and Conceptual Background 2.1 What empathy actually is Empathy is a loose word, and the looseness causes real confusion in debates about machines. It is helpful to split it into at least four components. Cognitive empathy is the ability to work out what another person is thinking or feeling. It is a form of accurate inference. A skilled clinician who notices that a patient's cheerfulness is a mask is exercising cognitive empathy. Affective empathy is the sharing of another person's emotional state, feeling some echo of their distress or relief. Compassionate concern, sometimes called empathic concern, is the motivation to act for the other person's benefit. It goes beyond understanding and beyond feeling into caring. Expressed empathy is the communicative behaviour: the words, tone, pauses and gestures that let the other person know they have been understood. This four part split matters because machines can perform some of these and not others. A language model can produce excellent expressed empathy. It can also perform something that functionally resembles cognitive empathy, in the sense of inferring emotional states from text with high accuracy. What it cannot do, on any current understanding, is feel, and it cannot care in the sense of having a stake in the outcome. Perry (2023) argues that this gap is not a technical shortfall to be closed with better models but a categorical difference, because the value of human empathy lies partly in the fact that another conscious being chose to spend their limited attention on you. Montemayor, Halpern and Fairweather (2022) make a similar in principle argument for healthcare, holding that empathic care involves a moral relationship that cannot be simulated. The opposing position is also serious. Inzlicht, Cameron, D'Cruz and Bloom (2024) point out that human empathy is not the reliable resource we imagine. It is biased towards people who look like us, it fades under fatigue, and it collapses under scale. A machine does not suffer compassion fatigue at four in the morning. If a lonely person receives a response that helps them feel understood, the argument goes, the metaphysics of the responder may matter less than we like to think. This is the #empathy_gap debate in its sharpest form, and it is unresolved. 2.2 Administrative burden and why it matters The second literature concerns #administrative_burden: the costs that people bear when they interact with organisations. These costs are usually divided into learning costs (working out what is required), compliance costs (filling in forms, gathering documents, attending appointments) and psychological costs (stress, stigma, the feeling of being processed rather than helped). Burden falls on two sides. It falls on the professional, in the form of documentation, coding, reporting and audit. It falls on the citizen, patient, student or customer, in the form of forms, portals and queues. Madsen, Lindgren and Melin (2022) show that digital self service does not simply remove burden but often moves it, turning the citizen into what they call an accidental caseworker who now performs administrative labour that a staff member used to perform. This is a crucial point for our argument, because the same displacement can happen with #algorithms. Removing a task from a professional's desk does not delete the task. It relocates it, either to a machine, to another worker, or to the person being served. The professional side of the burden is best documented in medicine. Documentation is consistently identified as a leading driver of #burnout, and time in the electronic record has become a standard proxy for how much of a clinician's life is spent on the shell rather than the core. This is why ambient documentation tools have been adopted faster than almost any other clinical technology in recent years. 2.3 From street level to system level Public administration scholarship supplies the third literature. The classic image of the frontline worker is of someone with discretion: a person who applies general rules to particular cases and who can bend, soften or accelerate a rule when the human situation demands it. Discretion is where empathy lives in a bureaucracy. It is the space in which an officer can say, in effect, the rules say no, but I can see what is happening here, so let me find another way. Digitisation compresses that space. Ranerup and Henriksen (2022) describe how automated decision making in Swedish social services redistributes agency between humans and technology, producing what they call digital discretion, in which the caseworker's judgment is increasingly channelled through what the system will accept. Considine, McGann, Ball and Nguyen (2022), asking whether robots can understand welfare, show how machine bureaucracies in welfare to work schemes reshape the relationship between adviser and jobseeker. Roehl and Crompvoets (2023) analyse the tension between automated decision making and the principles of good administration. Schiff, Schiff and Pierson (2022) develop the idea of public value failure to describe cases where government adoption of AI satisfies efficiency criteria while failing the values that justify public service in the first place. Two further problems appear in this literature. Busuioc (2021) argues that #accountability becomes structurally difficult when decisions are produced by systems whose reasoning nobody can fully reconstruct, which creates a gap between the person who is affected and any human who can be held responsible. Grimmelikhuijsen (2023) shows experimentally that #transparency about how an algorithm decides affects whether people perceive the decision as trustworthy, which suggests that the felt legitimacy of a decision depends not only on its content but on whether a person can see and question the reasoning behind it. The consequence is that automation in public services does not only save time. It also changes who is allowed to be flexible, and how much room there is for a human to notice that this particular case is not like the others. That room is the raw material of #relational_care. 2.4 Algorithmic management and the workplace The fourth literature concerns #algorithmic_management, the use of software to allocate tasks, monitor performance and evaluate workers. Zhang, Cooke, Ahlstrom and McNeil (2025) synthesise this field and identify a recurring set of concerns: pervasive surveillance, embedded bias, #deskilling, dehumanisation and worker alienation, alongside genuine gains in coordination and efficiency. Parent Rocheleau, Parker, Bujold and Gaudet (2024) developed a validated instrument to measure how workers actually experience algorithmic management, which reflects how quickly this has become a mainstream feature of working life rather than a gig economy curiosity. The relevance to empathy is direct. When a manager's judgment is replaced by a dashboard, the manager loses the practice of judgment. When a worker is evaluated by metrics that cannot see context, the worker learns to perform the metric rather than the work. And when a middle manager is measured on throughput, the manager will not protect a subordinate's time for a difficult conversation with a distressed client, because that conversation does not appear anywhere in the reporting system. 