Dawnload the Book (PDF): Introduction In 1516 a London lawyer named Thomas More published a small book in Latin about an island that did not exist. Its inhabitants held all property in common, worked six hours a day, rotated their houses by lot, and used gold to make chamber pots and the chains of their slaves. More called the island Utopia, a coined word that plays on two Greek roots: ou-topos, "no place," and eu-topos, "good place." The pun has outlived almost everything else about its author. Five centuries later the word, and its dark twin, dystopia, are part of ordinary political speech. Commentators call a proposed tax "utopian" to mean it is naive; they call a surveillance law "Orwellian" or a consumer trend "a bit Brave New World" to mean that it is sinister. We reach for imagined societies whenever we want to say that the present is going wrong. This book is about that habit, and about the literature that trained us in it. It follows a single line of development: from More's island, through the socialist utopias of the late nineteenth century, into the great dystopias of the twentieth, through the self-questioning utopias of the 1970s, and out into the climate fiction of our own time. It is not a survey of everything ever written in these modes. The field is enormous, and a list of titles would teach very little. Instead the book pursues one argument, and uses a limited number of works, read closely, to make it. The argument Every imagined society is a diagnosis of the present disguised as a place or a time. A utopia isolates the institution its author holds most responsible for human misery, whether that is private property, wage labour, the patriarchal family, the unaccountable state, the machine or the carbon economy, and imagines it abolished or transformed. A dystopia isolates the same kind of institution and imagines it perfected. Both work by what the critic Darko Suvin called estrangement: they make the familiar strange so that we can see it as something made, and therefore as something that might be unmade. But the diagnosis is only half of the story. The other half, and the one this book treats as decisive, is the question of the bridge: how do we get from here to there? Every utopia must say something, even if only by silence, about the route from the reader's world to the better one. More cut his island off from the mainland with a trench. Edward Bellamy put his hero to sleep for more than a century. William Morris, alone among the Victorian utopians, wrote a chapter describing the revolution. The great dystopias of the twentieth century are, among other things, stories in which the bridge has collapsed: the rebel is caught, the heretic is cured, the wall holds. And the most interesting speculative writing of the last fifty years, from Ursula K. Le Guin to Kim Stanley Robinson, has turned back to the bridge itself, treating the process of change, rather than its endpoint, as the proper subject of political imagination. The claim of this book, then, is that the political force of utopian and dystopian writing depends less on the picture of the good or bad society than on what the work does with transition. A utopia with no bridge is either a satire or a daydream. A dystopia with no exit risks teaching resignation. The genre's history is the history of writers trying, with changing materials and changing fears, to imagine not only a different world but a way into it. Terms A few definitions will keep the argument clear. By utopia this book means a detailed description of an imagined society that its author presents as considerably better than the one the reader lives in. By dystopia it means a description of an imagined society presented as considerably worse. Both are forms of speculative literature, a broad term covering any fiction whose world differs from ours in a deliberate, reasoned way, and which includes most science fiction. The word anti-utopia is sometimes used for works that attack utopian thinking itself, arguing that the pursuit of perfection breeds tyranny. Not every dystopia is an anti-utopia, and the difference matters: George Orwell's Nineteen Eighty-Four warns against a particular political tendency, not against the hope for a better world. Ideology is used in the sense given to it by the sociologist Karl Mannheim in Ideology and Utopia (1929). For Mannheim, ideologies are systems of ideas that tend to stabilise the existing order, while utopias are ideas that transcend that order and push against it. The distinction is not always clean. Utopias can become ideologies once their believers hold power, and dystopias can serve the status quo by suggesting that any alternative would be worse. Much of what follows is an attempt to read particular works for which way they push. Finally, critical utopia and critical dystopia are terms from the scholars Tom Moylan, Raffaella Baccolini and Lyman Tower Sargent. They describe works that keep an awareness of their own limits: utopias that show conflict and imperfection within the good society, and dystopias that retain a flicker of possible resistance. These terms will become important in the later chapters. The route The book runs in roughly chronological order, because the argument is historical: the diagnosis changes as the anxieties of each period change. Chapter 1 reads More's Utopia in the context of Tudor enclosure and humanist political thought, and asks why its bridge is a moat. Chapter 2 follows utopia's migration from distant islands into the future, a shift that made it a program rather than a thought experiment, through Bellamy, Morris and Charlotte Perkins Gilman. Chapter 3 traces the turn toward dystopia around the start of the twentieth century, in H. G. Wells, E. M. Forster and Yevgeny Zamyatin, as the machinery of progress began to look like a cage. Chapter 4 is devoted to the two dystopias that set the terms of the century's political vocabulary, Aldous Huxley's Brave New World and Orwell's Nineteen Eighty-Four, and to the long argument between them about whether power works through pain or through pleasure. Chapter 5 moves to the postwar decades, when affluence, advertising, automation and behavioural science supplied new nightmares and at least one sincere utopia. Chapter 6 examines the feminist, anarchist and ecological utopias of the 1970s, which rescued utopian writing by making it self-critical. Chapter 7 considers the dystopias of the neoliberal era, when it became, as one well-known formulation puts it, easier to imagine the end of the world than the end of capitalism. Chapter 8 turns to climate speculation, the field in which the question of the bridge has returned with the greatest urgency. The conclusion draws out what this history implies for how we read, and use, imagined societies now. Readers will find a handful of tables along the way, each comparing a group of works on the same few points. They are aids, not arguments. The arguments are in the prose. Why it matters It would be possible to treat all of this as a chapter in literary history, of interest mainly to students of the novel. That would be a mistake. Imagined societies do real political work. Bellamy's Looking Backward inspired a network of political clubs across the United States within a few years of its publication. Nineteen Eighty-Four supplied the vocabulary with which several generations have argued about state power and truth. The red cloaks and white bonnets of Margaret Atwood's The Handmaid's Tale have been worn by protesters outside legislatures on several continents. When a society cannot picture an alternative to itself, it loses one of its main tools for self-correction; when it can picture only catastrophe, it may drift toward the fatalism that makes catastrophe more likely. We live, by most accounts, in a period of low utopian energy and high dystopian supply. Streaming services, bookshops and cinemas are full of collapsed futures. The question this book finally asks is what that imbalance tells us, and whether the tradition that began on More's island still has something to offer a century facing a problem, climate change, that cannot be solved by imagining a better society somewhere else. The answer, it will argue, lies in the part of the tradition that was always hardest to write: the bridge. Chapter 1. More's Island and the Politics of Nowhere Thomas More was thirty-eight when Utopia appeared in Leuven in late 1516, printed under the supervision of his friend Erasmus. He was a successful London lawyer, an undersheriff of the city, and a rising figure who would within a few years enter the service of Henry VIII, eventually becoming Lord Chancellor before his execution in 1535 for refusing to accept the king's supremacy over the English church. The book he wrote in his late thirties is short, witty, learned and deeply strange. It is also the founding text of a genre, and the choices More made in it, about where to put the good society, who describes it and how seriously we are meant to take it, have shaped every utopia since. The most important of those choices is easy to overlook because it is so physical. Utopia is an island. It was not always one. According to the book's account, the land was once joined to a continent, until a conqueror named Utopus, having subdued its rough inhabitants, ordered a channel fifteen miles wide to be dug across the isthmus, cutting the new society off from its neighbours. The good society, in other words, begins with an act of separation. That trench is the first answer in literature to the question this book is concerned with: how do you get from the world as it is to a better one? More's answer is that you don't. You go somewhere else, or you dig a moat. The two books of Utopia Utopia is divided into two parts, and they are very different. Book II, which More seems to have written first, during a diplomatic mission to Flanders in 1515, is the description of the island itself. Book I, written afterwards in London, is a dialogue set in Antwerp among three characters: More himself, his real friend the humanist Peter Giles, and a fictional traveller called Raphael Hythlodaeus, a Portuguese sailor who claims to have voyaged with Amerigo Vespucci and to have lived for five years among the Utopians. Book I is where the diagnosis happens. Before Hythlodaeus describes the island, he describes England, and the picture is savage. He recounts a dinner at the household of Cardinal John Morton, where a lawyer praises the English practice of hanging thieves, sometimes twenty at a time on a single gallows, and wonders why theft persists nonetheless. Hythlodaeus answers that punishment cannot cure a crime that poverty compels. Men steal because they have no other way to live, and they have no other way to live because the land that once fed them is being taken away. Landlords, seeing greater profit in wool than in grain, were enclosing common fields and turning arable land to pasture, evicting the tenants who had farmed it. In the book's most famous image, the sheep, usually so meek and small-eating, have become so greedy and wild that they devour men themselves. This is not a fantasy. Enclosure was a live political issue in early Tudor England, the subject of royal commissions and statutes in the decades around the book's composition, and historians still argue about how much depopulation it caused. What matters for our purposes is the structure of More's argument. He identifies a specific institution, the private and profit-seeking use of land, and traces from it a chain of consequences: eviction, vagrancy, theft, execution. The social order that hangs thieves is the same order that manufactures them. And that diagnosis leads, by the end of Book I, to the radical claim that Hythlodaeus puts forward and that the character of More resists. As long as there is private property, and money is the measure of all things, there can be no justice and no lasting prosperity. The only cure is to abolish property altogether. Book II then shows what that cure looks like. There are fifty-four cities on the island, all built to much the same plan, each with its surrounding farmland. Citizens rotate between town and country, spending two years at a time on the farms. Everyone works, including most of those who in Europe would be idle, such as the nobility, their armed retainers and a large portion of the clergy, and because everyone works, six hours a day is enough to supply all needs. Goods are stored in common warehouses from which each household takes what it requires without payment. Houses have doors that open at a touch and are exchanged by lot every ten years, so that no one comes to think of a building as his own. Gold and silver, useless for any practical purpose, are used for chamber pots, for the chains of slaves and for the ornaments of children, who put them aside as they grow up, much as European children put aside toys. When ambassadors from a distant land arrive in Utopia dressed in gold chains and jewels, the Utopians take the humblest of their servants for their masters. Utopia's shadows A reader who comes to Utopia expecting a simple picture of paradise is soon unsettled. The island has slaves, drawn partly from criminals, both its own and those purchased from other countries, and partly from foreign labourers who choose servitude in Utopia over poverty at home. It practises a form of colonialism, sending surplus citizens to settle land on the mainland that its native inhabitants leave uncultivated, and making war on those who resist. It hires mercenaries to fight its wars and prefers to win by bribery and assassination rather than open battle. Travel between cities requires permission. Premarital sex is severely punished. Religious toleration is extensive by the standards of 1516, but those who deny the immortality of the soul or divine providence are barred from public office. Private conversation about public affairs outside the council or the popular assembly is a capital offence. Readers have taken these features in very different ways. Some treat them as evidence that More's ideal society was always authoritarian, an early blueprint for the regimented states that twentieth-century anti-utopians would condemn. Karl Popper, in The Open Society and Its Enemies (1945), traced the dangers of what he called utopian engineering back to Plato's Republic, More's most obvious model, and his critique has often been extended to More himself. Others see the unsettling details as signs that Utopia is not a blueprint at all but a satire, an exercise in the humanist art of arguing both sides of a question. The names support this reading. Hythlodaeus means something like "dispenser of nonsense." The capital city, Amaurot, is the "dim" or "phantom" city; its river, Anyder, is "waterless"; its chief magistrate, Ademus, is "without a people." The whole apparatus of the book announces, to anyone who knows Greek, that it is a joke. But a joke about what? Here the structure matters again. At the end of Book II the character of More, having listened politely, says that many of the Utopian customs seem to him absurd, above all the community of goods, which would destroy the nobility and magnificence that people regard as the true ornaments of a commonwealth. He adds, in a closing line that has puzzled readers ever since, that there are many things in the Utopian commonwealth that he wishes rather than expects to see in European cities. The author has placed his own name on the side of common sense, and then let common sense sound a little foolish. The book neither endorses nor rejects its island. It holds the question open. The best account of this double movement is that Utopia is a machine for estrangement. By showing a society in which gold is despised, More makes his readers notice how arbitrary their own reverence for gold is. By showing a society in which everyone works six hours, he makes visible the idleness of the European aristocracy and the overwork of its labourers. The island does not need to be desirable in every detail to perform this function. It needs only to be coherent enough that its differences from Europe read as choices rather than accidents. Once a reader has seen that a society could organise property differently, the existing organisation of property stops looking like nature. The problem of counsel Book I contains a second argument, less often remembered but just as important for the tradition, about whether a person with radical ideas should enter politics at all. Peter Giles and the character of More urge Hythlodaeus to put his wisdom at the service of a king. Hythlodaeus refuses. Princes, he says, care more for war than for peace and more for acquiring new kingdoms than for governing the ones they have. A counsellor who proposed Utopian measures in a royal council would be laughed at or ignored, and one who stayed in office would end up endorsing the very policies he meant to oppose. The character of More replies with a defence of what he calls a more civil philosophy, one that knows its stage and adapts itself to the play in hand. If you cannot make things good, you can at least make them less bad; you should not abandon the ship in a storm because you cannot control the winds. This exchange is the first appearance in utopian literature of the problem of the bridge in explicitly political form. Hythlodaeus has the vision but no route to it; he has concluded that the existing order cannot be reformed from within, and he has no theory of how it might be overturned. The character of More has a route, gradual influence from inside the court, but it leads nowhere near Utopia. The book leaves the two positions facing each other. It is hard not to read this against More's own life: within two years of publication he had joined the royal council, and he spent the rest of his career trying to make things less bad before the winds finally broke him. The island's trench, then, is not only a geographical fact but a political confession. More could imagine the abolition of property, but he could not imagine a sequence of events by which Tudor England would get there. So the good society had to be elsewhere, founded by a conqueror in some unrecorded past, and reachable only by a traveller whose name means nonsense. The imagined world is complete; the path to it is missing. More's successors on the map For more than two centuries after More, most utopias followed his spatial model. They were located on islands, in hidden valleys or in unexplored regions, discovered by voyagers and described in reports home. The pattern fitted an age of maritime exploration in which Europeans were encountering, and conquering, societies organised very differently from their own. It also allowed authors to present radical ideas at a safe distance. Two seventeenth-century examples show how flexible the form was. Tommaso Campanella, a Dominican friar imprisoned in Naples for his part in a conspiracy against Spanish rule, wrote The City of the Sun in Italian in 1602, publishing a Latin version in 1623. His city, ruled by a priest-philosopher and three ministers representing power, wisdom and love, abolishes private property and, following Plato, the private family; officials arrange sexual unions for the sake of good breeding. The walls of the city are painted with the whole of human knowledge, so that children learn simply by walking around. Francis Bacon's New Atlantis, published in 1627, the year after his death, took a different path. Its island of Bensalem keeps private property and the patriarchal family but is organised around Salomon's House, a state-sponsored institution of natural philosophers who conduct experiments, gather knowledge from around the world and direct the application of science to human welfare. Bacon's utopia is not a critique of property but a vision of what organised research could achieve, and it became a direct inspiration for the founders of the Royal Society later in the century. Set side by side, More, Campanella and Bacon show that from the beginning utopia was not one politics but a form available to many. What they share is the method: take the institution you think matters most, whether property, the family or knowledge, and imagine it reorganised from the ground up, then show the reorganisation working. What they also share is the missing bridge. None of these societies is reached by reform or revolution. Each is found, already complete, by a traveller who has wandered off the map. Why the island matters It is tempting to treat the island setting as a mere convention, a piece of stage machinery that later writers would discard. But the spatial utopia carries a particular politics. A society located elsewhere, in the present, is a standing rebuke: it shows that people just like us, at this very moment, live otherwise, and so our arrangements cannot be excused as the only possible ones. At the same time it is politically inert. Nobody can move to Utopia. The only use of the island is to change how we see home. This is the utopia as mirror, and it has never gone away. When contemporary readers encounter a well-built imaginary society, from Le Guin's anarchist moon to the matriarchal commune of a feminist novel, part of its effect is still the one More discovered: the sudden recognition that what we took for human nature is a set of institutions. But the mirror has limits. It can show what is wrong, and it can show that another arrangement is conceivable. It cannot show how to get from one to the other. That problem, which More dramatised in the stalled argument between Hythlodaeus and his friends, would dominate the next great phase of utopian writing, when the good society moved off the map and into the future. The change was not merely one of setting. A society placed in the future is no longer a mirror but a destination. It implies that the present will turn into it, which means that someone, eventually, has to explain the turning. The trench across Utopus's isthmus could be dug once and forgotten. A gap in time has to be crossed. Chapter 2. From Island to Future: The Victorian Blueprint Sometime in the second half of the eighteenth century, utopia moved house. It left its islands and hidden valleys and took up residence in the future. The change is usually dated to Louis-Sébastien Mercier's L'An 2440, published anonymously in 1771, in which a Parisian falls asleep after an argument with an English philosopher and wakes nearly seven centuries later in a reformed Paris. The book was banned in France, widely pirated, and translated into English as Memoirs of the Year Two Thousand Five Hundred. It was not a great work of literature, but it made a decisive formal innovation. The good society was no longer somewhere else. It was somewhen else, and specifically it was here, in the reader's own city, after the passage of time. The historian Reinhart Koselleck and others have linked