2.5 Machine generated empathy The final literature is the newest and the most unsettling. Ayers et al. (2023) compared physician answers to public health questions with answers generated by a chatbot. A panel of licensed healthcare professionals, blind to the source, preferred the chatbot's answers in the large majority of cases and rated them substantially higher on both quality and empathy. Part of the explanation is prosaic: the machine wrote much longer answers, and length correlates with perceived warmth. Busy doctors write clipped answers because they are busy. Ovsyannikova, de Mello and Inzlicht (2025) pushed the finding harder across four preregistered experiments. Third party readers rated AI generated empathic responses as more compassionate and more responsive than human ones, and this held when the AI was compared with trained crisis responders, and even when the source was openly disclosed. Sorin et al. (2024) reviewed the wider literature and found broadly consistent evidence that large language models produce responses that people evaluate as empathic. But the picture is not one sided. Yin, Jia and Wakslak (2024) found that while AI generated messages made recipients feel more heard than messages from untrained humans, simply labelling a message as machine written reduced the feeling of being heard. The label changed the experience even when the words did not change. Rubin et al. (2025) similarly examined how people value empathy that is perceived to come from a human as against a machine, and found that the perceived source carries independent weight. Zohar, Bloom and Inzlicht (2026) argue against what they call frictionless AI, suggesting that the effort and cost involved in human care are part of what gives it meaning, so removing all friction may remove some of the value. The synthesis is uncomfortable but clear. Machines can produce empathic text that readers often prefer. Machines cannot produce empathic relationships, because the value of being cared for depends partly on knowing that someone chose to care. This is the #authenticity problem, and it will not be solved by better models. 2.6 Emotional labour and the economics of attention One further idea connects these literatures. In caring and public facing work, warmth is not simply a personal quality that some workers happen to possess. It is a task. It is part of the job description, whether or not it appears in writing, and it consumes energy in the same way that any other task consumes energy. This is what is meant by #emotional_labour: the managed performance of appropriate feeling as a condition of employment. Once empathy is understood as labour rather than as personality, several things follow. It can be rationed. It can be exhausted. It can be crowded out by other tasks that are measured while it is not. And, critically, it can be automated, or at least its outward performance can be, which is exactly what emotionally fluent language models now make possible at near zero marginal cost. This reframing explains why the debate about whether machines are more empathic than humans is often badly posed. Under current conditions, professionals are not permitted to spend their empathy freely. They are working against clocks and targets that treat every minute of unhurried attention as a cost. When a machine that faces no such constraint is compared against a human who does, the comparison is not between machine capability and human capability. It is between an unconstrained system and a constrained one. The finding that the unconstrained system produces warmer text tells us at least as much about the constraint as about the system. 3. Conceptual Framework: Empathy Displacement and Reinvestment The literature above supports a framework with three claims. 3.1 The three kinds of automation Not all automation touches connection in the same way. It is useful to separate three types. Task automation removes clerical work: transcription, form filling, scheduling, coding, summarising. In principle it does not change any relationship. The professional still meets the person; the machine simply writes it up afterwards. Ambient documentation is the clearest example. Decision automation removes or shapes judgment: eligibility scoring, risk prediction, triage, automated grading, shortlisting. Here the relationship is affected, because a decision that once required a human to look at a case is now made, or heavily pre framed, by a model. #digital_discretion shrinks. Relational automation removes the encounter itself: the chatbot that answers the patient, the automated feedback that replaces the tutorial, the virtual assistant that handles the complaint. Here the relationship is not merely shaped but substituted. Most public debate collapses these three. Vendors sell task automation and quietly ship decision automation. Institutions adopt decision automation and drift into relational automation because it is cheaper. Keeping the three separate is the first analytical move this article recommends. 3.2 The time dividend fallacy The dominant argument for automation in caring professions is what we can call the time dividend argument: automation frees minutes, and the freed minutes will be spent on people. The framework proposed here holds that this is a fallacy in its simple form. Freed time is not automatically relational time. Time released by automation goes somewhere, and where it goes is determined by the incentive structure of the institution, not by the intentions of the professional. There are, broadly, four destinations: Reinvestment. The time goes back into the encounter. Consultations lengthen, feedback becomes personal, home visits happen. Reabsorption. The time is captured by the institution as capacity. Caseloads rise, clinic slots increase, class sizes grow. The professional is no less rushed than before; there are simply more people to rush. Recovery. The time goes to the worker's own life: leaving on time, less evening work, less exhaustion. This is a genuine good, and it may indirectly support empathy by reducing burnout, but it is not the same as more relational time. Redundancy. The time is eliminated by cutting posts. The institution banks the saving. Only the first destination produces an #empathy_dividend directly. The third produces one indirectly and slowly. The second and fourth produce none at all, and may reduce connection below its pre automation level. 