this shift to a broader transformation in how Europeans experienced time. Under the influence of Enlightenment ideas of progress, the future stopped being a repetition of the past and became an open field of possibility, something that human beings could shape. Once that happened, the island utopia began to look quaint. Why imagine a better society on the far side of the ocean when one could imagine the present society improving into it? The temporal utopia, sometimes called uchronia, fitted a century that believed in improvement. But the move carried a cost that took a hundred years to become fully visible. A future utopia makes an implicit promise: this is where we are going. And that promise invites a question the island never had to face. How, exactly, will we get there? The great utopias of the late nineteenth century are the first to confront that question head on, and the most interesting thing about them is how differently they answer it. Bellamy's sleeping man Edward Bellamy's Looking Backward: 2000–1887 appeared in Boston in 1888 and became one of the best-selling American books of the century. Its hero, Julian West, is a wealthy young Bostonian who suffers from insomnia and has himself put into a mesmeric trance in a sealed underground chamber. The house burns down, the mesmerist leaves town, and West sleeps undisturbed until the year 2000, when he is dug out by a physician, Dr Leete, who with his daughter Edith becomes West's guide to the new world. The Boston of 2000 has no poverty, no strikes, no political parties and almost no crime. The reason is that the economy has been consolidated into a single enterprise owned and managed by the nation. Every citizen serves in what Bellamy calls the industrial army from the age of twenty-one to forty-five, beginning with a period of general labour and then choosing a trade. Everyone receives an equal share of the national product, issued as an annual credit recorded on a card and drawn down at public stores. Unpleasant work is made attractive by shorter hours rather than higher pay. Retirement at forty-five frees citizens for leisure, culture and self-improvement. The whole system is administered by a hierarchy of officers culminating in a general-in-chief, who is the president of the nation. Bellamy's diagnosis of the Gilded Age is sharp. In one well-remembered passage, Dr Leete asks West to picture the society of 1887 as a great coach, pulled along a hilly road by the mass of humanity, while a fortunate few sit on top in comfortable seats, calling encouragement to those in harness and living in constant fear of being jolted off. The riders are not wicked. They simply cannot imagine the world arranged differently, and they believe that the difference between themselves and the pullers is somehow natural. The image captures, as well as anything in the literature of the period, the way inequality presents itself as fate. Most revealing, however, is Bellamy's account of the transition. There was no revolution. According to Dr Leete, the process of business consolidation that so alarmed Americans of the 1880s simply continued. Small firms were absorbed by great corporations; the great corporations merged into trusts; and at last, early in the twentieth century, the nation recognised that the logical conclusion of this development was a single trust owned by the people as a whole. The change came peacefully, by general agreement, once everyone saw that it made sense. The capitalist economy, in effect, evolved into its successor under its own momentum. This is a bridge of a very particular kind. It requires no conflict, no organisation, no sacrifice. History does the work. The hero, significantly, sleeps through it. For Bellamy's readers the message was both reassuring and energising: the good society was the natural destination of trends already under way, and one needed only to help the country recognise it. Within a year of publication, readers had founded the first Nationalist Club in Boston to promote Bellamy's ideas, and dozens more followed across the country. The movement fed into the Populist politics of the 1890s before fading. Bellamy wrote a sequel, Equality (1897), to answer his critics. Morris's answer One of those critics was William Morris, the designer, poet and socialist, who reviewed Looking Backward in the socialist paper Commonweal in 1889. Morris objected to almost everything about Bellamy's future. He disliked its centralisation, its military metaphors and its assumption that the good life consisted of short working hours followed by consumption. Bellamy's citizens, in Morris's reading, were still wage-earners of a kind, working at tasks they did not choose for an authority they did not control, and compensated by leisure. The goal of socialism, for Morris, was not to reduce labour to a minimum but to make labour itself pleasurable, an expression of human creativity rather than a burden to be minimised. Morris answered Bellamy in fiction. News from Nowhere, serialised in Commonweal in 1890 and published as a book in 1891, is the story of William Guest, a socialist who goes home to Hammersmith after a quarrelsome meeting and wakes to find himself in a transformed London sometime in the twenty-first century. The factories and slums are gone. The Thames is clean enough for salmon. The Houses of Parliament survive only as a storehouse for manure. There is no money, no central government in the old sense, no prisons and no formal marriage law. People work because they enjoy it, at crafts and trades they have chosen, and the decorated objects they make are given away rather than sold. Much of the book is taken up with a leisurely journey up the Thames to a haymaking, and its emotional climax is Guest's encounter with an old house near the river's upper reaches, closely modelled on Morris's own home at Kelmscott Manor. If Bellamy's utopia is a well-run corporation, Morris's is a village fête that never ends. Critics have found it static, nostalgic, even faintly medieval, and there is truth in the charge. But Morris did something that Bellamy did not. In a long chapter titled "How the Change Came," Guest's elderly informant, Hammond, recounts the history of the transition, and it is a story of conflict. Strikes spread; the workers organise; the government responds with repression, including a massacre of demonstrators in Trafalgar Square, an event Morris modelled on the real "Bloody Sunday" of November 1887, which he had witnessed. A period of civil war follows before the old order collapses. Even then, Hammond explains, it took generations for people to lose the habits of mind that capitalism had given them and to rediscover the pleasure of work. Morris's bridge, in other words, is built out of struggle, violence and slow cultural change. It is not pleasant reading, and it is not meant to be. But it is the first sustained attempt in the English utopian tradition to answer the question More left open: how does the actual world become the better one? Morris's answer is that it happens through collective political action, at great cost, and that the transformation of institutions must be followed by a transformation of desire. Gilman's separate country The third utopia to set beside Bellamy and Morris comes from the next generation and takes a different angle entirely. Charlotte Perkins Gilman, best known today for her short story "The Yellow Wall-Paper," was a Nationalist in her youth, a lecturer on economics and a prolific writer on the condition of women. Her book Women and Economics (1898) argued that women's economic dependence on men distorted both sexes and held back human development. In 1915 she serialised Herland in her own magazine, The Forerunner. Herland returns, strikingly, to the spatial model. Three American men, a sociologist named Vandyck Jennings who narrates, a romantic named Jeff and an aggressive adventurer named Terry, stumble on an isolated country somewhere in an unmapped highland, inhabited only by women. Two thousand years earlier, a combination of war and natural disaster killed or cut off the men, and the surviving women developed the ability to reproduce by parthenogenesis. Their society is organised around motherhood, understood not as a private relationship but as the collective, rational cultivation of the next generation. Education, forestry, food production and town planning are all conducted with the same careful, cooperative intelligence. There is no war, no poverty and virtually no crime. The comedy of the book comes from the men's assumptions. Terry expects a country of women to be chaotic or to need a master; Jeff expects them to be delicate and in need of chivalrous protection; Van, the scientist, is slowly educated out of his preconceptions. The Herlanders, for their part, are puzzled by everything the men tell them about their own world: that women in America are economically dependent, that motherhood is a private burden, that some babies are unwanted. The estrangement runs both ways, and Gilman uses it to expose gender as a set of social arrangements rather than natural facts. Gilman's bridge, however, is an accident. The men were removed by catastrophe, and the women's development followed from a biological miracle. As in More, the good society is reached by isolation. This is revealing. For a feminist writing in 1915, the route from a male-dominated society to an egalitarian one was hard to picture, and Gilman's fantasy removed the obstacle by removing the men. Later feminist utopias, as Chapter 6 will show, would return to this problem with more ambition. The three blueprints compared As Table 1 sets out, these three books diagnose different root problems and propose different cures, but the most revealing contrast is in the last column, the route by which the good society arrives. Table 1. Three utopias of the Progressive era compared. Work Root problem Organisation of work Route to utopia Bellamy, Looking Backward (1888) Competitive capitalism, waste Industrial army; equal credit Peaceful evolution of trusts Morris, News from Nowhere (1890) Alienated labour, ugliness Voluntary craft; no money Strikes, repression, civil war Gilman, Herland (1915) Women's dependence on men Cooperative, motherhood-centred Catastrophe and isolation The contrast between Bellamy and Morris became, in miniature, the great argument within socialism for the next century: centralised administration or decentralised self-management, reform through the logic of development or rupture through struggle. It also anticipated, remarkably closely, the arguments that later dystopias would make against utopia. The anxieties that Morris expressed about Bellamy's regimented Boston, that it would turn people into functionaries and reduce life to production and consumption, would reappear in far darker form in Zamyatin and Huxley. The Marxist objection Behind all three stands a challenge that shaped left-wing attitudes to utopia for a century. In Socialism: Utopian and Scientific, published as a pamphlet in 1880 from chapters of a longer polemic, Friedrich Engels drew a sharp line between the "utopian socialists" of the early nineteenth century, such as Henri de Saint-Simon, Charles Fourier and Robert Owen, and the "scientific socialism" he and Karl Marx claimed to have founded. The utopians, in Engels's account, had sketched attractive pictures of future societies and hoped to realise them by persuasion or by founding model communities. They had failed because they did not understand the historical forces that actually drive social change. Scientific socialism did not need a blueprint. It needed an analysis of capitalism's own contradictions, which would produce the working-class movement that would overthrow it. Marx himself was famously reluctant to describe the society that would follow capitalism, and the reluctance became a principle. To write a detailed utopia was to substitute private imagination for historical analysis, and to risk distracting the movement from the real work of organisation. For many socialists, then, Bellamy and Morris were at best useful propaganda and at worst a distraction. Yet the objection cuts both ways. A movement that refuses to picture its goal may find, on taking power, that it has no idea what to build, or that the picture it lacked has been supplied by whoever controls the state. The German philosopher Ernst Bloch, writing in the mid-twentieth century in The Principle of Hope, argued that Marxism had impoverished itself by suppressing the utopian impulse, which he saw everywhere in human culture, from daydreams and fairy tales to architecture and music, as an anticipation of what does not yet exist. The sociologist Ruth Levitas, in The Concept of Utopia (1990), later defined utopia broadly as the expression of the desire for a better way of being. On these views the utopian imagination is not a rival to political analysis but a precondition of it: one must want something before one can plan for it. What the temporal utopia changed By 1900 the future-set utopia had become a recognised genre, especially in the United States, where Bellamy's success inspired dozens of imitators and replies. Its move into time had changed the political meaning of the form. More's island was a mirror: it showed Europe its own image reversed. Bellamy's Boston and Morris's London were destinations, and destinations can be argued about in practical terms. Is centralisation efficient? Is a moneyless economy workable? Would people really work without wages? These are questions about programmes, not about moral vision alone. This practicality was a strength. It is why Looking Backward could inspire a political movement in a way that More's Utopia never did. But it was also a vulnerability. A programme can be tried, and a programme that has been tried can fail. When, in the twentieth century, states arose that claimed to be building the future that the Victorian utopians had described, whether planned economies organised on something like an industrial army or technocracies directed by experts, the utopian blueprint began to look less like a promise than a threat. The next chapter follows that reversal, as writers who had grown up on Bellamy's optimism began to imagine what his well-run nation might feel like from the inside. Chapter 3. The Machine and the Cage: The Birth of Modern Dystopia The word "dystopia" is older than the genre it now names. John Stuart Mill used a version of it in a speech to the House of Commons in 1868, attacking the government's policy on Irish land as too bad to be practicable, and coining "dys-topians" as the opposite of utopians. But the literary form did not take shape until the decades around 1900, and when it did, it came not from utopia's enemies but from inside the utopian tradition itself. The first great dystopias were written by people who had absorbed the Victorian dream of a rationally organised future and had begun to wonder what it would feel like to live there. This chapter follows that turn through three writers: H. G. Wells, who was both the most energetic utopian and the most influential dystopian of his generation; E. M. Forster, who wrote a single story in direct reply to Wells; and Yevgeny Zamyatin, a Russian naval engineer who, having lived through a revolution that promised the future, wrote the novel from which much of the twentieth century's dystopian imagination descends. What unites them is a shift in the diagnosis. For More, Bellamy and Morris, the enemy was an institution of the old order, whether property, competition or wage labour. For the new dystopians, the enemy was increasingly the very machinery that was supposed to deliver the new order: technology, centralised organisation and the application of reason to every sphere of life. Wells between hope and horror H. G. Wells's first novel, The Time Machine (1895), sends its unnamed Time Traveller to the year 802,701. He finds a humanity divided into two species. The Eloi are small, beautiful, childlike creatures who live idly above ground in the ruins of great buildings. The Morlocks are pale, apelike beings who live underground among machinery, emerge at night and, it gradually becomes clear, feed on the Eloi. The Traveller's first theory is that he has arrived in a communist paradise gone soft. His later and darker conclusion is that the two species are the descendants of the Victorian leisured class and the Victorian industrial working class, separated by centuries of social stratification into different biological kinds. The upper class, freed from all necessity, has decayed into helplessness; the lower class, confined to the depths, has become its predator. The book extrapolates a present tendency, class segregation, into biological fact, and then reverses the expected power relation. It is a dystopia of social evolution. But it is not an argument against utopia as such. Wells spent much of the next forty years writing utopias, most seriously in A Modern Utopia (1905), in which two travellers find themselves on a duplicate Earth run by a voluntary order of disciplined, ascetic administrators called the samurai. Wells believed passionately in planning, in science and in a world state, and in books such as The Shape of Things to Come (1933) he imagined the old nations being swept away by a benevolent technocratic elite. Wells's divided legacy matters here because it shows how close the two modes lay. His When the Sleeper Wakes (1899) borrows Bellamy's device almost exactly. A man falls into a trance in the 1890s and wakes two centuries later to find that, through the compound interest on his investments, he owns most of the world. But the London he wakes into is not a cooperative commonwealth. It is a vast, glass-roofed megacity in which the labouring masses, dressed in the blue canvas of a single Labour Company, toil in underground factories while an oligarchy rules in his name. The Sleeper becomes a figurehead for a popular revolt, only to find that its leader is another would-be tyrant. The same premise that in Bellamy's hands produced a peaceful end to class conflict produces in Wells's hands its technological intensification. Forster's machine E. M. Forster was a novelist of personal relations, drawing-rooms and English country houses, and he is not usually thought of as a writer of speculative fiction. But in 1909 he published in the Oxford and Cambridge Review a long story, "The Machine Stops," which he later described as a reaction to one of the earlier heavens of H. G. Wells. It is set in a future in which humanity lives underground, each person alone in a hexagonal cell, all needs supplied by a global Machine. People rarely meet in person. They communicate through a kind of video screen, exchange ideas in endless lectures delivered remotely, and regard direct experience of the surface, of their own bodies or of one another, as vulgar and faintly repellent. The Machine is worshipped, its manual treated as a sacred book. The protagonist, Vashti, is a lecturer on music who is content with her cell. Her son Kuno, who lives on the other side of the world, is not. He has found a way to the surface and discovered that people still live there. He is threatened with Homelessness, which means expulsion to the surface and presumed death. The story ends, as its title promises, with the Machine breaking down. Nobody knows how to repair it, because the knowledge has been lost in generations of reliance, and the underground civilisation dies. In the last moments Kuno tells his mother that humanity will begin again with the people on the surface. Readers today are often startled by how precisely the story anticipates features of digital life: isolation mediated by screens, the preference for secondhand ideas over direct experience, the dependence on systems no individual understands. But Forster's diagnosis is not really about technology in the narrow sense. It is about what happens to a society that has delegated all its choices to an automatic system. The Machine has no ruler. There is no villain. Its tyranny is the product of convenience and habit, sustained by the willing consent of its inhabitants. This is a new kind of nightmare, and it would return, as the next chapter will show, in Huxley. The story also has a peculiar relationship to the bridge. The good society, if there is one, lies on the other side of catastrophe. The Machine cannot be reformed; it can only stop. Kuno's hope for the surface-dwellers is a hope that the collapse will clear the ground. This pattern, in which renewal requires breakdown, would become one of the dominant shapes of speculative fiction, and it returns with particular force in the climate stories discussed in Chapter 8. Jack London and the failed revolution Before turning to Zamyatin it is worth pausing on an American book that shows the bridge collapsing in a different way. Jack London's The Iron Heel (1908) is presented as a manuscript written by Avis Everhard, the widow of a socialist leader named Ernest Everhard, describing the rise of an oligarchy in the United States in the early twentieth century. As the great trusts consolidate their power, they crush the unions, co-opt the most skilled workers into a privileged caste, and drown a popular uprising in Chicago in blood. The manuscript breaks off mid-sentence. What makes the book remarkable is its frame. The manuscript has been edited, with explanatory footnotes, by a scholar living some seven centuries later in a socialist utopia, the Brotherhood of Man, which eventually replaced the Iron Heel. The utopia exists, but only in the margins, and the reader learns that it took centuries of oligarchic rule before it arrived. London was writing in part against the optimism of Bellamy and of the socialist movement of his day. The trusts, in his account, would not peacefully evolve into public ownership. They would fight, and they would win for a very long time. The Iron Heel is a bridge story that insists the bridge is far longer, and far bloodier, than anyone had hoped. Orwell later admired its understanding that a ruling class could hold on to power through organised violence and the cultivation of a loyal elite. Zamyatin's One State Yevgeny Zamyatin was a Bolshevik before the revolution, arrested and exiled under the tsars. He was also a naval engineer who spent much of 1916 and 1917 in England supervising the construction of Russian icebreakers at Newcastle, an experience he satirised in his novella The Islanders. He returned to Russia in time for the October Revolution, and in 1920 and 1921, amid civil war, famine and the consolidation of Bolshevik power, he wrote We. The novel is set more than a thousand years in the future, in the One State, a glass city enclosed by a Green Wall that shuts out the wild nature beyond. Its citizens have numbers instead of names. The narrator, D-503, is a mathematician and the chief engineer of the Integral, a spaceship that will carry the One State's principles to other planets. Life is regulated by the Table of Hours, which fixes the moment of rising, eating, working and walking for everyone at once. Sexual relations are permitted by pink coupons, which entitle a citizen to lower the blinds of his transparent apartment for a set hour. The state is headed by the Benefactor, re-elected unanimously every year on the Day of Unanimity. Dissent is punished by the Machine of the Benefactor, which reduces the offender to a puddle of clean water. D-503 begins the novel as a true believer. His journal, which he is writing for the inhabitants of the planets the Integral will reach, is a hymn to the beauty of mathematical order. Freedom, he explains, is the source of all unhappiness; the One State has solved the problem of happiness by abolishing it. Then he meets I-330, a woman who smokes, drinks, laughs at the State and leads him beyond the Green Wall, where people covered in fur live in the forest. D-503 develops what the State diagnoses as a soul. The resistance plans to seize the Integral. It fails. The State announces a new surgical procedure, the Great Operation, which removes the imagination, and D-503 submits to it. In the final entry he calmly records watching I-330 being tortured, and expresses confidence that reason will prevail. We established the grammar of the modern dystopia. It gave later writers the numbered citizen, the transparent dwelling, the regulated sexuality, the ritual election, the leader as benevolent father, the rebellious woman who awakens the male protagonist, the zone of wildness beyond the wall and, above all, the ending in which the rebel is cured rather than killed. Orwell reviewed the book in Tribune in 1946, noting its resemblances to Brave New World, and the structure of Nineteen Eighty-Four, which he was then writing, follows Zamyatin's closely. What Zamyatin was attacking It would be easy to read We as a straightforward satire of Soviet communism, and in the Soviet Union it was read that way. It was refused publication; an English translation appeared in New York in 1924, and Zamyatin was hounded until, after a letter to Stalin, he was permitted to emigrate in 1931. He died in Paris in 1937. The novel was not published in Russia until 1988. But Zamyatin's target was broader and stranger than any one regime. The One State is a parody not only of Bolshevik collectivism but also of the whole utopian tradition, and of the industrial rationalism that Zamyatin had seen in Newcastle as much as in Petrograd. The Table of Hours is modelled explicitly on the work of Frederick Winslow Taylor, the American engineer whose time-and-motion studies promised to make every movement of the worker efficient; D-503 praises Taylor as the greatest genius of the ancients. The glass city echoes the Crystal Palace, which Dostoevsky's narrator in Notes from Underground had mocked as the symbol of a rational utopia in which human beings would be reduced to piano keys. And the One State's claim that happiness requires the abolition of freedom repeats the argument of the Grand Inquisitor in The Brothers Karamazov. Zamyatin's own essays make the underlying philosophy clear. In "On Literature, Revolution, Entropy and Other Matters" (1923) he argued that there is no final revolution, just as there is no final number. Every revolution that succeeds hardens into a new orthodoxy, a state of entropy, and must be overturned by the next heresy. The danger of utopia, on this view, is not that it is impossible but that it is conceived as an ending. A society that believes it has reached perfection must regard all further change as illness. Hence the Great Operation: in a completed world, imagination is a disease. This is the decisive shift. The Victorian utopias had worried about the bridge, how to get from the present to the good society. Zamyatin's dystopia worries about the far bank. What happens when you arrive? If the good society is a final state, then its maintenance requires stopping history, and stopping history requires stopping the people who make it. We turns the utopian destination into a prison precisely because it is a destination. From diagnosis to warning By the 1920s, then, the dominant imaginative relationship to the future had changed. In the space of a generation the machine, the plan and the organised society had passed from being the means of liberation to being the threat. Wells continued to write utopias, but the most powerful speculative works of the period were warnings. The diagnosis was no longer that a particular institution, private property or competitive markets, caused suffering, but that the very project of rational social design, whether capitalist or socialist, contained a tendency toward total control. This shift carried its own political ambiguity. A warning against the perfect state could serve many ends. It could defend individual freedom against bureaucratic socialism, as many Western readers took We to do. It could equally be read as a warning against capitalist rationalisation, as Zamyatin's use of Taylor suggests. And it could slide into a general suspicion of all attempts at collective improvement, the anti-utopianism that Popper would later systematise. Which way a given dystopia pushes depends on the details: on what it identifies as the source of control, and on whether it leaves any opening for resistance. The two most famous dystopias of the century, to which we now turn, answer those questions very differently. Hashtags: #UtopiaAndDystopia #SpeculativeLiterature #PoliticalIdeology #UtopianLiterature #DystopianLiterature #PoliticalImagination #Estrangement #ImaginedSocieties #CriticalUtopia #CriticalDystopia #AntiUtopia #ThomasMore #LookingBackward #NewsFromNowhere #Herland #TheMachineStops #We #BraveNewWorld #NineteenEightyFour #IdeologyAndUtopia #PoliticalTransition #SocialCritique #ClimateFiction #SpeculativePolitics #FutureOfUtopianThought

Download the Book (PDF): Introduction Every payment system buys something. The question that matters is whether it buys what patients actually want. A health system that pays per visit, per test and per procedure buys visits, tests and procedures. It buys them efficiently, in enormous quantity, and with a precision that would be admirable if the goods in question were the point. They are not. A patient with a worn-out knee does not want a knee replacement for its own sake. She wants to walk to the shops without pain, to sleep through the night, to climb the stairs to her grandchildren's bedroom. A man with newly diagnosed type 2 diabetes does not want quarterly appointments and a growing list of prescriptions. He wants to keep his eyesight, his kidneys and his feet, and ideally to avoid needing most of the prescriptions at all. The services are inputs. The outcomes are what the patient is buying, whether or not anyone is paying for them. Fee-for-service medicine pays for the inputs and leaves the outcomes to professional conscience. For most of the twentieth century that arrangement worked tolerably well, because the inputs were limited and the professional conscience was strong. It works much less well now. Medicine can do far more than it could, most of it is delivered by teams spread across many organisations, and the dominant burden of disease has shifted from acute episodes that a single doctor can resolve to chronic conditions that unfold over decades. In the United States, national health spending reached $5.3 trillion in 2024, or 18 percent of the economy, according to the federal government's own actuaries. Careful estimates put a quarter or so of that spending in the category of waste: care that was duplicated, uncoordinated, overpriced or simply not needed. Other countries spend less, but none has escaped the underlying tension, because every system that pays for activity eventually discovers that activity is what it gets. "Value-based healthcare" is the name that has attached itself to the effort to change what gets paid for. The phrase has been used so loosely, by so many people selling so many things, that it is in danger of meaning nothing. This booklet uses it in a precise sense. Value, here, means the health outcomes that matter to patients divided by the total cost of achieving them across the whole cycle of care. Value-based payment means any arrangement in which what a provider is paid depends, at least partly, on that ratio rather than on the volume of services supplied. Everything else, the acronyms and the demonstration projects and the consulting frameworks, is machinery for approximating that idea. The argument of this book The central claim of the chapters that follow is that payment reform succeeds to the extent that three things line up: the unit of payment matches the unit in which patients experience value; the measurement of outcomes is trustworthy enough to bear financial weight; and the time horizon and population over which a provider is accountable are long and broad enough for good care, including prevention, to pay off. When all three line up, the evidence shows that payment change can reduce spending without harming quality, and sometimes while improving it. When any one of them is misaligned, the reform either achieves little or produces new distortions that can be as damaging as the ones it was meant to cure. This is a more demanding standard than the popular story, which treats the move away from fee-for-service as a simple matter of will: stop paying for volume, start paying for value, and let providers sort out the rest. The real history of the past fifteen years is more instructive. Bundled payments for joint replacement did cut spending, but almost entirely by sending patients home instead of to nursing facilities, not by making surgery better. Accountable care organisations did save money, but the savings were modest, concentrated among physician-led groups, and repeatedly threatened by the way their spending targets were set. Pay-for-performance programmes did change what clinicians recorded, sometimes more than what they did. Patient-reported outcome measures did reveal enormous variation in how much patients actually benefit from surgery, but getting enough patients to fill in the questionnaires has proved harder than designing them. And prevention, the part of medicine every reformer invokes, remains the part that payment systems are worst at buying, because its benefits arrive years later and often land in someone else's budget. None of this means value-based payment has failed. It means it is an engineering problem rather than a slogan, and engineering problems are solved by understanding mechanisms. That is what this booklet sets out to supply. How the book is organised The first chapter examines what fee-for-service actually rewards and why its defects are structural rather than moral: no one needs to be greedy for a volume-based system to produce too much of some things and too little of others. It also gives fee-for-service its due, because a reform that forgets what the old system did well will reproduce its failures in a new form. The second chapter defines value carefully, drawing on the framework developed by Michael Porter and Elizabeth Teisberg and on the practical classification of payment models that American payers now use to track their progress. It asks what counts as an outcome, whose costs count, and why the level at which value is measured, the medical condition rather than the individual service, determines almost everything that follows. Chapters three and four take the two main instruments of reform in turn. Bundled payments pay a single price for an episode of care, such as a hip replacement and the ninety days that follow it. Accountable care organisations take responsibility for the total cost and quality of care for a defined population over a year or more. Each chapter traces the logic of the instrument, the major experiments in the United States and elsewhere, what the evaluations have found, and why the results look the way they do. Chapter five turns to measurement from the patient's side: patient-reported outcome measures, the questionnaires that ask patients directly about pain, function and quality of life. England has collected them for every NHS hip and knee replacement since 2009; Medicare is now tying hospital payment to them. The chapter explains what these instruments capture that clinical data miss, and what it takes to make them usable for payment rather than merely for research. Chapter six confronts the uncomfortable fact that every measure used for payment becomes a target, and every target invites behaviour aimed at the measure rather than the goal. Risk adjustment, patient selection, coding intensity and the design of financial benchmarks are where value-based programmes most often go wrong, and understanding them is the difference between a reform that works and one that merely redistributes money. Chapter seven addresses prevention directly. It explains why fee-for-service underpays for prevention, why most value-based models only partly fix the problem, and what payment arrangements, from primary care capitation to hospital global budgets to decade-long accountable care contracts, actually shift the economics in prevention's favour. The eighth chapter draws the threads together into practical guidance on managing the transition: how to sequence risk, how to choose between voluntary and mandatory models, what capabilities provider organisations must build, and how to protect the institutions that serve the poorest patients. The conclusion sets out what the evidence now supports, what remains genuinely unsettled, and what payers, providers and policymakers should do differently as a result. A note on scope Much of the evidence in this book comes from the United States, for a simple reason. The US has been the world's largest laboratory for payment reform, because its fragmented, activity-based system created the most pressure for change and because Medicare, through the Center for Medicare and Medicaid Innovation, has run dozens of formally evaluated experiments since 2011. But the economics are universal. England's pay-for-performance scheme in primary care, Sweden's bundled payment for joint replacement, the Netherlands' bundled contracts for chronic disease and the Spanish experience with capitated regional contracts all appear where they sharpen the argument. A reader in a tax-funded system will find that the vocabulary differs, but the problems of measurement, risk and time horizon are identical. The book is written for people who need to understand payment reform well enough to make decisions about it: clinicians moving into leadership roles, managers in hospitals and health plans, policymakers and their advisers, and students of health economics and policy. It assumes no technical training in economics. It does assume a willingness to follow a mechanism through to its consequences, because that is where the lessons of the past fifteen years are to be found. Chapter 1: What Fee-for-Service Buys The simplest way to pay a doctor is to pay for what the doctor does. A consultation has a price, a blood test has a price, an operation has a price, and the bill is the sum of the prices. This arrangement is so natural that it hardly seems like a policy choice at all. It is how we pay plumbers, lawyers and mechanics. It is how medicine was paid for most of its history, and it remains the foundation of payment in the United States, in much of continental Europe for outpatient physicians, and in private practice almost everywhere. Fee-for-service is often described as though it were a corruption of medicine, a system that turns healers into salespeople. That description is unhelpful, because it locates the problem in character rather than in structure. The trouble with fee-for-service is not that it makes doctors greedy. It is that it pays for a particular thing, the discrete service, and that thing is only loosely related to what patients need. The results follow from that mismatch even when every clinician in the system is conscientious. The economics of paying for activity In most markets the buyer decides how much to buy. Health care is different in a way that the economist Kenneth Arrow identified in a famous 1963 paper: the patient typically does not know what he needs, and relies on the seller to tell him. The doctor is both adviser and supplier. Arrow argued that the ethical codes of medicine, the trust placed in physicians, and the non-profit form of many hospitals were social institutions that had grown up precisely to manage this conflict. They are real and they matter. But they operate at the margin of a financial incentive that points in one direction. Under fee-for-service, each additional service adds to revenue, and the person deciding whether the service is needed is the person who is paid for providing it. The empirical literature on what happens as a result is large and consistent. When Medicare fees change for reasons unrelated to patient need, the quantity of services moves with them. Jeffrey Clemens and Joshua Gottlieb, studying an administrative change in how Medicare set regional fees, found that a two percent increase in payment rates led to roughly a three percent increase in the provision of care, with the largest responses in elective, discretionary services and little measurable effect on patient health. Jonathan Gruber and Maria Owings showed that when falling birth rates squeezed obstetricians' incomes in the 1970s and 1980s, caesarean delivery rates rose, with the increase concentrated where the income loss was largest. These studies do not show that doctors perform unnecessary operations for money. They show that at the margin, where clinical judgment could reasonably go either way, the payment system tilts the decision. The more consequential distortion is not overuse but misallocation. Fee-for-service does not merely pay for activity; it pays different amounts for different activities, and those amounts are set administratively, not by any market in outcomes. In Medicare, physician fees are set by the Resource-Based Relative Value Scale, introduced in 1992 on the basis of research led by William Hsiao at Harvard. The scale was meant to price services by the work, practice expense and malpractice risk involved. In practice, the relative values are updated with heavy input from a committee convened by the American Medical Association, dominated by specialty societies, and the result has been persistently generous pricing for procedures and imaging relative to the time-intensive cognitive work of primary care, geriatrics and psychiatry. A cardiologist who performs a stress test earns more per hour than one who spends that hour talking a patient through the decision not to have one. Prices shape supply, and supply shapes practice. Specialties that are paid well for procedures attract more trainees, invest in more equipment, and develop more indications for the procedures they perform. Specialties paid poorly for time struggle to recruit. The Milbank Memorial Fund's 2025 scorecard estimated that primary care accounts for only about 4.6 percent of total US health spending, and a smaller share of Medicare spending, despite a large body of evidence that the supply of primary care is associated with better population health and longer life. A payment system that prices procedures above conversations will, over decades, build a health system with too many proceduralists and too few generalists, and no individual decision along the way needs to be wrong for that outcome to emerge. What fee-for-service cannot see A second structural defect is fragmentation. Fee-for-service pays each provider separately for each service. It has no concept of an episode of illness, a course of treatment or a patient's health over time. The surgeon is paid for the operation; the hospital for the admission; the rehabilitation facility for each day of stay; the home health agency for each visit; the primary care physician for the follow-up appointment. Nobody is paid for making the pieces fit together, and nobody is financially responsible for the whole. The consequences are familiar to anyone who has watched an elderly relative move through the system. Tests are repeated because results do not travel. Medications are changed at discharge and not reconciled. A patient is discharged to a nursing facility because that is where the hospital has a relationship, not because it offers the best recovery. Complications that arise after discharge generate new bills rather than costs to the provider whose care produced them. Indeed, under pure fee-for-service, a hospital that causes a surgical infection is paid again to treat it. The system pays more for worse care in exactly those cases where quality fails. The same blindness extends to time. Fee-for-service pays for services delivered now. It places no value on illnesses prevented, hospitalisations avoided or years of healthy life gained, because those are non-events that generate no claims. A primary care practice that invests in nurse outreach to keep its diabetic patients' blood sugar under control reduces future amputations, dialysis and admissions, but the savings accrue to hospitals and insurers, not to the practice, which has simply spent money on unbillable work. Chapter 7 returns to this problem in detail, because it is the most important and least solved problem in payment reform. The Dartmouth Atlas of Health Care, built over several decades by John Wennberg, Elliott Fisher and colleagues, documented the aggregate result. Medicare spending per beneficiary varied by a factor of two or more across American regions, and the variation was not explained by differences in illness or prices. High-spending regions used more hospital days, more specialist visits and more intensive care at the end of life. They did not achieve better survival, better function or higher patient satisfaction. Fisher and colleagues' 2003 analyses in the Annals of Internal Medicine concluded that the additional spending in high-spending regions bought more care but not better outcomes. The finding has been refined and debated since, with later work showing that some of the variation reflects differences in health that the original adjustments missed, but the core observation has survived: much of what the system spends is driven by local supply and practice style rather than by need. The Institute of Medicine's 2012 report Best Care at Lower Cost estimated that about $750 billion of US health spending in 2009 was wasted. A 2019 review by William Shrank and colleagues in JAMA put annual waste at between $760 billion and $935 billion, roughly a quarter of total spending, divided among failures of care delivery and coordination, overtreatment, pricing failures, fraud and administrative