3.3 The three conditions From this it follows that automation supports empathy only when three conditions hold together. Condition 1: Release. The technology must actually save meaningful time or cognitive load. This is an empirical question and the answer is often smaller than the marketing suggests. Condition 2: Protection. The released time must be ring fenced against reabsorption. Without a deliberate protective mechanism, such as a commitment not to increase caseload following implementation, the default institutional response is to convert saved time into volume. Condition 3: Reinvestment capability. The professional must retain the skill, the authority and the relational competence to use the protected time well. If discretion has been narrowed, if the professional has been deskilled, or if they have lost the habit of unhurried attention, the protected time will not become connection. It will become awkward silence. The three conditions can be stated compactly. #empathy = f(released time x protected time x relational capability). If any factor is zero, the product is zero. This is the core of the Empathy Displacement and Reinvestment framework, and the rest of the article tests it against evidence. 4. Method This article is an integrative narrative review. It is not a systematic review and does not claim exhaustive coverage. The aim is conceptual synthesis across fields that use different vocabularies for the same underlying phenomenon. Selection. Literature was identified through targeted searching of academic databases and journal collections in medicine and health informatics, public administration and policy, organisational studies and employment relations, education research, and psychology. Search terms combined an automation term (artificial intelligence, algorithm, automated decision making, ambient documentation, algorithmic management) with a relational term (empathy, compassion, human connection, discretion, trust, care). Inclusion criteria. Preference was given to peer reviewed work published from 2021 onwards, with priority to empirical studies, randomised trials, systematic reviews and validated instruments. Conceptual and normative work was included where it clarified definitions or offered a theoretical mechanism. Vendor material, promotional white papers and non peer reviewed commentary were excluded from the evidence base, although they were read to understand the claims being made in practice. Analysis. Sources were coded against the three conditions of the framework (release, protection, reinvestment capability) and against the three types of automation (task, decision, relational). Where evidence conflicted, the conflict is reported rather than resolved. Limitations of the method. Narrative synthesis is vulnerable to selection bias. The evidence base is also unbalanced: healthcare is heavily researched, while social work, further education and small business #customer_service are thinly covered. Most of the empirical work comes from high income countries with well resourced digital infrastructure, which limits transferability. These limits are revisited in Section 8. 5. Evidence Across Five Domains 5.1 Healthcare: the best case for automation, and its limits Healthcare offers the strongest test of the time dividend argument, because the burden is severe, the technology is mature, and the relational stakes are obvious. What the evidence shows on release. Ambient documentation tools listen to a consultation and draft the clinical note. Lukac et al. (2025) reported a pragmatic randomised trial involving 238 outpatient physicians across fourteen specialties, comparing two commercial ambient scribe products against usual care. The results were more modest than the enthusiasm surrounding the technology. On the primary outcome, the time physicians spent writing in the note, only one of the two products produced a statistically significant reduction against control, and the size of that reduction was in the region of a tenth of the baseline. Secondary measures were more encouraging: physicians in the intervention groups reported lower task load and less work exhaustion. Read carefully, this is an important result. It suggests that the main benefit of #ambient_scribes may not be minutes saved but load lifted. The clinician who does not have to type while listening is doing one job instead of two. That is a cognitive gain, and cognitive bandwidth is precisely what empathy requires. A clinician who is composing a note in their head while a patient describes their fear is not fully present, however kind their expression. What the evidence shows on protection. Here the picture is worse. Observational work on ambient scribes has repeatedly found that consultation length does not increase and clinic throughput does not change. The saved documentation time does not become longer visits. It becomes shorter evenings, which is a real benefit to the clinician, but it does not by itself become more #patient_experience of being heard, unless the encounter itself changes in quality. Some studies do report improvements in eye contact and patient reported engagement, which suggests that the encounter can change in quality even when it does not change in length. This is the most hopeful finding in the field: automation may improve connection not by buying more minutes but by improving the minutes that already exist. What the evidence shows on reinvestment capability. This is the least studied and most worrying dimension. If a machine drafts the note, the clinician's practice of clinical reasoning through writing may weaken over time. The note is not only a record; it is a thinking tool. There is not yet good longitudinal evidence on whether note drafting by machine produces #deskilling in reasoning, and this is a serious gap. The empathy performance question. Alongside documentation, healthcare is where machine generated empathy has been tested most directly. The Ayers et al. (2023) finding that blinded professional evaluators rated chatbot answers to patient questions as both higher quality and dramatically more empathic than physician answers is often reported as a story about machine warmth. It is at least as much a story about human depletion. The physicians in that comparison were answering questions for free, quickly, in the margins of overloaded lives. Their answers were short because their time was short. The machine's answers were long because the machine has no time constraint. What the study reveals, arguably, is not that machines are more empathic than doctors but that our systems have made doctors write like machines while machines have learned to write like the doctors we wish we had. This reading is supported by Kerasidou et al. (2021), who