complexity. Not all of this is attributable to fee-for-service. Pricing failures reflect market power, and administrative waste reflects the multiplicity of payers. But overtreatment and failures of coordination, which together account for a large share of the estimate, are exactly the behaviours that activity-based payment rewards or fails to discourage. What fee-for-service does well A reform that sees only the defects of the old system will reproduce them in a new form, so it is worth being precise about what fee-for-service does well. First, it rewards effort and productivity. A physician paid per service has a reason to see patients promptly, to add clinic sessions when demand rises, and to work efficiently. Salaried and capitated systems frequently struggle with the opposite problem: long waits, rationing by queue, and too little attention to access. The waiting lists that have long been a political problem in the English NHS are, among other things, a consequence of paying hospitals and consultants in ways that did not reward extra activity. When England introduced activity-based payment for hospitals under the name "Payment by Results" in the 2000s, one explicit aim was to reduce waiting times by paying for the operations that patients were queuing for. Second, fee-for-service protects patients against one specific hazard: stinting. When a provider receives a fixed payment regardless of what it does, the incentive runs toward doing less. For patients whose needs are clear and whose treatments are effective, doing less is harmful. A system that pays a fixed amount per patient per year gives the provider a reason to avoid expensive patients, to delay costly treatments, and to cut corners that patients cannot see. The managed care backlash in the United States in the late 1990s, when health maintenance organisations were widely perceived as denying needed care, was a political response to exactly this hazard, and it set back capitated payment for a decade. Third, fee-for-service is administratively simple and transparent. A service is delivered, a claim is submitted, a price is paid. The provider bears no insurance risk and needs no actuarial capacity. Small practices and rural hospitals can participate without sophisticated data systems. Every alternative model requires something fee-for-service does not: attribution of patients to providers, calculation of spending benchmarks, risk adjustment, quality measurement, reconciliation of accounts months after the care was delivered. Those requirements favour large organisations with the capital to build analytic infrastructure, which is one reason value-based payment has accelerated consolidation in some markets. Fourth, and least appreciated, fee-for-service is the plumbing on which almost every alternative model still runs. Most bundled payments and accountable care arrangements in the United States do not replace fee-for-service claims. They overlay them. Providers continue to bill for each service, and at the end of the year the total is compared with a target and a bonus or penalty is calculated. The claim remains the unit of data from which spending is measured and diseases are recorded. This is not a defect of reform design; it is a practical necessity. But it means that the move "from fee-for-service to value" is, in most places, a move from pure fee-for-service to fee-for-service with an outcome-linked adjustment layered on top. The spectrum of payment units The useful way to think about payment models is not as a binary choice between volume and value but as a spectrum defined by the size of the unit being paid for. At one end is the individual service. Moving along the spectrum, the unit grows: a day of care, a hospital admission, an episode spanning an operation and its recovery, a year of care for a condition, a year of all care for a person. At the far end is a global budget for an entire population. As the unit grows, two things happen. The provider is given more freedom to substitute one input for another, because it is paid for the whole package rather than each component. A provider paid per episode of joint replacement can choose home physiotherapy instead of a rehabilitation facility and keep the difference. And the provider bears more financial risk, because the cost of the package varies with factors, including the patient's underlying health and sheer chance, that the provider does not control. The design problem of every value-based model is to capture the first benefit without imposing too much of the second. Table 1 sets out the main payment units, the incentive each creates, and the characteristic hazard that comes with it. The pattern it shows is that no unit is free of distortion. Each solves some of the problems of the unit smaller than itself and introduces new ones. Table 1. Payment units, the incentives they create and their characteristic hazards. Payment unit Example Main incentive Characteristic hazard Individual service Physician fee schedule Provide more services Overuse, fragmentation Hospital admission Diagnosis-related groups (DRGs) Shorter, cheaper stays Early discharge, more admissions Episode of care Bundled payment for joint replacement Efficient care across settings More episodes, selection of easier cases Population, shared savings Accountable care organisation Lower total spending, quality targets Benchmark gaming, coding intensity Population, full capitation Medicare Advantage, integrated systems Keep enrolled population healthy and cheap Stinting, risk selection Pay-for-performance add-on Hospital value-based purchasing Meet measured quality targets Teaching to the measure The history of the diagnosis-related group is a useful illustration of how this spectrum works in practice. Before 1983, Medicare paid hospitals retrospectively for their costs, which gave hospitals no reason to economise. The introduction of the Inpatient Prospective Payment System that year replaced cost reimbursement with a fixed payment per admission, determined by the patient's diagnosis-related group. The effect on length of stay was immediate and large: hospitals discharged patients sooner, because each additional day was now a cost rather than a source of revenue. The DRG was, in effect, the first large-scale bundled payment, and it worked as intended within the boundary of the admission. But the boundary mattered. Once the admission became the unit of payment, hospitals had reasons to do more admissions, to discharge patients to post-acute facilities that were paid separately, and to code diagnoses as severely as possible to move patients into better-paid groups. The spending that DRGs squeezed out of hospital stays partly reappeared in skilled nursing facilities, home health agencies and readmissions. Growth in Medicare post-acute spending in the late 1980s and 1990s was, in part, a consequence of a payment reform that had solved the problem inside the hospital by pushing costs outside it. This is the pattern to hold in mind throughout the rest of this book: each payment unit draws a boundary, and behaviour changes most at the boundary. From activity to value If fee-for-service buys services, and each larger payment unit buys a larger package of services with its own new distortions, what would it mean to buy value directly? The answer requires two things that fee-for-service never needed. The first is a definition of the outcomes that the payment is supposed to purchase, precise enough to measure and robust enough to bear financial consequences. The second is a way of attributing responsibility for those outcomes to someone who can be paid for them, over a period long enough and a scope broad enough that the provider can actually influence them. Neither is straightforward. Outcomes in medicine are multidimensional, delayed and heavily influenced by factors outside any provider's control, including the patient's own behaviour, genetics, income and neighbourhood. Responsibility is diffuse, because most patients with serious illness are treated by many providers across many settings. And the measurement systems that exist were mostly built for billing, not for assessing whether patients got better. The rest of this book is about how reformers have tried to solve these two problems, and what the results teach. The next chapter begins with the first: what exactly is the value that value-based payment is supposed to buy? Chapter 2: Defining Value "Value" is one of those words that everyone in health care uses and few define. Pharmaceutical companies speak of the value of their drugs; hospital systems advertise value-based care; insurers sell value-based networks; governments promise value for taxpayers' money. Much of this usage means nothing more than "good" or "cheaper". If value-based payment is to be more than a slogan, it needs a definition precise enough to measure and to pay for. Outcomes divided by cost The definition that has done most to shape the field comes from Michael Porter, the Harvard strategy scholar, and Elizabeth Teisberg, in their 2006 book Redefining Health Care, restated by Porter in a widely cited 2010 article in the New England Journal of Medicine. Value, they argued, is health outcomes achieved per dollar spent. It is a ratio, not a slogan, and both its numerator and denominator have specific meanings. The numerator is outcomes, not processes. Whether a clinician followed a guideline, ordered a test or prescribed a drug is a process. Whether the patient survived, recovered function, avoided complications and returned to work is an outcome. Processes matter only insofar as they produce outcomes. Most quality measurement in health care has historically focused on processes, because they are easier to record and more directly under the clinician's control, but a process measure is at best a proxy for value and at worst a distraction from it. The denominator is the total cost of achieving the outcome, across the whole cycle of care. Not the price of a single service, not the cost to one department or one provider, but the full cost of all resources used to treat the patient's condition from diagnosis through recovery. A hospital that shortens length of stay by discharging patients sicker, only for them to be readmitted, has not reduced the cost of the cycle; it has moved it. A surgeon who uses a cheaper implant with a higher failure rate has not improved value if the revision surgery five years later is counted. Porter's framework makes a second claim that is easy to miss but essential. Value must be measured around the patient's medical condition, defined as an interrelated set of circumstances best addressed in an integrated way. Knee osteoarthritis is a condition. Type 2 diabetes is a condition. Breast cancer is a condition. For primary and preventive care, the relevant unit is a patient population with similar needs, such as healthy children, frail elderly people or adults with multiple chronic illnesses. The reason this matters is that outcomes are condition-specific. What counts as success for a patient with heart failure bears no relation to what counts as success for a patient having cataract surgery, so any meaningful measure of value has to be built condition by condition. This has direct implications for payment. Fee-for-service pays for services that cut across conditions and patients; a chest X-ray has the same price whether it is ordered for pneumonia, heart failure or no good reason at all. Value-based payment, in Porter's sense, would pay for the care of a condition, over the cycle of that condition, with some part of the payment contingent on the outcomes achieved. Bundled payments, discussed in the next chapter, are the most direct attempt to do this for acute episodes. Condition-based capitation, a fixed annual payment for managing a chronic disease, is the analogue for long-term illness. What counts as an outcome Porter proposed that outcomes for any condition could be organised in a three-tier hierarchy. The first tier is the health status achieved or retained: survival, and the degree of health or recovery, such as freedom from pain, ability to walk or return to normal activity. The second tier is the process of recovery: how long it took to return to normal function, and the disutility of the care itself, including complications, treatment errors and their consequences. The third tier is the sustainability of health: whether the recovery lasted, whether the condition recurred, and what long-term consequences of treatment emerged, such as radiation-induced cancers or implant failures. The hierarchy is useful because it makes explicit how much of what patients care about is missing from conventional measurement. Clinical records capture survival and major complications reasonably well. They capture very little about pain, function, the time taken to return to work, or the patient's own judgment of whether the treatment was worth it. Those things can only be learned by asking the patient, which is why patient-reported outcome measures, the subject of Chapter 5, sit at the heart of any serious attempt to measure value. A related framework, the "Triple Aim", was set out by Donald Berwick, Thomas Nolan and John Whittington of the Institute for Healthcare Improvement in Health Affairs in 2008. It proposed that health systems should pursue three goals simultaneously: improving the experience of care, improving the health of populations, and reducing per capita costs. The Triple Aim differs from Porter's framework in emphasis. Porter focuses on conditions and the providers who treat them; Berwick and colleagues focus on populations and the systems that serve them. The Triple Aim was later extended to a "Quadruple Aim" by Thomas Bodenheimer and Christine Sinsky, adding the well-being of the clinicians and staff who deliver care, on the argument that burned-out clinicians cannot deliver the other three. In practice, the population perspective has shaped accountable care organisations and the condition perspective has shaped bundled payments, and a working value-based system needs both. European thinking has added another dimension. In 2019 the European Commission's Expert Panel on Effective Ways of Investing in Health published an opinion arguing that value in health care should be understood through four pillars: personal value, meaning care appropriate to each patient's goals; technical value, meaning the best possible outcomes from available resources; allocative value, meaning equitable distribution of resources across all patient groups; and societal value, meaning the contribution of health care to social participation and cohesion. The panel's concern was that a narrow focus on outcomes per dollar for the patients in front of a provider could neglect the patients who never reach that provider at all. That concern is real, and it recurs in Chapter 6 when we examine how value-based programmes can penalise institutions serving disadvantaged populations. Measuring the denominator If outcomes are hard to measure, costs are in some ways harder. Most hospitals do not know what it costs them to treat a particular patient with a particular condition. Their accounting systems were built to allocate costs to departments and to generate charges for billing, and the "cost" of a service in a hospital ledger is usually a charge multiplied by a department-wide cost-to-charge ratio, which bears little relation to the resources actually consumed. Robert Kaplan and Michael Porter proposed a remedy in a 2011 Harvard Business Review article: time-driven activity-based costing. The method maps every step in a patient's care pathway, estimates the time each clinician and each piece of equipment spends on that step, and multiplies by the cost per minute of each resource. The result is a bottom-up estimate of what the cycle of care actually costs, which can then be compared across providers, across patients and across alternative pathways. Hospitals that have applied the method have frequently discovered that expensive resources, such as surgeons and operating theatres, were being used for tasks that cheaper resources could perform, and that the largest cost differences between pathways lay not in the headline procedures but in the days of care before and after. Accurate costing matters for payment design because a bundled price or a capitation rate that bears no relation to the real cost of care will either overpay efficient providers or drive out those that treat complex patients. It also matters for providers, because an organisation that does not know its own costs cannot tell whether a value-based contract will make or lose money until it is too late. There is a subtler point about the denominator. Costs to whom? A provider's cost is not the same as the payer's spending, which is not the same as the cost to society. A bundled payment that reduces Medicare spending by discharging patients home rather than to a nursing facility has reduced the payer's cost, but some of the cost of recovery may have shifted to family carers who now provide unpaid help. A value calculation from the payer's perspective will count the saving and miss the shift. Most real-world value-based programmes measure value from the payer's perspective, because that is what the payer can observe, but a reader evaluating them should remember what is being left out. Value in a worked example An illustration shows why the ratio matters more than either of its parts. Consider two hypothetical hospitals performing knee replacements for similar patients. Hospital A has a short length of stay and a low price per admission, and its finance department regards it as the more efficient of the two. Hospital B spends more on the admission itself: it runs a pre-operative education class, has a physiotherapist see every patient on the day of surgery, and assigns a nurse to telephone each patient in the week after discharge. Now follow the whole cycle. Suppose Hospital A discharges a larger share of its patients to a rehabilitation facility, because its patients have not been prepared to go home, and suppose a larger share of its patients return to the emergency department with pain or wound problems in the month after surgery. Suppose, finally, that when patients are asked a year later how well they can walk and whether they still have pain, Hospital B's patients report larger improvements. On the measure most hospitals track, the cost of the admission, Hospital A looks better. On the measure that matters, outcomes achieved per dollar across the full cycle, Hospital B may deliver far more value, because its extra spending in the admission is more than offset by lower spending afterwards and its outcomes are better. The example is invented, but the pattern it describes is exactly what bundled payment evaluations have found in real hospitals, as the next chapter shows. It makes three general points. First, the cost that matters is the cost of the cycle, not of any single component, and cutting a component can raise the total. Second, the outcomes that distinguish good care from mediocre care, in this case pain and function a year later, are often invisible in administrative data and can only be learned by asking patients. Third, under fee-for-service, Hospital B is punished twice: it is paid the same for the admission despite spending more on it, and it loses the downstream revenue that Hospital A's rehabilitation referrals and emergency visits generate for the wider system. Value-based payment is, at bottom, an attempt to make Hospital B's approach pay. The same logic applies to populations. A primary care practice that spends more on nurse care managers for its patients with heart failure will look expensive on any measure of practice cost. If those care managers prevent hospital admissions, the practice has increased value across the population it serves, but under fee-for-service it bears the cost while hospitals and insurers capture the savings. The unit at which value is measured, whether the service, the episode or the population, determines whose efforts are recognised. A classification of payment models In practice, the US health system tracks its progress toward value-based payment using a classification developed by the Health Care Payment Learning and Action Network, a public-private partnership convened by the Department of Health and Human Services in 2015. The framework sorts payments into four categories according to how strongly they are linked to value, and each year the network surveys payers on how their spending is distributed across them. Category 1 is pure fee-for-service with no link to quality. Category 2 is fee-for-service with some link to quality, such as bonuses for reporting quality data or for meeting performance targets. Category 3 covers alternative payment models built on fee-for-service architecture, divided into 3A, where providers share in savings but bear no losses, and 3B, where they share in both. Category 4 covers population-based payments, in which providers receive a prospective payment for a condition or for all of a population's care, and fee-for-service claims are no longer the basis of payment. Table 2 shows how US health care payments were distributed across these categories in 2023, as reported in the network's most recent survey. The figures are striking in two respects. First, even after more than a decade of reform, more than half of all payments remained in categories 1 and 2, that is, fee-for-service with at most a modest link to quality. Second, the pattern differed sharply by line of business. Medicare Advantage, where insurers themselves are paid a capitated rate and often pass that capitation on to provider groups, had by far the largest share of population-based payment. Commercial insurance, which covers most working-age Americans, lagged behind. Table 2. Share of US health care payments by payment category, 2023. Line of business Categories 1–2 Category 3A Category 3B Category 4 Commercial 60.8% 17.6% 13.4% 8.2% Medicaid 56.3% 22.6% 11.9% 9.2% Traditional Medicare 58.0% 8.3% 21.8% 11.9% Medicare Advantage 35.7% 21.3% 9.0% 34.0% All lines combined 54.8% 16.7% 14.3% 14.2% Source: Health Care Payment Learning and Action Network, 2024 APM Measurement Effort (calendar year 2023 payments). The classification is useful but it can mislead. It measures the form of payment contracts, not their strength. A contract in Category 3A may put only a small fraction of a provider's revenue at stake, with generous benchmarks that make savings easy to claim; a Category 2 pay-for-performance programme might, in principle, have larger financial consequences. The shift from Category 1 toward Category 4 is also not an unambiguous improvement. Population-based payment carries the hazards of stinting and risk selection discussed in Chapter 1, and a large share of Category 4 payment in Medicare Advantage reflects capitation