argue that empathy in healthcare is a systemic property, not merely a personal one. If that is right, then the correct response to the Ayers finding is not to replace physician communication with machine communication. It is to ask why physicians are answering patients in fifty words at midnight. Human machine collaboration. A more promising model appears in Sharma, Lin, Miner, Atkins and Althoff (2023), who studied a system that suggested empathic phrasing to peer supporters in an online mental health support platform rather than writing the messages for them. The peer supporters, still the authors of their own messages, produced more empathic responses with the assistance than without, and the gain was largest among supporters who found it hardest to express empathy in the first place. This is #hybrid_intelligence in its most defensible form. The machine did not replace the relationship. It coached the human inside the relationship. The human remained the one who cared, and the person receiving support was still being helped by a person. 5.2 Education: marking, feedback and the meaning of being seen Teaching has its own version of the documentation crisis. Marking is time consuming, repetitive and, at scale, cognitively exhausting. It is also, in principle, one of the most intimate acts of teaching, because feedback is where a teacher demonstrates that they have actually read this particular student's particular work. Release. Automated grading works, within limits. Floden (2025) compared human and machine grading of university exams and found that the model could grade at a level that was, in some respects, comparable with human graders, while also showing differences in consistency that make unsupervised use unwise. Yavuz, Celik and Yavas Celik (2025) found high reliability for rubric based essay grading by large language models when the rubric was explicit and the task well structured. The pattern across this literature is consistent: machines do well on structured, criterion referenced tasks and poorly on creativity, originality and nuance. Protection. The question that education research has barely begun to ask is what happens to the time. If a lecturer saves twelve hours of marking, do those hours become office hours, or do they become a larger cohort? Institutional finance suggests the second. The #student_teacher_relationship is not a line item in a budget, and so it is not protected by one. Reinvestment capability. There is a specific risk in education that does not appear as sharply elsewhere. Marking is not only assessment; it is how a teacher learns who their students are. The teacher who reads sixty essays knows which student is bluffing, which is stuck, which has quietly improved and which has quietly given up. That knowledge is the foundation of relational teaching. Outsource the reading, and the teacher may retain the time but lose the knowledge, arriving at the tutorial with a free hour and nothing to say. This is #empathy_atrophy in its most concrete form: not a loss of feeling, but a loss of the informational basis on which feeling acts. The design implication is important. In education, the safest form of automation is not the one that reads the work for the teacher, but the one that removes the clerical scaffolding around the work: the spreadsheet, the mark entry, the plagiarism check, the deadline chasing, the attendance return. 5.3 Welfare and public administration: where discretion goes to die Public services present the hardest case, because here automation touches decisions, not just tasks, and because the people affected are usually those with the least power to object. Considine et al. (2022), examining machine bureaucracies in welfare to work programmes, question whether automated systems can grasp the messy reality of what a claimant actually needs. Ranerup and Henriksen (2022) show how automated decision making in social services redistributes agency between the caseworker and the system. Roehl and Crompvoets (2023) place this within a debate about whether algorithmic bureaucracy can satisfy the requirements of good administration at all. The mechanism by which automation damages empathy in this domain is not mysterious. It works like this. First, a scoring system is introduced to help caseworkers prioritise. It is presented as decision support, not decision making. Second, the caseworker learns that deviating from the score requires written justification, while following it requires nothing. The cost of empathy has been raised and the cost of compliance has been lowered. Third, deviations decline. Not because caseworkers stopped caring, but because caring now carries an administrative penalty. Fourth, the system's outputs begin to look increasingly accurate, because behaviour has adapted to the system. Fifth, the remaining discretion is formally withdrawn, on the evidence that nobody was using it. This is the quiet death of #digital_discretion, and it happens without any single actor intending it. It is a structural outcome, not a moral failing. Busuioc (2021) identifies the accompanying #accountability problem: when the decision emerges from a system, the citizen who wants to know why has nobody to ask. Grimmelikhuijsen (2023) shows that the perceived trustworthiness of automated decisions depends on whether people can see the reasoning, which means that opacity is not a neutral technical property but a direct input into whether citizens feel treated as persons. There is also the burden displacement problem identified by Madsen, Lindgren and Melin (2022). When services move to self service portals, the citizen becomes the caseworker: uploading, formatting, chasing, resubmitting. For a confident, literate, well resourced citizen this is a minor irritation. For a person in crisis, with poor literacy, unstable housing and a failing phone, it is an exclusion mechanism. #digital_welfare that removes staff without removing complexity does not reduce administrative burden. It privatises it, and it privatises it onto exactly the people least able to bear it. Schiff, Schiff and Pierson (2022) would classify this as a public value failure: the system meets its efficiency targets while failing the purpose that justified its existence. The empathy loss here is not the loss of a warm feeling. It is the loss of the human being who could have said: I can see this form is impossible for you, sit down, we will do it together. 