paid to insurers whose profitability depends partly on how intensively they code their enrollees' diagnoses, a problem examined in Chapter 6. The categories describe the plumbing of payment. Whether that plumbing actually buys value depends on the details. Value to whom, over what period The definition of value has one more dimension that matters for everything that follows: time. The value of a hip replacement becomes apparent within a year. The value of better diabetes control becomes apparent over a decade. The value of childhood vaccination or smoking cessation becomes apparent over a lifetime. Any payment model has to choose a time horizon over which outcomes are measured and costs are counted, and that choice determines which kinds of care the model rewards. Short horizons favour interventions whose benefits are quick and visible. They systematically undervalue prevention, chronic disease management and anything whose returns are delayed. Long horizons are better in principle, but they require long-term relationships between payers, providers and patients, and in health systems where people change insurers frequently, no single payer may be around to collect the returns. This is the "wrong pocket" problem in its temporal form: the entity that pays for prevention is often not the entity that benefits from it. Porter's definition of value, outcomes over costs across the full cycle of care, is the right standard. The rest of this book is largely about the compromises that have been made in trying to apply it: shortening the cycle to something measurable, narrowing the outcomes to something collectable, and dividing the costs among payers and providers in ways that each can bear. Those compromises are unavoidable. The test of a good payment model is whether its compromises leave the incentives pointing in roughly the right direction. The next two chapters examine the two main models that have been tried at scale, starting with the one that most directly embodies the idea of paying for the care of a condition: the bundled payment. Chapter 3: Bundled Payments A bundled payment is a single price for an episode of care. Instead of paying the surgeon, the hospital, the anaesthetist, the rehabilitation facility and the home health agency separately for each thing they do, the payer sets one amount for everything related to, say, a hip replacement from the day of surgery to ninety days afterwards. If the providers deliver the episode for less, they keep some or all of the difference. If it costs more, they bear some or all of the excess. The idea is appealing because it corresponds closely to how patients experience care. A patient having a hip replaced does not think of her treatment as a set of separate transactions with separate organisations. She thinks of it as one thing, the operation and her recovery from it, and she judges it by whether she can walk without pain at the end. A bundle draws a boundary around that one thing and makes someone responsible for all of it. The mechanism The logic of a bundle rests on three effects. First, it rewards efficiency within the episode. Because the price is fixed, every unnecessary test, every avoidable day in hospital and every unneeded visit becomes a cost rather than a source of revenue. Second, it rewards coordination across settings. Because the hospital, surgeon and post-acute providers share one budget, they have a reason to work together: to plan discharge before admission, to choose the least intensive setting in which the patient can safely recover, and to prevent the complications that would otherwise generate costly readmissions. Third, if the bundle includes complications, it creates something like a warranty. A provider that causes a surgical infection pays to treat it, rather than being paid again. A bundle also has a structural limit. It fixes the price per episode, but it does not fix the number of episodes. A provider paid a profitable bundled price for knee replacements has the same incentive as under fee-for-service to perform more knee replacements, including on patients whose benefit is marginal. A bundle rewards doing each episode well; it says nothing about whether the episode should have happened at all. This is why bundled payments work best for conditions where the decision to treat is relatively clear, such as hip fracture, and are weakest for elective procedures where the indication is discretionary, such as spinal fusion for back pain. It is also why bundles are ultimately a complement to population-based models, which do have a reason to ask whether the episode was needed, rather than a substitute for them. Two design choices determine how a bundle behaves. The first is the target price: how much the payer pays per episode. If the price is set from the provider's own historical costs, efficient providers are penalised for their past efficiency and inefficient ones are rewarded for their past waste. If it is set from regional or national averages, the reverse happens, and providers treating sicker or poorer patients may find the price unachievable. Most modern bundles blend the two, starting from a provider's history and moving toward a regional benchmark over time, with adjustments for patient risk. The second is the episode definition: what services are included, for how long, and which patients qualify. A narrow definition, such as the hospital stay alone, is easy to administer but leaves the most important savings, in post-acute care, outside the boundary. A broad one, such as ninety days of all related care, captures more but exposes providers to more variation they cannot control. How reconciliation works In practice, almost no modern bundle is paid as a single cheque in advance. The usual arrangement is "retrospective reconciliation". All providers continue to bill fee-for-service as normal throughout the year. At the end of a performance period, the payer adds up all the claims for each episode, compares the total with the target price, and either pays the accountable provider a reconciliation payment or asks for a repayment. A simple hypothetical example shows how this works. Suppose a hospital's target price for a joint replacement episode is set at $25,000, which already includes a small discount relative to its expected spending. Over the year, the hospital's episodes average $23,500 in total claims across the surgeon, the hospital stay, post-acute care and any readmissions. The difference of $1,500 per episode, subject to a quality adjustment and a cap on the maximum gain, is paid to the hospital as a reconciliation payment. If the episodes had averaged $26,500, the hospital would owe Medicare some or all of the $1,500 excess, depending on its risk track. The hospital can, if the rules permit and within legal limits, share its gains with the surgeons and post-acute providers who helped achieve them, an arrangement known as gainsharing. Three features of this mechanism shape behaviour. The first is that the discount built into the target price is how the payer captures its share of savings. If a hospital merely matches its historical spending, the payer still saves the discount; the hospital earns nothing. The second is that the hospital bears the cost of care it does not provide directly: if its patients go to an expensive rehabilitation facility or are readmitted elsewhere, those claims count against its target. That is what gives it a reason to manage the whole episode. The third is that the reconciliation arrives months after the care, which weakens the connection between clinical decisions and financial consequences, and means that a hospital without its own analytic capacity may not know how it is performing until long after it could have changed course. Early experiments The bundled payment is older than the phrase "value-based care". In 1991 Medicare launched the Participating Heart Bypass Center Demonstration, which paid a small number of hospitals a single negotiated price covering both hospital and physician services for coronary artery bypass surgery. The demonstration ran until 1996, and its evaluation reported savings to Medicare of around ten percent on the bundled services, driven mainly by shorter stays and fewer tests, with no evidence that quality suffered. It was not extended, partly because the participating hospitals were self-selected high performers and partly because the administrative machinery for bundling was cumbersome in an era before electronic claims processing. The idea was revived in 2006 by Geisinger Health System in Pennsylvania, a physician-led integrated system with its own health plan. Geisinger's ProvenCare programme began with elective coronary bypass surgery. Its cardiac surgeons agreed on forty best-practice steps drawn from national guidelines, built them into the electronic record so that each step was checked before the patient could move on, and offered the health plan a single price covering the operation and ninety days of related follow-up care, including the treatment of any complications, at no extra charge. In effect, Geisinger sold a warranty. The programme's early results, published by Alfred Casale and colleagues in the Annals of Surgery in 2007, reported high rates of compliance with the best-practice steps and improvements in several outcomes, including readmissions. ProvenCare was later extended to other procedures and chronic conditions. Its significance lay less in any single result than in demonstrating that clinicians could redesign a pathway around a fixed price and share the gains. Outside the United States, the most influential experiment was OrthoChoice, introduced by the Stockholm County Council in Sweden in 2009. The council paid a fixed price for elective hip and knee replacements in relatively healthy patients, covering the pre-operative assessment, the surgery, the implant, and a first follow-up visit, and it made providers financially responsible for treating related complications for a multi-year warranty period. Patients were free to choose among providers, and outcomes were tracked through Sweden's national joint registries, among the oldest and most complete in the world. Subsequent analyses reported lower costs, fewer complications and shorter waiting times, while noting that the design deliberately excluded the sickest patients, which makes its results easier to achieve than they would be for an unselected population. OrthoChoice mattered because it paired a bundle with a registry that measured outcomes over years, not just days. The Netherlands took a different approach in 2010, introducing bundled payment for chronic conditions: type 2 diabetes, chronic obstructive pulmonary disease and cardiovascular risk management. Insurers contracted with "care groups", typically organisations of general practitioners, for an annual price covering the full package of primary and routine specialist care for each patient. The care group then subcontracted with dietitians, podiatrists, laboratories and specialists. Early evaluation found improvements in care coordination and protocol adherence, but also that the bundles did not reduce overall spending, in part because specialist and hospital care for complications lay outside them. The Dutch experience illustrates a general lesson: a bundle that covers only part of a condition's costs will shift spending toward the parts it leaves out. Medicare at scale The large-scale test of bundled payment came through the Center for Medicare and Medicaid Innovation, created by the Affordable Care Act in 2010. Its first major bundling initiative, the Bundled Payments for Care Improvement programme, ran from 2013 to 2018 and allowed hospitals and physician groups to volunteer for bundles covering a wide range of conditions. Its successor, BPCI Advanced, ran from October 2018 to the end of 2025, again voluntarily, with ninety-day episodes and a single set of rules. The most scientifically informative experiment was mandatory. The Comprehensive Care for Joint Replacement model, launched in April 2016, required hospitals in randomly selected metropolitan areas to accept bundled payment for hip and knee replacements. Random selection meant that CJR could be evaluated almost like a clinical trial, by comparing hospitals in the chosen areas with hospitals in areas that were not chosen. CJR ran, in modified form, until the end of 2024. Table 3 summarises what the major evaluations found. The pattern is consistent. Bundles reduced spending per episode by a few percent, without measurable harm to quality, and almost all of the reduction came from one place. Table 3. Selected evaluations of Medicare episode-based payment models. Model Design Spending effect per episode Quality Main driver CJR, first two years Mandatory, randomised areas −$812 (−3.1%) institutional spending No change in complications Fewer discharges to post-acute facilities CJR, performance year 6 Mandatory −$1,012 (−3.5%) Similar to controls Fewer rehabilitation facility discharges BPCI Advanced, eight model years Voluntary Reductions of about $700–$1,100 Maintained Post-acute care use Oncology Care Model, first three years Voluntary −$297, less than the $704 monthly care payments Mostly unchanged Supportive care drugs Source: Barnett et al., NEJM 2019; CMS CJR performance year 6 evaluation; CMS BPCI Advanced final evaluation summary, 2026; Keating et al., JAMA 2021. The Barnett study in the New England Journal of Medicine found that in the first two years of CJR, institutional spending per episode fell by $812, or 3.1 percent, relative to control areas, driven largely by a 5.9 percent relative reduction in the share of patients discharged to post-acute facilities. There was no significant change in complication rates, and, importantly, no evidence that hospitals were avoiding high-risk patients. By the sixth performance year, the official evaluation found a relative reduction of $1,012 per episode, or 3.5 percent, again driven mostly by reduced use of inpatient rehabilitation facilities, and for the first time CJR produced net savings to Medicare after accounting for payments to hospitals, of about $54 million that year. BPCI Advanced, according to its final evaluation summary published by CMS in September 2026, produced reductions of roughly $700 to $1,100 per episode and more than $800 million in net savings to Medicare across its eight years, though it did not meet the Innovation Center's criteria for expansion. The Oncology Care Model, which paid oncology practices a monthly fee per patient on top of fee-for-service in exchange for accountability for six-month chemotherapy episodes, reduced episode spending by less than the monthly fees cost, producing a net loss. Why the savings came from post-acute care The consistent finding that bundled payment savings come mainly from post-acute care deserves explanation, because it is revealing about both bundles and fee-for-service. Post-acute care in the United States, meaning skilled nursing facilities, inpatient rehabilitation facilities, long-term care hospitals and home health, was for decades one of the most variable components of Medicare spending. Two patients with identical hip replacements might go home with outpatient physiotherapy, or spend two weeks in a rehabilitation facility, depending largely on local practice patterns and on which facilities had relationships with the hospital. Under fee-for-service, the hospital had no reason to care which option was chosen, because the post-acute provider was paid separately, and the facility had every reason to keep beds full. Evidence that institutional rehabilitation produced better outcomes than home recovery for typical joint replacement patients was weak. A bundle made that choice visible and gave the hospital a reason to make it carefully. The hospitals that did best in CJR and BPCI Advanced typically did three things: they began discharge planning before surgery, setting expectations with patients and families that most would go home; they built preferred networks of post-acute providers who agreed to share data and follow agreed pathways; and they tracked patients after discharge to catch problems before they became readmissions. None of this required new medical technology. It required someone to be responsible for the whole episode. The flip side is that bundles did little to change the core of the episode itself: the surgery, the anaesthesia, the implant, the hospital stay. Those components had already been squeezed by the DRG system and by earlier efficiency efforts. It also means that once post-acute use has been rationalised, the easy savings are gone. A second generation of bundles cannot expect to find the same slack, and repeated rebasing of target prices to each year's lower spending risks pushing prices below what efficient care actually costs, a version of the "ratchet" problem examined in the next chapter. The next generation Medicare's current bundled payment model, the Transforming Episode Accountability Model, began on 1 January 2026 and is scheduled to run for five years. TEAM is mandatory for acute care hospitals in selected geographic areas and covers five surgical episodes: lower extremity joint replacement, surgical treatment of hip and femur fracture, spinal fusion, coronary artery bypass grafting, and major bowel procedures. Its episodes are shorter than CJR's, running from the hospital stay or outpatient procedure to thirty days afterwards, and its target prices include a discount relative to expected spending, which in the model's current rules is 1.5 percent for bypass surgery and bowel procedures and 2 percent for the other three. Hospitals can choose among tracks with different levels of risk, including an initial year with no downside risk, and safety-net hospitals are given longer on-ramps and a lower-risk track. Three features of TEAM show how bundle design has evolved. First, it adds a patient-reported outcome measure, the hospital-level measure of pain and function after hip and knee replacement discussed in Chapter 5, to the quality measures that adjust its payments. This is the first time a major Medicare bundle has made patients' own reports of their outcomes part of the payment calculation. Second, it requires hospitals to refer patients to a primary care provider on or before discharge, an explicit attempt to connect a short surgical episode to the longer relationship in which the patient's health will actually be managed. Third, its shorter thirty-day window reflects a judgment that most of the variation hospitals can influence falls in the first month, and that longer windows expose hospitals to too much noise. Whether TEAM will produce larger savings than its predecessors is not yet known. Its first results will not be available until well into 2027. The evidence from CJR and BPCI Advanced suggests realistic expectations: savings of a few percent per episode, concentrated in post-acute care, with quality held steady. That is a real gain across millions of episodes, but it is not a transformation. What bundles can and cannot do The lessons of fifteen years of bundled payment can be stated simply. Bundles work when the episode is clearly defined, the decision to treat is not highly discretionary, and there is significant variation in how care is delivered across settings. They work by making one party responsible for coordination and giving it a reason to eliminate avoidable spending at the boundaries between providers. They do not, by themselves, discourage unnecessary episodes, reward prevention, or improve the core clinical quality of the procedure unless outcome measures are built in. A bundle is, in the end, a partial answer. It moves the payment unit from the service to the episode, which is closer to how patients experience care. But patients experience more than episodes. They experience years of living with conditions, and the most important decisions, whether to have surgery at all, whether a chronic disease is controlled well enough to avoid the episode entirely, happen outside any episode's boundary. Those decisions are the domain of population-based models, and the most widespread of these, the accountable care organisation, is the subject of the next chapter. Hashtags: #ValueBasedHealthcare #FeeForService #QualityOutcomes #HealthcarePaymentReform #PatientOutcomes #HealthcareValue #OutcomeBasedCare #BundledPayments #AccountableCareOrganizations #PatientReportedOutcomeMeasures #HealthcareQuality #CostOfCare #PopulationHealth #SharedSavings #Capitation #PayForPerformance #RiskAdjustment #HealthcarePrevention #CareCoordination #EpisodeBasedPayment #ValueBasedPayment #HealthcareEfficiency #PaymentIncentives #HealthSystemTransformation #FutureOfValueBasedHealthcare

Download the Book (PDF): Introduction In 2015, BlackBerry bought a company called Good Technology for $425 million. Good had been valued privately above $1.1 billion, had filed to go public, and employed hundreds of people who held stock and options in it. When the sale closed, the preferred shareholders were largely made whole and the common stock was worth about forty-four cents a share. Employees who had exercised options at several dollars a share, and paid tax on the difference between the strike price and what they had been told the stock was worth, had paid tax on a gain that turned out never to have existed. Nobody committed fraud. No court found any wrongdoing in the sale price. The documents did exactly what they said they would do, and every one of those documents had been signed, years earlier, by people acting reasonably in rooms where the possibility of a $425 million outcome looked like an excellent day. That gap — between what a set of contracts obviously says and what people who signed them believe they say — is the subject of this book. Venture capital is usually explained as a business of judgement. The stories it tells about itself are stories about conviction: the investor who saw something nobody else could see, the founder who was right when everyone said otherwise, the meeting that changed everything. Those stories are not false. They are also not where most of the money is won and lost. Most of the money is won and lost in arithmetic that nobody in the room performed, because performing it requires knowing that it is there. The claim of this book is that venture outcomes are produced by a small number of mechanisms — ownership, dilution, the preference stack, the power-law distribution of returns, and the interest rate at which future cash is discounted — and that these mechanisms are exact. They have correct answers. The answers do not depend on anyone's view of a market or a founder. And the people who are surprised by their own outcomes are almost never people who made a bad judgement call; they are people who signed something whose consequences they had not calculated. Consider how ordinary the surprises are. A founder is offered money at a valuation and accepts, not realising that the option pool she has agreed to is carved out of her side of the table and that the headline number therefore values her shares at two thirds of what she thinks. A seed investor accepts a term in a later round giving the new money seniority over his, and discovers at a modest exit that seniority meant the difference between recovering his capital and recovering nothing. A limited partner re-ups with a manager reporting a three-times return, not having distinguished between value the manager has estimated and cash the manager has actually returned. An angel signs four post-money safe agreements over two years, each one feeling like a small thing, and arrives at a Series A having given away half the company before a priced round has occurred. Each of those is a solvable problem. Each requires nothing more than knowing which calculation to run, and running it before signing rather than afterwards. This book runs those calculations. It does the cap-table arithmetic properly, with real numbers and round figures, and shows exactly who pays for an option pool and exactly what happens when a convertible instrument and a new option pool have to be resolved in the same transaction — a genuinely circular problem with a clean solution that most spreadsheets get wrong. It works a full exit waterfall through four different sale prices and identifies the precise point at which a preferred holder is better off abandoning its preference. It states the anti-dilution formulas in words and shows the difference between a full ratchet and a weighted average on identical facts: in one worked case, 706,000 extra shares against 12,000,000. It builds a fund model and demonstrates why a single position usually has to return the whole fund, and what that implies for cheque size, ownership targets and the decision about whether to write the next cheque into a company that has stopped working. It also takes the macroeconomics seriously rather than gesturing at it. Venture capital holds the longest-duration assets in finance, and the arithmetic of discounting explains why a two-percentage-point change in the rate of interest can change what a fast-growing company is worth by half or more, why private valuations lag public ones by four to six quarters, and why a fall in public equity markets produces a venture fundraising drought a year later through a mechanism that has nothing to do with anyone's opinion of venture capital. That last point matters in September 2026, because the Federal Reserve has just started raising rates again while the venture industry deploys capital at the fastest pace in its history. What this book will not do is tell you which companies to back. There is no method for that, and anyone selling one should be treated with suspicion. Nor will it argue that the asset class is a bubble, or that it is not. Both positions are currently defensible from the same data: US startups raised a record $412.7 billion in the first half of 2026, and the distributions actually returned to limited partners have been running at a fraction of the long-run average for four years. It is entirely possible to look at that pair of facts and reach opposite conclusions, and the argument between them is not resolvable by arithmetic. What arithmetic can do is tell you what has to be true for either view to pay off, which is a more useful thing to know than which view is fashionable. The book is also not a legal text. It describes provisions that appear in the model financing documents used across the American market and the case law that has shaped them, but the law here is genuinely jurisdiction-specific and genuinely detailed, and no book substitutes for counsel who has read your actual charter. What it can do is tell you which clauses are worth paying a lawyer to argue about, and which ones founders spend their negotiating capital on for no reason. Four sorts of reader should find it useful. A founder about to price a round, who wants to understand what she is actually agreeing to and where the leverage in the negotiation really sits. An operator holding options in a private company, who would like to know what those options are likely to be worth under various outcomes and why the answer is so often less than the headline valuation implies. An investor — whether a partner at a fund, an angel writing small cheques, or an analyst at a limited partner — who wants the mechanics stated precisely rather than approximately. And anyone professionally adjacent to this market who is tired of reading about venture capital in a register that is either promotional or contemptuous, and would prefer to read about how it works. No prior vocabulary is assumed. Every term is defined where it first appears, including the ones that practitioners use so casually that they have stopped noticing they are jargon. Where a figure is well established it is stated precisely; where it is uncertain or comes from a source that should be treated with care, it is described qualitatively rather than given false precision. There are no invented statistics in this book, which means there are fewer statistics in it than there might otherwise be. One further warning about what the numbers in this book are for. Every worked example uses round figures — 8,000,000 shares, a $10 million post-money valuation, a $2.5 million cheque — chosen so the arithmetic can be followed without a spreadsheet. Real cap tables are not round, and a real closing involves share counts with seven significant figures and rounding conventions that occupy a paragraph of the stock purchase agreement. None of that changes the structure of the calculation. The point of a clean example is to make the mechanism visible, and once it is visible, the messy version becomes a matter of substituting numbers rather than of understanding. A word about tone. It is easy to write about venture capital either as a heroic enterprise or as a racket, and both framings are available in quantity. Neither is accurate, and neither is helpful to someone who has to sign a document next week. The structures described here were, for the most part, designed by intelligent people to solve real problems: a liquidation preference makes it rational to pay a premium for a minority stake in a company with no earnings, a carried interest arrangement aligns a manager with the capital they deploy, a valuation cap compensates an early investor for taking risk before a price exists. Every one of them also produces outcomes that the people affected by it did not anticipate. Both things are true, and treating either as the whole story leads to bad decisions. What follows is an attempt to state, precisely and in prose, how the money actually moves. It starts with the market as it stands, because the arithmetic only matters once you know what it is being applied to — and the market as it stands in late 2026 is stranger than at any point in the industry's history. Chapter 1 A Record Year That Almost Nobody Is Having American startups raised roughly $412.7 billion in the first half of 2026. That is the largest half-year total in the history of the asset class, comfortably ahead of the 2021 peak that everyone spent the following four years apologising for. If you read only that number, you would conclude that venture capital had recovered, that the correction which began in late 2021 was over, and that the industry had returned to something like normal operation. Almost every other number contradicts it. Deals of $100 million or more absorbed roughly 87.5 percent of that capital. Whatever is left — something under an eighth of the total — is what funded every seed round, every Series A and every Series B in the United States for six months. Companies working on artificial intelligence took around 86 percent of the dollars. A decade ago, AI companies were about one deal in seven; in early 2026 they were more than two deals in five, and because their rounds are far larger than everyone else's, their share of dollars ran far ahead of their share of deals. The aggregate is a record. The distribution underneath it is the narrowest the industry has ever recorded. The same concentration appears on the other side of the table, among the funds themselves. US venture firms raised about $74.8 billion in the first half of 2026 — close to the whole of the previous year's total, so on the face of it another recovery. But funds of a billion dollars or more took roughly 68 percent of it, against something closer to 36 percent the year before. Three firms — Andreessen Horowitz, Thrive Capital and Founders Fund — took something approaching half of all the capital raised. Funds below $50 million were about two-thirds of the vehicles that closed and raised about four percent of the money. First-time funds are on track for their weakest year since 2016. This is the shape of the market a portfolio strategy has to survive: enormous aggregate liquidity at the top, a structural drought everywhere else, and a return profile that will be decided by whether a handful of very large private companies convert paper value into cash. The distribution problem Venture capital has two ways of telling a limited partner how a fund is doing, and they are not the same kind of statement. TVPI — total value to paid-in capital — divides everything the fund holds, realised and unrealised, by the money the LP has actually sent in. A TVPI of 2.5x means the general partner believes the portfolio is worth two and a half times what the LP contributed. DPI — distributions to paid-in capital — divides only the cash that has been sent back. A DPI of 0.3x means the LP has received thirty cents for every dollar committed and spent. TVPI is an opinion. It rests on marks, and marks in venture are largely a function of the price of the last round, which is a price set by a small number of buyers in a negotiation neither party has any incentive to describe as disappointing. DPI is a fact. Money either arrived in the LP's account or it did not. Funds raised in 2020 and 2021 have an unusual amount of the first and an unusual scarcity of the second. They deployed into the highest-priced entry environment on record, marked those positions up through 2021, and then held them through a period in which the exit market effectively closed. Distributions across the industry fell in 2024 to the lowest level since the financial crisis — on the order of six percent of net asset value returned in a year, against a long-run average closer to eighteen or nineteen percent. The recovery since has been partial. Recent one-year marked returns in the high teens sit alongside cash distributions of roughly twelve percent of asset value: the paper is moving faster than the money. For a limited partner, this is not an abstraction. Endowments, foundations and pension funds run on spending rules. A university that draws five percent of its endowment annually needs cash, and when its venture managers stop distributing, the cash has to come from somewhere else — from selling public equities at whatever price the market offers, from borrowing, or from declining to make new commitments. All three happened between 2022 and 2025, and the third is why the fundraising market for anyone who is not already enormous has remained brutal even as headline totals recovered. There is a second-order effect that matters more than it first appears. A general partner's ability to raise the next fund has historically depended on the marks in the current one, because marks arrive faster than exits and a fund raising in year four has almost nothing realised to show. That worked when the relationship between marks and eventual cash was reasonably stable. After 2021 it stopped being stable, and sophisticated LPs began discounting unrealised value heavily — applying their own haircuts to reported TVPI, or simply refusing to underwrite a manager without meaningful DPI from a prior fund. The median gap between a firm's funds has compressed to under two years at the top of the market and stretched indefinitely everywhere else. Firms that can raise are raising faster than ever. Firms that cannot are discovering that a 3.8x TVPI is not a fundraising document. The window opened, for a very small number of companies The most consequential exit event of the decade so far happened in June 2026, when SpaceX listed. Within days its market capitalisation passed $2 trillion, well ahead of the roughly $1.5 trillion valuation that had been reported as its target in the run-up. A single listing generated more exit value than the industry had seen in years, and it flattered every aggregate statistic for the half-year in which it occurred. It also tells you almost nothing about the exit environment for the other 950-odd private companies in the United States carrying billion-dollar-plus valuations. Since 2022, fewer than fifty venture-backed American companies have gone public in any single year. The 2025 cohort — CoreWeave in March, Circle and Chime in June, Figma in July, Klarna in September — was genuinely a reopening after three years of near-total closure, and it was celebrated as such, but it was a reopening measured in dozens, not hundreds. Klarna is the instructive case. The Swedish payments company was marked at $45.6 billion in 2021. In July 2022 it took a down round at $6.7 billion — an eighty-five percent cut, executed in public, which at the time was treated as a scandal and in retrospect looks like unusually honest pricing. When it finally listed in 2025 it did so at a small fraction of its 2021 mark. Every investor who came in at the 2021 price lost money on a company that is, by any operating measure, a real and substantial business. The company was fine. The entry price was not. Two structural facts about exits have changed in ways that matter for portfolio construction. The first is that companies now consume vastly more capital before they reach a meaningful exit. The median company exiting above $500 million had raised something in the region of $324 million of venture capital, against roughly $157 million a decade earlier. The exit price has not doubled to match. Which means the multiple on invested capital available at any given exit price has compressed, mechanically, for everyone. The second is that the step-up from the last private round to the exit price has collapsed. In 2021, a company exiting was typically doing so at a substantial premium to its final private mark — on the order of sixty percent. That premium is now closer to fifteen percent. The last private round used to be a staging post on the way to a higher public price. It is now, frequently, the high-water mark. Crossover investors who piled into late-stage private rounds in 2021 on the theory that they were buying a discount to an imminent IPO discovered that they had in fact been the exit. Prices have reset, unevenly The valuation correction that began in late 2021 was real, was severe, and is largely over — for the companies that survived it. Carta, which administers cap tables for a large share of the American startup market and therefore sees pricing data others do not, put the down-round rate at about 11.4 percent of priced rounds in the first quarter of 2026. That is back to where it sat in 2019 and 2020. At the depth of the reset it was roughly double that: something like one financing in five was priced below the prior round. A company raising at a lower price than last time was, for about two years, an ordinary event rather than a crisis, and the vocabulary of the industry adjusted accordingly. But the average conceals a split that is now the dominant fact of early-stage pricing. In the first quarter of 2026, the median post-money valuation for a Series A at an AI foundation-model company was in the region of $300 million. The median Series A for a company not working on AI was around $55 million. These are not two ends of a distribution; they are two different markets that happen to share a vocabulary, a set of documents and a group of investors. Series B and Series C medians rose year on year — by about seventeen and thirteen percent respectively — but that aggregate rise is substantially an AI effect propagating through the stages. Underneath the pricing, the harder number is graduation. The proportion of seed-funded companies that reach a Series A within two years has fallen materially from where it stood for the 2018 and 2019 cohorts. The precise figure depends on the data source, the cohort definition and how long you wait, but the direction is unambiguous and the magnitude is large: something in the region of one in six of a recent seed cohort graduating within two years, where a pre-2020 cohort would have managed something closer to a third. The seed round got bigger and the bar for the next round got higher at the same time, and the gap between them swallowed a great many companies. This is why bridge rounds, seed extensions and second seeds became ordinary rather than embarrassing. A company that raised a seed round in 2022 on 2021 expectations, failed to reach the metrics a 2024 Series A required, and was not going to die, had to raise something. Often that something was a small round from existing investors on a convertible instrument at the old price or worse. Those instruments are still sitting on cap tables, and they will convert. The concentration nobody has a hedge for Diversification is the one free lunch in finance, and venture capital has spent four decades explaining that it achieves diversification through numbers: hold thirty companies, accept that most will fail, and let the distribution do its work. That argument depends on the failures being independent of one another. A portfolio of thirty companies failing for thirty unrelated reasons behaves very differently from a portfolio of thirty companies failing for one reason. In 2026, an enormous share of American venture capital is exposed to a single reason. When 86 percent of the dollars go into one technology theme, the funds deploying those dollars do not hold diversified portfolios in any meaningful economic sense. They hold one position expressed thirty ways. The companies share suppliers — a small number of chip vendors and cloud providers. They share customers, to a degree that has attracted real scrutiny, since the largest buyers of AI services are frequently also the largest investors in, and suppliers to, the companies selling them. They share a demand assumption: that enterprise and consumer willingness to pay for these systems will grow fast enough to justify the capital expenditure being committed today. If that assumption is wrong, it will be wrong for all of them simultaneously. This does not mean the assumption is wrong. It means the industry has built a portfolio whose payoff depends on one macro-scale question and has done so while describing the result as diversified. The distinction matters for anyone constructing a fund, and it matters more for a limited partner who has committed to eight venture managers and believes, reasonably, that eight managers is a portfolio. If all eight are deploying into the same theme at the same time at prices set by the same handful of marginal buyers, the LP owns one bet with eight fee streams attached. The mechanical consequence shows up in the valuations. A median AI foundation-model Series A at around $300 million post-money prices in an outcome that is very large and reasonably probable. For the fund that wrote that cheque to earn a venture return on it, the company has to exit in the billions — and not a few billion, if the fund is large and its ownership modest. There is no arithmetic under which a $300 million entry price and a $900 million exit is a good venture outcome; it is a modest multiple on a position that consumed a great deal of reserve capital and a decade of fund life. The entry price commits the investor to needing an exceptional outcome, and the price at which capital is entering commits a very large share of the industry to needing exceptional outcomes at the same time. What the drought did to behaviour Four years without distributions changed the industry's habits in ways that outlast the drought itself. Companies stayed private longer, not because private ownership is superior but because the alternative was unavailable. The consequence is that the median venture-backed company now reaches a scale, a revenue base and a shareholder count that would once have made it obviously public, while remaining private. It accumulates employees holding options that cannot be exercised without a taxable event and cannot be sold. It accumulates investors from funds whose ten-year lives are expiring. It accumulates layers of preferred stock, each with its own rights, each negotiated in a different market environment. Firms responded by building liquidity mechanisms that did not previously exist at scale. Company-sanctioned tender offers, in which the company arranges for a buyer to purchase employee and early-investor shares at a set price, moved from rare to routine at large private companies. Direct secondary purchases became a recognised strategy with dedicated funds behind it. Limited partners sold fund stakes, and general partners began moving assets into continuation vehicles rather than forcing a sale or a write-off. Reported venture secondary activity in 2026 runs into the tens of billions of dollars, heavily concentrated in the stock of SpaceX, OpenAI and Anthropic, and US venture fund stakes have been changing hands at something in the region of 78 cents on the dollar of carrying value — which is itself a market price for the credibility of venture marks. Firms also began treating reserves differently. A fund that expects its winners to exit in year seven can deploy aggressively into new companies and hold a moderate reserve. A fund that expects its winners to stay private into year twelve must hold reserves to defend its ownership through three or four more rounds, and must plan for the possibility that the fund's legal life expires before the company's private life does. Sequoia's 2021 restructuring into an open-ended vehicle was the most visible acknowledgement of this: a deliberate escape from the ten-year fund life because the ten-year fund life no longer matched the holding period of the assets. Each of these adaptations is sensible. Together they describe an asset class that has quietly changed shape — from a business of finding companies, owning them for seven years and selling them, into a business of finding companies, owning them for twelve years, and managing a complicated set of intermediate liquidity events along the way. The documents governing those funds and those companies were mostly written for the old shape. Why the arithmetic is the point Read the paragraphs above and you can construct two entirely coherent narratives. One says that venture