5.4 The workplace: managing people by dashboard Zhang et al. (2025) synthesise the algorithmic management literature and describe a set of well documented risks: surveillance, bias, deskilling, dehumanisation and alienation. Parent Rocheleau et al. (2024) provide a validated measure of how workers experience these systems, which signals that the phenomenon is now stable enough to be measured routinely. The relevance to empathy operates on two levels. At the first level, algorithmic management affects the empathy that managers extend to workers. A manager who sees a person sees context: the bereavement, the sick child, the difficult client. A manager who sees a dashboard sees a number that is below target. #workforce_wellbeing suffers not because managers become cruel but because the interface removes the information on which mercy depends. Reducing a person to a metric is not merely an epistemic simplification. It is a moral one, because it deletes exactly the details that would have made a compassionate exception seem reasonable. At the second level, algorithmic management affects the empathy that workers can extend to clients. #frontline_workers who are timed, tracked and scored on handling time will end conversations that ought to continue. A care worker paid by the visit and routed by an optimisation engine cannot sit for ten extra minutes with a frightened elderly client, because those ten minutes are a variance to be explained. The system has not forbidden compassion. It has simply made compassion expensive and made haste free. That is enough. This is the clearest illustration of why the individual virtue framing of empathy is inadequate. You cannot solve this with kindness training. The worker's kindness is not the binding constraint. The routing algorithm is. 5.5 Customer service: the frontier of relational automation Customer service is where relational automation is furthest advanced, and it therefore offers a preview of what happens when the encounter itself is handed over. The evidence here is genuinely double edged. Automated systems answer instantly, at any hour, without irritation, and they never make the customer feel stupid for asking. For routine matters this is not merely acceptable, it is better than the human alternative, and it is better precisely because it removes the human capacity for contempt. The failure mode is different. It appears at the exact moment when the situation stops being routine. The bereaved spouse trying to close an account, the person disputing a charge they cannot afford, the customer whose problem does not fit any of the eleven menu options: these are the cases where connection is needed and where automated systems fail, often by looping. The harm is not simply that the problem goes unsolved. The harm is the experience of speaking clearly, at length, about something that matters, and being answered by something that has not understood and cannot be made to understand. That experience communicates a message about the person's worth, and the message is that they are not worth a person. This suggests a design rule of general application, developed further in Section 7: automate the routine, but build a fast, visible, unpenalised route to a human for the non routine, and make the route to the human easier than the route around it. 5.6 Cross domain synthesis Reading the five domains together produces a pattern that is more orderly than any single field would suggest. The domains differ sharply in which of the three conditions fails. In healthcare, release is real but modest, protection almost always fails, and reinvestment capability is untested. In education, release is real for structured tasks, protection is entirely unaddressed, and reinvestment capability is actively threatened because the automated task, reading student work, is itself the source of relational knowledge. In welfare administration, release is often illusory because the burden is displaced rather than removed, protection is irrelevant because the problem is not time but authority, and reinvestment capability is destroyed directly by the narrowing of discretion. In the workplace, the technology is not aimed at freeing time at all; it is aimed at extracting it, and the empathy loss is a design outcome rather than a side effect. In customer service, the encounter itself has been substituted, and the harm concentrates entirely at the boundary between routine and non routine cases. Three general lessons follow. First, the risk to connection rises sharply as one moves from task automation to decision automation to relational automation. Almost all the defensible gains sit in the first category. Almost all the serious harms sit in the second and third. This is a useful heuristic for anyone asked to evaluate a proposed system. Second, the harm rarely arrives as an announced decision. It arrives as drift. No hospital board votes to shorten the human relationship. No ministry announces that caseworkers may no longer be flexible. The change happens through a hundred small, individually reasonable adjustments to what is easy and what is costly, and the aggregate is only visible in retrospect. This is why the protective commitments recommended in Section 7 must be made in advance and in writing. Afterwards, there will be no moment at which anyone can object, because there will have been no single event to object to. Third, the people who lose most are consistently those with the least capacity to advocate for themselves: the patient who cannot articulate their symptoms, the student who will not ask for help, the claimant whose life does not fit the form, the customer whose problem is not on the menu. Automation, in every domain reviewed, works best for the confident and the typical. It fails at the edges. And the edges are precisely where empathy was always most needed and where its absence does the most damage. 6. Discussion: Six Mechanisms Pulling the domains together, six mechanisms explain how automation of administrative work changes empathy and connection. 6.1 The reabsorption mechanism Saved time flows to whoever has the power to claim it. Professionals rarely have that power. In the absence of an explicit protective commitment, capacity gains become volume gains. This is the single most important reason why the time dividend argument fails in practice, and it is an institutional fact, not a technological one. Any organisation that adopts automation while simultaneously raising throughput targets has answered the question of what the time is for, whatever its mission statement says. 6.2 The discretion narrowing mechanism Decision automation raises the cost of the compassionate exception and lowers the cost of compliance. Over time, professionals stop making exceptions, and the capacity to make them atrophies. Because the change is gradual and each individual step is defensible, there is rarely a moment at which anyone can object. 6.3 The empathy atrophy mechanism Empathy is a practice, and practices decay when unused. If the machine drafts the difficult letter, the professional does not practise composing the difficult letter. If the machine reads the essays, the teacher does not build a picture of the student. The concern here is not that automation makes people callous. It is that automation removes the daily repetitions through which relational skill is maintained, in the same way that navigation software has, for many drivers, quietly removed the practice of knowing where they are. 6.4 The disclosure paradox Machine generated messages are often rated as warmer than human ones, until the reader learns that a machine wrote them, at which point the felt sense of being heard drops (Yin, Jia, & Wakslak, 2024). This creates a genuine ethical trap. #disclosure reduces the benefit; concealment obtains the benefit through deception and destroys #trust if discovered. There is no clever escape from this trap at the level of message design. The only stable resolution is institutional: be transparent about what is automated, and ensure that what is automated is not the part where being cared for matters. Notably, Ovsyannikova et al. (2025) found that AI responses were still rated as more compassionate even when the source was disclosed, which complicates the picture considerably. The reconciliation may be that disclosure damages the sense of relationship more than it damages the assessment of message quality. A reader can judge a message to be good and still feel that nobody was there. Zohar, Bloom and Inzlicht (2026) capture this in their argument against frictionless AI: part of what makes care valuable is that it cost someone something. 6.5 The #empathy_washing mechanism Organisations increasingly deploy warm, apologetic, emotionally fluent automated communication as a substitute for actually changing the conditions that produce distress. The insurer that sends a beautifully empathic automated rejection has not become kinder. It has become better at appearing kind while rejecting. This is a serious and under examined risk, because emotionally fluent language models make it cheap to perform care at exactly the moment when care is being withdrawn. The performance of empathy can become a substitute for the provision of it, and the better the performance becomes, the more effectively it conceals the substitution. 6.6 The bandwidth mechanism, and why it gives grounds for hope Not every mechanism is negative. The clearest positive finding across the healthcare evidence is that removing simultaneous cognitive load improves the quality of presence even when it does not increase its quantity. A clinician not typing is a clinician looking. A teacher not chasing attendance returns is a teacher available at the door after class. This mechanism does not depend on the institution protecting time. It works within the existing encounter, which makes it robust to reabsorption. The strategic implication is significant. Institutions and professionals seeking an #empathy_dividend should prioritise automation that reduces load during the encounter over automation that saves time around it, because the first kind of benefit is much harder for the institution to take back. 7. Toward Design: Principles for Preserving Connection The following principles follow from the framework and the evidence. They are addressed to institutions, to professionals, and to students who will inherit these systems. 7.1 Automate the clerical, not the relational Draw an explicit line. On one side put transcription, form filling, coding, scheduling, data entry, compliance reporting, mark entry, routing, and record retrieval. Automate these aggressively. On the other side put the delivery of bad news, the assessment of a person's situation, the exercise of an exception, the response to distress, and the first contact with someone in crisis. Do not automate these, and do not let them be automated by accident through the slow accumulation of decision support. The test is simple. If the task's purpose is to produce a record, automate it. If the task's purpose is to reach a person, do not. 7.2 Protect the dividend explicitly Any automation programme in a caring or public facing profession should be accompanied by a written, auditable commitment about what happens to the released time. In the absence of such a commitment, the time will be reabsorbed. The commitment should be specific: consultation lengths will not be shortened; caseloads will not rise for eighteen months; office hours will increase by a stated amount. Vague pledges to focus on what matters most are worth nothing, because they are unfalsifiable. This is the single highest leverage intervention available, and it is not technical. It is a matter of #policy_design and industrial relations. 7.3 Keep the human in the loop where it counts, and be honest about where it does not #human_in_the_loop has become a comforting phrase that often means very little. A human who rubber stamps two hundred algorithmic recommendations an hour is not meaningfully in the loop. Genuine human oversight requires that the human has the time to disagree, the information to disagree intelligently, and the institutional safety to disagree without penalty. If those three conditions are absent, the oversight is decorative and should be described as such. Practical mechanisms include requiring no justification for deviating from a recommendation while requiring justification for accepting a recommendation in a flagged high stakes case, thereby reversing the incentive gradient that kills discretion. 7.4 Build the warm handover Every automated system that interacts with the public should have a visible, fast, low friction route to a human being, and reaching that human should never be presented as a failure of the user. This is the #warm_handover principle. The handover should carry context, so that the person does not have to repeat their story, because repetition of a painful story to a fresh listener is itself a harm. The design metric that matters is not deflection rate. Deflection rate is the number of people prevented from reaching a human, and optimising for it is optimising against connection. The metric that matters is resolution with #dignity. 7.5 Practise disclosure as a default Tell people when a message was generated by a machine. Yes, this will reduce the warmth they feel. That reduction is not a bug in the disclosure; it is accurate information about their situation, and they are entitled to it. An institution that relies on concealment to generate perceived warmth is running a deception, and deception is a poor foundation for #trust. The corollary is that if disclosure would make a communication feel cold, the communication probably should not have been automated in the first place. Disclosure is therefore a useful design test, not merely an ethical duty. 7.6 Preserve the practice Deliberately retain some of the work that automation could take, in order to keep the skill alive. Clinicians should still write some notes. Teachers should still read a sample of scripts themselves, chosen not at random but including the students they know least well. Managers should still meet the people whose numbers look bad. This is not nostalgia or inefficiency. It is maintenance of the informational and relational base on which good judgment depends. Aviation understood this decades ago: pilots hand fly the aircraft periodically, precisely because automation is good, so that when the automation fails there is still a pilot. 