capital has recovered: record deployment, record fund sizes, a reopened IPO market, a technology cycle of genuine historical significance. The other says that venture capital is in serious trouble: a distribution drought, collapsed graduation rates, a vintage of funds that cannot return capital, and a capital base concentrated in a handful of firms making a handful of enormous bets on a single sector. Both narratives are true of different parts of the same market, and neither is useful for deciding anything. What decides outcomes is not which story you believe but what the documents say and what the arithmetic does. A founder who raises at a $300 million post-money valuation in 2026 and exits at $400 million in 2030 has, in the language of the market, built a successful company and had a good outcome. Whether that founder makes any money depends on the size of the preference stack sitting above the common stock, on whether the preferred converts, on how many option pools were carved out of pre-money valuations along the way, and on how many convertible instruments were sitting unconverted when the price was set. Those are arithmetic questions with exact answers, and they are routinely answered wrong — not by people who cannot do the arithmetic, but by people who did not know the arithmetic was there. A limited partner deciding whether to re-up with a manager reporting 3.2x TVPI and 0.4x DPI on a 2021 fund faces the same kind of question. The answer depends on what the marks are made of, on how much of the fund's remaining value sits in one or two positions, on whether the manager has reserves left to defend those positions, and on the fee and carry structure that determines what share of any eventual realisation actually reaches the LP. Again: arithmetic, with exact answers, routinely answered badly. The rest of this book is that arithmetic. It begins where the money begins — with the question of how a deal reaches an investor at all, and why, in a market where the returns are this concentrated, being able to see a deal turns out to matter more than being able to judge it. Chapter 2 Access Is the Scarce Input A seed fund with $100 million and a four-year investment period will make something in the region of thirty first cheques. To make thirty investments it will typically look seriously at several hundred companies and encounter, in some form, a few thousand. The ratio of things seen to things bought sits somewhere between fifty and a hundred to one. Those numbers get quoted as evidence of selectivity, and investors like quoting them for that reason. They are better read as evidence of a capacity constraint. A partner who can hold perhaps two hundred substantive conversations a year, of which maybe forty become real diligence processes, is not filtering an efficient market down to its best elements. They are sampling a very large population through a very small aperture, and where they point the aperture determines most of what they will ever own. This is the first thing to understand about venture returns, and it is consistently underweighted in how the industry describes itself. The public story of venture capital is a story about judgement: the partner who saw what others missed, the contrarian bet, the thesis that turned out to be right. The private reality is that judgement operates on whatever arrives, and what arrives is determined by structural position — who returns your calls, who forwards you an introduction, who thinks of you when a company they know is about to raise. A superb picker with mediocre access will underperform a mediocre picker with superb access, and the arithmetic of the power law explains why. The base rate problem Correlation Ventures examined more than 21,000 financings between 2004 and 2013 and found that roughly 65 percent of venture deals returned less than the capital invested in them. Around four percent returned more than ten times. Data collected by Horsley Bridge across some 7,000 investments over two decades and widely cited since points the same way: about six percent of deals produced about sixty percent of the returns. Take the second number seriously and the strategic implication is uncomfortable. If six percent of deals produce sixty percent of returns, then the entire return of a portfolio is decided by whether it contains one or two members of that six percent. Everything else — whether the other ninety-four percent are ranked correctly, whether the mediocre investments return 0.4x or 1.1x, whether the failures fail quickly or slowly — moves the fund's result by relatively little. Now consider what that does to the value of selection skill. Suppose an investor is genuinely good, and can identify, among the companies they see, which ones are more likely to be exceptional. That skill is worth a great deal if the exceptional companies are in the set they see. It is worth nothing at all if they are not. Selection skill is a multiplier on access; it cannot substitute for it. A fund that never gets shown the six percent can rank its own deal flow perfectly and still return 0.8x. There is a corresponding error on the other side, which is to conclude that judgement does not matter and that the correct strategy is to index the market by writing very small cheques into everything. This fails for a different arithmetic reason, which is that ownership determines what a winner is worth to you. Owning one percent of a company that exits at $5 billion returns $50 million before dilution and preference effects — transformative for a $30 million fund, irrelevant for a $500 million one. Indexing gets you exposure to the six percent at a position size too small to matter. The problem is genuinely two-sided: you need to see the exceptional companies and you need to own enough of them. Where deals actually come from The phrase "proprietary deal flow" is used to mean a deal that only one investor sees. At the seed stage such deals exist, and they are a real source of edge for investors with unusual networks — a former operator whose ex-colleagues call them first, an investor embedded in a specific technical community, a fund with scouts inside large technology companies. A scout programme, in which a fund gives small amounts of capital to well-connected operators to invest on its behalf in exchange for a share of the upside, is a purchase of access, quite explicitly. At Series A and beyond, genuinely proprietary deals are close to nonexistent, and where they exist they should be treated with suspicion. A company with strong metrics raising a Series A will run a process, because its seed investors will tell it to. If a good Series A opportunity is available to exactly one firm at a price no one is competing on, the likeliest explanation is not that the firm has magical access. It is that the other firms saw it and passed. This is adverse selection, and it is the most reliable way for a fund with weak access to lose money: it will consistently win the deals it should not have won. What actually separates funds at the competitive end of the market is not exclusivity but preference — whether a founder choosing between three term sheets chooses yours. That preference is manufactured by brand, by the specific partner's reputation with founders who have worked with them, by the reference calls founders make to the firm's existing portfolio, and by speed. A fund that can go from first meeting to signed term sheet in nine days beats a fund that takes five weeks, not because founders are impatient but because a fast process costs the founder less of the thing they cannot replace, which is their own attention. Speed has a price, and the price is diligence. The 2021 market discovered how high that price can be: firms competing for allocation compressed diligence to the point where multi-hundred-million-dollar cheques were written on the basis of a deck and a reference call, and a number of those cheques went into companies whose reported metrics did not survive contact with an audit. The discipline that separates a fast process from a reckless one is knowing in advance which questions are the ones that can kill the deal, and asking those first. Everything else can be done after the term sheet. Accelerators occupy a peculiar position in this system. Y Combinator has become, in effect, a scheduling mechanism for the seed market: it concentrates several hundred companies into a two-day period, which both solves the access problem for investors and destroys their ability to move at a differentiated pace. Batch companies are seen by everyone, priced competitively and closed fast. The accelerator has converted access into a queue, and the queue is efficient, which means the returns available from standing in it are ordinary. Signalling, and the cost of the wrong investor Every investor on a cap table transmits information about the company to every future investor, and the information is not always the information the founder intended. The clearest case is the multi-stage fund writing a seed cheque. A firm that manages several billion dollars and invests from seed through growth will sometimes put $2 million into a seed round. For the firm this is cheap optionality: a call option on the right to lead the Series A, priced at a rounding error against the fund. For the founder it looks like validation from a famous name. What it actually creates is a conditional signal. When the company raises its Series A, every prospective lead will ask the same question: is the existing large investor leading this round? If the answer is no, the prospective lead must explain to itself why the firm with the most information about this company, the most capital available, and an explicit strategy of following its seed investments, has chosen not to. Sometimes there are good answers — portfolio conflict, a partner departure, a fund at the end of its investment period. Usually there are not, and the round becomes materially harder to raise. A seed fund that cannot lead a Series A transmits no such signal, because nobody expected it to. This is a real, if unglamorous, advantage of small specialised funds, and founders systematically underprice it in favour of the brand on the term sheet. Signalling runs in the other direction too, and here the founder has some control. Taking money from an investor who is conspicuously good at one thing tells the market that the company's problem is that thing. A round led by a firm known for enterprise go-to-market says the company has a product and needs distribution. A round led by a deep-technology specialist says the risk is technical. These signals are read, and they shape which firms self-select into the next round. The most damaging signal is the one founders create accidentally by running a round badly. A round that has been in the market for four months, has been seen by forty firms and has no lead is a public fact in a small industry. Investors talk. A partner deciding whether to spend two weeks on a company will ask a friend at another firm whether they have seen it, and "yes, months ago, we passed" ends the conversation. This is why experienced founders run rounds as compressed processes with a hard start date, and why they would rather fail fast with fifteen firms than drift with forty. How the decision actually gets made Deal flow arrives at a partnership, and partnerships have decision structures that shape outcomes as much as any individual's judgement does. The standard arrangement is that one partner sponsors a deal and carries it through the firm's process. That partner does the work, writes the memo, brings the founders to a partner meeting, and — in most firms — will sit on the board afterwards. Because the sponsoring partner's reputation inside the firm is attached to the investment, the firm's internal incentives push toward conviction rather than consensus. A deal that everyone is mildly positive about is usually a worse investment than a deal that one partner will fight for and two others think is wrong, and the better firms have explicit mechanisms to protect the second kind from being voted down. This matters because of what the power law does to the distribution of opinion. The companies that become enormous are, at the time of investment, almost always the subject of a plausible and articulate objection. Airbnb was strangers sleeping in strangers' homes. The objection is not stupid; it is the reason the opportunity exists at the price it exists. A partnership that requires unanimity systematically excludes exactly the deals where the return lives, because unanimity selects for the absence of a serious objection, and the absence of a serious objection usually means the absence of a serious mispricing. Firms that have survived multiple cycles tend to run some version of a rule that a single partner with sufficient conviction can commit a defined amount of capital without the others agreeing. The diligence itself divides into things that can be verified and things that cannot. Revenue can be verified: a diligence process that does not pull the actual billing data and reconcile it against the deck is not a diligence process. Customer concentration, churn cohorts, gross margin after the cost of serving customers rather than before it, the proportion of contracted revenue that is genuinely recurring rather than a pilot that renews annually if someone remembers — all of these are checkable, and all of them were routinely not checked in 2021, with predictable consequences. What cannot be verified is the thing the return depends on: whether this team, in this market, will do something that has not been done. The honest position is that diligence reduces the probability of an obvious error rather than raising the probability of an exceptional outcome. Reference calls are the closest available instrument, and their value lies almost entirely in the off-list references — the people the founder did not name, found through the investor's own network, who have no reason to manage the conversation. An on-list reference tells you the founder can choose references. Several firms have tried to systematise the unverifiable part, and the results are instructive. Quantitative sourcing models that rank companies on hiring velocity, web traffic, developer adoption or repeat-founder status do find real signal, and several firms built respectable portfolios on them during the 2010s. What they do not appear to do is find the tail. The characteristics that a model can observe are, by construction, characteristics that other investors can also observe, which means they are priced. The tail outcome tends to look, in the data available at the seed stage, like an ordinary company with an unusual founder, and the unusual part is not in the dataset. What sourcing looks like when one sector takes most of the money The 2026 market has bent all of this in a specific direction. When a single theme absorbs the overwhelming majority of dollars, competition for the companies in that theme becomes extreme and competition for everything else collapses. Inside the theme, rounds are frequently preempted: an investor who has been tracking a company offers a term sheet before the company has decided to raise, at a price above where a competitive process would clear, precisely to avoid the process. Preemption is a transfer from the investor to the founder, and it is rational for the investor only if the alternative is losing the deal entirely. It is also a reliable indicator that prices are high, because nobody preempts in a buyer's market. Outside the theme, the opposite dynamic applies. A capable company building unglamorous software for an unglamorous industry is raising in a market with very few active buyers, and the terms reflect it. Median non-AI Series A pricing sits at roughly a sixth of AI foundation-model pricing. For a fund willing to operate there, access has stopped being the binding constraint — those companies would very much like to meet you — and selection has become the constraint again, along with the harder question of whether a company priced reasonably in a sector with no capital flowing into it can ever raise the rounds it will need to become large. For the founder, the practical consequence is that a fundraise is a scheduling problem before it is a persuasion problem. The mechanics that work are unglamorous and consistent: build the list of target firms before any meeting happens, rank it, and get warm introductions to the whole list within the same ten-day window, so that first meetings cluster rather than stretch. Take the lower-priority firms first, because the pitch improves with repetition and the early meetings are where you discover which objection you have not yet answered. Refuse to send a deck to a partner who will not take a meeting, since a deck circulating without a conversation attached generates passes that stay on the record. And decline to name a price in the first meeting: the price is set by whether there is competition, and there is no competition until several firms are in diligence simultaneously. The single most common founder error is treating the round as a sequence rather than a process — approaching firms one at a time, waiting for each answer before starting the next conversation. A sequence takes six months, produces no competition, and by the fourth month has generated the public fact that the company has been raising for a long time. The identical set of firms approached simultaneously produces a decision in six weeks and at least the possibility of two term sheets, which is the only thing that moves price. This is the trade that defines portfolio strategy in a concentrated market. You can compete for access in the sector where the returns are believed to be, and pay entry prices that require exceptional outcomes to work. Or you can have easy access in sectors where entry prices are reasonable, and accept that the next round is not guaranteed to exist. Both are defensible. What is not defensible is assuming the second market's prices with the first market's exit assumptions, which is the error that produced the 2021 vintage. Underneath both is the same structural truth. The company that will produce most of a fund's return is out there being formed right now, and the fund's entire result depends on whether anyone in that company's world happens to think of calling you. Everything else in this book — the cap-table mathematics, the preference stacks, the reserve models, the waterfalls — operates on the set of companies that made it through that aperture. Widening the aperture is the highest-leverage thing a venture firm does, and it is almost never the thing that appears on the strategy slide. The moment a company does arrive, and the conversation turns to price, the arithmetic begins — and the first thing it does is establish who owns what. Chapter 3 Pre-Money, Post-Money, and the Pool Somebody Has to Pay For A founder is offered $2.5 million at a $7.5 million pre-money valuation. She owns 8,000,000 shares, which is all the stock that exists. The offer sounds like it gives away a quarter of the company, and it does — but only under one of four sets of terms she is likely to be shown, and under the other three she will end up with between 47 and 64 percent. Every one of those outcomes is consistent with the sentence "$2.5 million at a $7.5 million pre-money". The difference is entirely in the definitions. This is not a trick, and the investors proposing these structures are not being dishonest. The definitions are standard, they are written down in the stock purchase agreement, and they are the same ones every experienced participant uses. They are simply not the definitions a reasonable person would infer from the headline. Shares, not percentages The first discipline is to stop thinking in percentages and start thinking in shares and a price per share. Percentages are outputs. Shares and prices are the things the documents actually create. Post-money valuation is defined as pre-money valuation plus the amount invested. Here that is $7.5 million plus $2.5 million, or $10 million. The new investor's ownership is the amount invested divided by the post-money: $2.5 million divided by $10 million, which is 25 percent. To turn that into shares, ask how many new shares must be issued so that they represent 25 percent of everything outstanding afterwards. Call the new shares X. Then X divided by 8,000,000 plus X equals 0.25, which rearranges to X equals 8,000,000 divided by 3, or 2,666,667 shares. The price per share is the investment divided by the shares: $2.5 million divided by 2,666,667, which is $0.9375. There is a shortcut worth memorising, because it is the identity that makes everything else in this chapter legible. The price per share in a priced round equals the pre-money valuation divided by the pre-money fully diluted share count. Here: $7.5 million divided by 8,000,000 shares equals $0.9375. Same answer. The pre-money valuation is, definitionally, a valuation of the shares that exist before the money arrives, which means every share that is deemed to exist before the money arrives shares in that $7.5 million. That last sentence is where the money is. "Fully diluted" does not mean shares outstanding. It means shares outstanding plus every share that could come into existence through an instrument already issued: options granted to employees, options authorised but not yet granted, warrants, and — this is the contested part — options that do not yet exist but that the new investor requires be authorised as a condition of the investment. The option pool shuffle No sensible investor funds a company with no stock available for future employees. A Series A company that intends to hire a head of engineering, a head of sales and thirty others over the next two years needs equity to offer them, and it needs to know how much before it hires. So the term sheet will specify an unallocated option pool — typically between 10 and 20 percent of the post-money capitalisation at a priced seed or Series A — to be in place at closing. The question that determines who pays for it is whether the pool is counted in the pre-money fully diluted share count or created afterwards. Take the same deal with a 15 percent post-money unallocated pool carved out of the pre-money. Now three ownership constraints must hold simultaneously: the investor holds 25 percent, the pool is 15 percent, and the founder's 8,000,000 shares are whatever is left, which is 60 percent. If 8,000,000 shares are 60 percent of the total, the total is 8,000,000 divided by 0.60, or 13,333,333 shares. The investor therefore receives 25 percent of that, or 3,333,333 shares, and the pool is 15 percent of it, or 2,000,000 shares. The price per share is now $2.5 million divided by 3,333,333 shares, which is $0.75. Check it against the identity. The pre-money share count is the founder's 8,000,000 