7.7 Measure what you claim to value Institutions measure throughput because throughput is easy to measure. If connection is genuinely a goal, it must be measured too, however imperfectly: patient reported feeling of being heard, student reported sense of being known, citizen reported experience of being treated as a person. Measures shape behaviour. An organisation that automates while measuring only efficiency has decided, in the only language institutions actually understand, that connection does not count. 7.8 Teach relational competence as a technical skill For students, and for the universities that teach them, the implication is that #ai_literacy and relational skill are not separate curricula. The professional of the coming decades will need to know what these systems can and cannot do, how to interrogate a recommendation, when to override it, and how to remain fully present with a person while a machine listens in the background. That last skill is genuinely new, and it is not currently taught anywhere. 7.9 A note addressed to students Most readers of this article will not be the people who choose which systems their future employer buys. They will be the people who have to work inside those systems, and they will arrive at a workplace where the tools are already installed and the assumptions already fixed. Three practical stances are worth carrying into that situation. Be precise about what the tool is doing. When someone says a system is helping with admin, ask whether it is writing records, shaping decisions, or replacing conversations. These are different, and the word admin is often used to smuggle the second and third past scrutiny. Ask where the time went. When an employer introduces a tool on the promise that it will free you to focus on people, note the promise, and then check, six months later, whether your caseload rose. That single question, asked persistently and out loud, is more useful than any amount of ethical theory. Keep your hand in. Do some of the work the machine could do for you, deliberately, and choose which parts to keep on the basis of what they teach you about the people you serve. The skill you do not practise is the skill you will not have on the day it matters, and it will matter on a day you did not schedule. 8. Limitations and Future Research This article has several limitations that should temper its conclusions. The evidence base is unbalanced. Healthcare dominates, because healthcare has money, research infrastructure and a strong tradition of trial based evaluation. Social work, further education, immigration casework, housing and small scale customer service are far less studied, and they may be precisely the domains where the risks are greatest, because they combine high relational stakes with low institutional resources and vulnerable service users. The evidence is also geographically narrow. Almost all the studies cited come from high income countries. How these dynamics play out in systems with very different staffing ratios, digital infrastructure and cultural expectations of professional relationships is largely unknown, and it should not be assumed that the findings transfer. Timeframes are short. Most studies of ambient documentation and automated grading measure outcomes over weeks or months. The mechanisms that concern this article most, particularly #empathy_atrophy and #deskilling, operate over years. There is currently no good longitudinal evidence on whether a clinician who has never written their own notes reasons differently after a decade, or whether a teacher who has never marked reads students differently after five years. These are the studies the field most urgently needs. The construct of empathy remains contested, and studies measure very different things under the same word. A study measuring third party ratings of message warmth is not measuring the same phenomenon as a study measuring whether a patient felt understood, which is not the same as a study measuring whether a clinician experienced concern. Comparing findings across these measures, as this article has had to do, is a real methodological weakness. Finally, the framework offered here is conceptual and has not been tested. It generates testable predictions, and this is the appropriate research agenda: The reabsorption prediction. In institutions that adopt task automation without an explicit protective commitment, caseload or throughput will rise within twenty four months, and measures of relational quality will not improve. This is directly testable with a comparison of institutions that did and did not make such commitments. The bandwidth prediction. Automation that removes cognitive load during the encounter will improve relational measures even where it does not increase encounter length, while automation that saves time outside the encounter will not, absent protection. The atrophy prediction. Over multi year horizons, professionals with heavy exposure to decision automation will show reduced willingness and reduced ability to make justified exceptions, relative to matched controls. The disclosure prediction. Disclosure will reduce perceived relational value more than it reduces perceived message quality, and this divergence will be larger in high stakes, ongoing relationships than in one off transactional contacts. Each of these is answerable with existing methods. None of them requires waiting for better technology. 9. Conclusion The argument that automation will restore human connection by clearing away the paperwork is not false. It is incomplete, and its incompleteness is dangerous, because it allows institutions to claim a relational benefit they have not actually paid for. The evidence reviewed here supports four conclusions. Automation does reduce some administrative load, and the most reliable benefit is cognitive rather than temporal. The clinician who is not typing while listening is more present, and this is a genuine and durable gain. Freed time does not become relational time on its own. It becomes whatever the institution's incentives make it. Without an explicit, auditable commitment to protect it, released time is reabsorbed as capacity, and the professional is left in exactly the same condition of hurry as before, now with a machine watching. Machines can generate language that people experience as warm, responsive and even compassionate, sometimes more so than the language of exhausted professionals. This should be read less as a triumph of machine empathy than as an indictment of the conditions under which human professionals are made to communicate. The right response is not to conclude that empathy has been solved. It is to ask why the humans were writing in fifty words at midnight. And the value of empathy depends, at least in part, on its source. People feel less heard when they learn a machine wrote the message, not because the words changed but because the relationship did. Being understood by a system that cannot care is not the same as being understood by a person who chose to. The friction, the cost, the fact that someone with limited time and their own troubles turned towards you anyway, is not an inefficiency in human care. It is much of what human care is. The task ahead is therefore not to decide whether to automate. That decision is already being made, in every hospital, ministry, school and call centre, mostly without the people affected being consulted. The task is to insist that automation be pointed at the shell and not the core, and to insist, loudly and in writing, on what the saved time is for. Machines should take the forms. People should keep the faces. If institutions are unwilling to state clearly which of these they intend, then the promise of a more human future is not a plan. It is an advertisement. Hashtags #empathy #automation #algorithms #human_connection #administrative_burden #digital_discretion #algorithmic_management #ai_ethics #future_of_work #relational_care #compassion #public_administration #healthcare #education #ai_literacy References Ayers, J. W., Poliak, A., Dredze, M., Leas, E. C., Zhu, Z., Kelley, J. B., Faix, D. J., Goodman, A. M., Longhurst, C. A., Hogarth, M., & Smith, D. M. (2023). Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Internal Medicine, 183(6), 589 to 596. https://doi.org/10.1001/jamainternmed.2023.1838 Busuioc, M. (2021). Accountable artificial intelligence: Holding algorithms to account. Public Administration Review, 81(5), 825 to 836. https://doi.org/10.1111/puar.13293 Considine, M., McGann, M., Ball, S., & Nguyen, P. (2022). Can robots understand welfare? Exploring machine bureaucracies in welfare to work. Journal of Social Policy, 51(3), 519 to 534. https://doi.org/10.1017/S0047279421000519 Floden, J. (2025). Grading exams using large language models: A comparison between human and AI grading of exams in higher education using ChatGPT. British Educational Research Journal, 51(1), 201 to 224. https://doi.org/10.1002/berj.4069 Grimmelikhuijsen, S. (2023). Explaining why the computer says no: Algorithmic transparency affects the perceived trustworthiness of automated decision making. Public Administration Review, 83(2), 241 to 262. https://doi.org/10.1111/puar.13483 Inzlicht, M., Cameron, C. D., D'Cruz, J., & Bloom, P. (2024). In praise of empathic AI. Trends in Cognitive Sciences, 28(2), 89 to 91. https://doi.org/10.1016/j.tics.2023.12.003 Kerasidou, A., Baeroe, K., Berger, Z., & Caruso Brown, A. E. (2021). The need for empathetic healthcare systems. Journal of Medical Ethics, 47(12), e27. https://doi.org/10.1136/medethics-2019-105921 Lukac, P. J., Turner, W., Vangala, S., Chin, A. T., Khalili, J., Shih, Y. T., Sarkisian, C., Cheng, E. M., & Mafi, J. N. (2025). Ambient AI scribes in clinical practice: A randomized trial. NEJM AI, 2(12). https://doi.org/10.1056/aioa2501000 Madsen, C. O., Lindgren, I., & Melin, U. (2022). The accidental caseworker: How digital self service influences citizens' administrative burden. Government Information Quarterly, 39(1), 101653. https://doi.org/10.1016/j.giq.2021.101653 Montemayor, C., Halpern, J., & Fairweather, A. (2022). In principle obstacles for empathic AI: Why we can't replace human empathy in healthcare. AI and Society, 37(4), 1353 to 1359. https://doi.org/10.1007/s00146-021-01230-z Ovsyannikova, D., de Mello, V. O., & Inzlicht, M. (2025). Third party evaluators perceive AI as more compassionate than expert humans. Communications Psychology, 3, Article 4. https://doi.org/10.1038/s44271-024-00182-6 Parent Rocheleau, X., Parker, S. K., Bujold, A., & Gaudet, M. C. (2024). Creation of the algorithmic management questionnaire: A six phase scale development process. Human Resource Management, 63(1), 25 to 44. https://doi.org/10.1002/hrm.22185 Perry, A. (2023). AI will never convey the essence of human empathy. Nature Human Behaviour, 7(11), 1808 to 1809. https://doi.org/10.1038/s41562-023-01675-w Ranerup, A., & Henriksen, H. Z. (2022). Digital discretion: Unpacking human and technological agency in automated decision making in Sweden's social services. Social Science Computer Review, 40(2), 445 to 461. https://doi.org/10.1177/0894439320980434 Roehl, U. B. U., & Crompvoets, J. (2023). Inside algorithmic bureaucracy: Disentangling automated decision making and good administration. Public Policy and Administration. Advance online publication. https://doi.org/10.1177/09520767231197801 Rubin, M., Arnon, H., Huppert, J. D., & Perry, A. (2025). Comparing the value of perceived human versus AI generated empathy. Nature Human Behaviour, 9, 2345 to 2359. https://doi.org/10.1038/s41562-025-02247-w Schiff, D. S., Schiff, K. J., & Pierson, P. (2022). Assessing public value failure in government adoption of artificial intelligence. Public Administration, 100(3), 653 to 673. https://doi.org/10.1111/padm.12742 Sharma, A., Lin, I. W., Miner, A. S., Atkins, D. C., & Althoff, T. (2023). Human AI collaboration enables more empathic conversations in text based peer to peer mental health support. Nature Machine Intelligence, 5(1), 46 to 57. https://doi.org/10.1038/s42256-022-00593-2 Sorin, V., Brin, D., Barash, Y., Konen, E., Charney, A., Nadkarni, G., & Klang, E. (2024). Large language models and empathy: Systematic review. Journal of Medical Internet Research, 26, e52597. https://doi.org/10.2196/52597 Yavuz, F., Celik, O., & Yavas Celik, G. (2025). Utilizing large language models for EFL essay grading: An examination of reliability and validity in rubric based assessments. British Journal of Educational Technology, 56(1), 150 to 166. https://doi.org/10.1111/bjet.13494 Yin, Y., Jia, N., & Wakslak, C. J. (2024). AI can help people feel heard, but an AI label diminishes this impact. Proceedings of the National Academy of Sciences, 121(14), e2319112121. https://doi.org/10.1073/pnas.2319112121 Zhang, M. M., Cooke, F. L., Ahlstrom, D., & McNeil, N. (2025). The rise of algorithmic management and implications for work and organisations. New Technology, Work and Employment, 40(3), 659 to 671. https://doi.org/10.1111/ntwe.12343 Zohar, E., Bloom, P., & Inzlicht, M. (2026). Against frictionless AI. Communications Psychology, 4, Article 39.