plus the pool's 2,000,000, or 10,000,000 shares. The pre-money valuation of $7.5 million divided by 10,000,000 shares is $0.75. Consistent. The pool was counted as part of the pre-money company, so it drew part of the pre-money value. Now look at what happened to the founder. Her 8,000,000 shares are worth 8,000,000 times $0.75, which is $6 million. Not $7.5 million. The $1.5 million difference is exactly the value of the 2,000,000 pool shares at the round price, and it came entirely out of her pocket. The investor still paid $2.5 million and still owns 25 percent; nothing about the investor's position changed. The pool was funded one hundred percent by the existing holders, which in this case means the founder alone. The alternative — creating the pool after the round closes — distributes the cost differently. Issue the investor's 2,666,667 shares at $0.9375, bringing the total to 10,666,667. Then authorise a pool equal to 15 percent of the new total: if P divided by 10,666,667 plus P equals 0.15, then P equals 10,666,667 times 0.15 divided by 0.85, or 1,882,353 shares. The company now has 12,549,020 shares. The founder holds 8,000,000 of them, or 63.75 percent. The investor holds 2,666,667, or 21.25 percent. The founder is 3.75 percentage points better off and the investor is 3.75 points worse off, and those 3.75 points are precisely 25 percent of the 15 percent pool — the investor's pro rata share of a cost they did not have to bear under the first structure. Founders negotiating a round routinely spend their effort on the pre-money number and accept the pool size as a technical matter for the lawyers. This gets the priorities backwards. Under a pre-money pool, moving the pool from 15 percent to 10 percent has a larger effect on the founder than most realistic movements in the headline valuation. Run it: with a 10 percent pool, the founder's 8,000,000 shares are 65 percent of the total, so the total is 12,307,692 shares, the investor takes 25 percent or 3,076,923 shares, and the price per share is $0.8125. The founder's stake is worth $6.5 million instead of $6 million. To get the same $6.5 million while conceding the 15 percent pool, she would need the post-money to rise to $10.83 million, which means the pre-money would have to go from $7.5 million to roughly $8.33 million — an eleven percent increase in the headline number. Almost no investor will move the valuation eleven percent in a negotiation. A great many will move the pool by five points, particularly if the founder arrives with a hiring plan showing what the company actually needs to grant over the next eighteen months, which is the only honest basis for sizing a pool in the first place. When a note converts in the same round The arithmetic above assumed a clean cap table. Most are not clean, because most companies raise something before they raise a priced round, and that something is usually a convertible instrument — a note or a SAFE — that turns into equity when the priced round closes. Suppose our founder previously raised $1 million on a convertible note with a $5 million valuation cap. A cap is a ceiling on the valuation at which the earlier money converts: whatever the priced round's valuation turns out to be, the note holder converts as though the company had been worth no more than $5 million. If the round prices above the cap, the note holder gets a better price per share than the new investor. The difficulty is that this creates a genuine circularity. The note's conversion price depends on the pre-money share count. The pre-money share count includes the option pool. The pool is defined as a percentage of the post-money total. The post-money total includes the shares issued to the note holder. Each quantity is defined in terms of the others, and there is no order in which you can compute them one at a time. The way out is to solve the constraints simultaneously rather than sequentially. Under the standard, investor-drafted convention, the note converts inside the pre-money, meaning the note shares are part of the pre-money capitalisation and the new investor still receives a full 25 percent. So three things must hold at once: the investor's shares are 25 percent of the final total, the pool is 15 percent of the final total, and the note holder receives $1 million worth of shares priced at the $5 million cap divided by the pre-money count applicable to the cap. Call the final total T. The founder's 8,000,000 and the note shares together must make up the remaining 60 percent. The pool is 0.15T. The note converts at $5 million divided by the founder's shares plus the pool — that is, $5 million divided by 8,000,000 plus 0.15T — and the number of note shares is $1 million divided by that price, which works out to one fifth of 8,000,000 plus 0.15T, or 1,600,000 plus 0.03T. Now impose the constraint. The founder's shares plus the note shares equal 0.60T, so 8,000,000 plus 1,600,000 plus 0.03T equals 0.60T. That gives 9,600,000 equals 0.57T, so T is 16,842,105 shares. Everything else falls out. The pool is 2,526,316 shares. The note converts at $5 million divided by 10,526,316, which is $0.475 a share, and buys 2,105,263 shares. The new investor's 25 percent is 4,210,526 shares, and the round price is $2.5 million divided by that, or $0.59375 — which is also $7.5 million divided by the 12,631,579 pre-money shares, as the identity requires. The founder now holds 8,000,000 of 16,842,105 shares, or 47.5 percent. The note holder has 12.5 percent. The pool has 15. The investor has 25. Two things about this deserve attention. The first is that the $1 million note cost the founder 12.5 points of ownership, and at the round price of $0.59375 her stake is worth $4.75 million rather than the $6 million she would have held without it. The second is that the new investor was not diluted at all. They paid $2.5 million and received exactly 25 percent, the same as if no note existed. The entire dilution from the converting instrument fell on the people who were already there. That is the drafting convention, not a law of nature. If the round is papered so that the note converts on top of the pre-money — so the pre-money valuation is a valuation of the founder's stock and the pool only, and the note shares are additional — the numbers change materially. The round price becomes $7.5 million divided by the founder's shares plus the pool, the note price is two thirds of that, and when you work the constraints through, the founder ends at roughly 50.2 percent, the note holder at 13.0, the pool at 15, and the new investor at 21.7 percent rather than 25. The new investor has absorbed their share of the earlier instrument. Nearly every model a founder builds assumes the second convention, because it is the intuitive one: new money arrives, everybody who was there before gets diluted together. Nearly every term sheet specifies the first. The gap between the two, in this small example, is 2.7 points of founder ownership and 3.3 points of investor ownership — and in a company with several million dollars of accumulated convertible instruments and a large pool, the gap routinely runs to ten points or more. Structure New investor Founder Option pool Note holder Price per new share No pool, no converting note 25.00% 75.00% — — $0.9375 15% pool carved out of the pre-money 25.00% 60.00% 15.00% — $0.7500 15% pool created after the round closes 21.25% 63.75% 15.00% — $0.9375 15% pre-money pool, $1M note at a $5M cap converting inside the pre-money 25.00% 47.50% 15.00% 12.50% $0.59375 Table 1: The same $2.5 million investment at a stated $7.5 million pre-money valuation, under four sets of definitions. Founder holds 8,000,000 shares before the round in every case. The rest of the stack A capitalisation table has several other inhabitants, and each of them can move a founder's eventual outcome by more than a round of valuation negotiation. Founder stock almost always vests, even though the founders already own it. The mechanism is reverse vesting: the founders keep their shares, but the company retains the right to repurchase the unvested portion at cost if they leave. A standard schedule runs four years with a one-year cliff, so a founder who leaves after ten months typically keeps nothing. Investors insist on this because the company's most valuable asset is the founders' continued attention, and because a co-founder who departs in year two holding thirty percent of the company outright is a fatal defect that no future investor will fund around. Founders who have already been working for eighteen months when the round closes should negotiate credit for that time, and usually get it. The tax treatment of that arrangement turns on a single filing. Section 83(b) of the Internal Revenue Code lets a holder of restricted stock elect to be taxed on its value at grant rather than as it vests. At grant, when the shares are worth essentially nothing, the tax is essentially nothing. Without the election, each tranche is taxed as ordinary income as it vests, at whatever the stock is worth then, which for a successful company becomes ruinous. The election must be filed within thirty days of the grant, and the deadline is absolute. It is the cheapest and most consequential piece of paperwork in a startup's first year. Acceleration provisions govern what happens to unvested stock on a sale. Single-trigger acceleration vests everything on a change of control; double-trigger vests it only if the change of control is followed by the holder being terminated without cause or resigning for good reason. Acquirers strongly prefer double-trigger, because they are paying for a team and single-trigger hands that team the proceeds and the freedom to leave on closing day. Double-trigger is market for founders and executives; broad single-trigger acceleration across a whole company is a real impediment to an acquisition and will be negotiated away, usually by the acquirer reducing the price. Then there are warrants, which are options issued to someone other than an employee — a venture debt lender, a strategic partner, a landlord. A venture debt facility typically carries warrant coverage of somewhere between five and twenty percent of the loan amount, meaning a $5 million facility might come with warrants over $500,000 of stock at the last round price. They dilute like any other option, they sit in the fully diluted count, and they are easy to forget because they live in a loan document rather than the equity file. Finally, the common stock and the preferred stock are not worth the same amount, and the company is required by law to say so. Section 409A of the tax code requires options to be granted with an exercise price at or above the fair market value of the common stock, established by an independent valuation. Because common lacks the liquidation preference, the protective rights and the board seat that the preferred carries, it prices at a substantial discount — often somewhere between a fifth and two fifths of the latest preferred price at early stage, narrowing as a company approaches an exit and the preference becomes less likely to bite. A company that has just raised at $10 a share may quite properly issue employee options at $2.50. The spread is not an accounting fiction. It is the market's estimate of how much of the company the preferred stack has already claimed. What to insist on seeing The defence against all of this is procedural and unglamorous: never accept a headline and never accept a percentage. Ask for the pro forma capitalisation table — the share-by-share, holder-by-holder schedule of who owns what immediately after closing — and ask for it before signing the term sheet rather than during the closing mechanics, when it is presented as a formality and nobody wants to reopen the deal. Three questions resolve almost every ambiguity. Is the option pool included in the pre-money fully diluted capitalisation? Do converting notes and SAFEs count inside the pre-money or outside it? And what, precisely, is the resulting price per share? The third question is a useful trap, because the price per share is the one number that cannot be rhetorically softened. If the investor says $7.5 million pre-money and the price per share comes out at $0.59, the founder knows that the effective valuation of her own stock is $4.75 million and can decide what she thinks about that with the actual figure in front of her. None of this makes the deal bad. A round that leaves a founder at 47.5 percent after clearing a note and funding a real option pool may be an excellent round. But it should be understood as a $4.75 million valuation of her holdings rather than a $7.5 million one, because the next conversation — about what she gets paid in an exit — starts from the number of shares she actually owns, and no amount of goodwill at signing changes that number afterwards. And the instruments that caused most of the trouble here were not priced rounds at all. They were the small, fast, cheap financings that companies now use for their first several million dollars, and which have their own arithmetic. Hashtags: #VentureCapitalPortfolioStrategy #DealSourcing #TermSheets #ExitMechanics #VentureCapital #PortfolioConstruction #PowerLawReturns #DealFlow #ProprietaryDealFlow #InvestorAccess #VentureSelection #OwnershipTargets #VentureFundStrategy #CapitalConcentration #TVPI #DPI #VentureDistributions #FollowOnReserves #CapTableModeling #OptionPoolShuffle #ConvertibleInstruments #LiquidationPreferences #ExitWaterfalls #VentureReturns #FutureOfVenturePortfolioStrategy

Download the Book (PDF): Introduction A venture-backed company is a set of promises about a future that has not happened yet. The founders promise to build something valuable. The investors promise money now in exchange for a particular slice of whatever gets built. The employees accept options that are worth nothing unless the company is eventually sold or listed at a price well above what the investors paid. Every one of those promises is written down, usually in a few hundred pages of documents that most of the people who sign them never read closely. The documents sit in a data room for years. Then one day a buyer makes an offer, or the market turns, or the money runs out, and the documents suddenly decide who gets what. This book is about those documents and the law that surrounds them. It is written for lawyers who advise founders, investors, and early-stage companies, and for lawyers who are about to start. It is also written for the founders, finance leads, and investors who have to live with what their lawyers negotiate and who want to understand why a single clause in a certificate of incorporation can move tens of millions of dollars between two groups of people who, at the time of signing, believed they were on the same side. The argument of this book The book makes one argument, and every chapter tests it against a different part of the life cycle. The argument is this: every term in a venture financing is a claim on a future exit, and the practitioner's job is to price those claims in numbers before the documents are signed, and then to run the process by which they are later enforced, because the Delaware courts judge conflicts between preferred and common stockholders far more by how the board behaved than by what the paper said. That argument has two halves, and both are neglected in practice. The first half is arithmetic. A liquidation preference, an anti-dilution adjustment, a valuation cap on a SAFE, or an option pool increase negotiated into a pre-money valuation is not a matter of legal drafting taste. Each one moves a precise number of dollars at a precise exit value, and the only way to understand what a client is agreeing to is to model it. A founder who is told that a 1x participating preference is "market" in some sector has learned nothing. A founder who is shown that, on a capitalization like her company's, the same preference transfers more than $14 million from common to preferred at a $75 million sale has learned what she is being asked to give up. This book therefore contains worked numerical examples throughout. They are hypothetical and are labeled as such; the companies in them do not exist. The mechanics they illustrate are the mechanics of the standard documents. The second half is process. The venture documents give preferred stockholders a bundle of contractual rights, and give the board, which the preferred holders often control, the discretion to decide when and how the company pursues a sale, a financing, a redemption, or a wind-down. Delaware law, which governs most venture-backed companies, treats the preferred holders' special rights as contractual and generally treats the directors' fiduciary duties as running to the common stockholders as the residual owners of the enterprise. When the interests of the two groups diverge, as they reliably do in a modest sale or a down round, the Court of Chancery asks whether the board understood the conflict and dealt with it through a fair process. In the leading cases discussed in Chapter 8, conflicted boards escaped damages only because the common stock was found to be worth nothing anyway. That is a narrow and expensive escape. The lesson is not that the contract controls, but that the contract controls only if the process holds. What the book covers The chapters follow the order in which a company meets the problems. Chapter 1 begins at formation: why nearly every venture-backed company is a Delaware corporation, how founder stock is issued and vested, why the Section 83(b) election is the most consequential piece of paper a founder will sign in the first month, and how the qualified small business stock exclusion under Section 1202 of the Internal Revenue Code was substantially expanded by the One Big Beautiful Bill Act in July 2025. Chapter 2 addresses the securities law that governs every private round: Regulation D, the two versions of Rule 506, the accredited investor definition, and the SEC staff's March 2025 guidance that made it far easier to verify investor status in a Rule 506(c) offering. Chapter 3 turns to the instruments that dominate early money, the post-money SAFE and the convertible note, and shows with a worked example how several SAFEs stack into dilution that founders routinely underestimate. Chapters 4 through 7 are the core of the negotiation. Chapter 4 takes the reader through a priced round on the National Venture Capital Association's model documents, including the October 2025 revisions, and works through a full capitalization table from formation to a Series A, including the option pool shuffle that quietly lowers the effective pre-money valuation. Chapter 5 addresses control: board composition, protective provisions, the voting agreement, and the 2024 amendment to the Delaware General Corporation Law that responded to the Court of Chancery's decision in West Palm Beach Firefighters' Pension Fund v. Moelis & Co. Chapter 6 is devoted to liquidation preferences, with a full exit waterfall at several sale prices. Chapter 7 addresses anti-dilution protection and down rounds, comparing full-ratchet and weighted-average formulas on the same hypothetical cap table, and then pay-to-play provisions and recapitalizations. Chapter 8 is the doctrinal center of the book: the fiduciary duties of directors when preferred and common stockholders want different things. It covers In re Trados Inc. Shareholder Litigation (2013), the Frederick Hsu Living Trust v. ODN Holding Corp. litigation (2017 and 2020), and the line of cases in which venture investors were held to account for dilutive financings, and then turns to the 2025 amendments to the Delaware General Corporation Law, the Delaware Supreme Court's February 2026 decision upholding them, and the argument over whether companies should leave Delaware for Texas or Nevada. Chapter 9 addresses the sale of the company: drag-along rights and their enforceability, appraisal waivers, management carve-outs, the golden parachute rules, and the post-closing mechanics that decide how much of a headline price actually reaches the stockholders. Chapter 10 addresses liquidity before and at the exit: secondary sales and company tender offers, the initial public offering and the direct listing, and the market conditions of 2025 and 2026, which produced both the largest IPO in history and a sobering reminder of how far public prices can fall from private valuations. A note on currency and scope The law in this area moved quickly between 2024 and 2026. Delaware amended its corporation statute twice in two years, first to validate stockholder agreements and then to create statutory safe harbors for conflicted transactions. Congress changed the qualified small business stock rules. The SEC staff relaxed accredited investor verification for certain offerings, the SEC allowed a shorter offering period for certain tender offers, the IRS began accepting Section 83(b) elections online, and the NVCA revised its model documents to address national security rules and tranched financings. The book states the law as it stood in late 2026, and gives dollar thresholds with the year to which they apply. Some matters were still in motion at the time of writing, including congressional proposals to rewrite the accredited investor definition and a proposal to create a dedicated business court in Nevada. Where the law is unsettled, the book says so. The focus is United States law, and principally Delaware corporate law and federal securities and tax law, because that is where the overwhelming majority of venture financings are documented. Comparative points appear where they help, most obviously in the discussion of Texas and Nevada as alternative domiciles. Finally, the book is written to explain, not to substitute for advice on a particular transaction. Nothing in it is legal, tax, or investment advice. Chapter 1: Founding the Company: Entity, Founder Stock, and the Tax Clocks The decisions made in the first month of a company's life are the cheapest to get right and the most expensive to fix. A founder who forms the wrong entity, issues stock without vesting, forgets to assign the code she wrote before incorporation, or misses a thirty-day tax deadline will usually not feel the consequences for years. When she does feel them, typically in diligence for a Series A or an acquisition, the fix costs legal fees, tax, and leverage at exactly the moment she can least afford to lose any of the three. This chapter covers the formation choices that matter most for a company that expects to raise venture capital: the choice of entity and state, the issuance and vesting of founder stock, the assignment of intellectual property, the Section 83(b) election, and the qualified small business stock exclusion, whose clock starts running the day the stock is issued. The Delaware C corporation and why investors insist on it Venture-backed companies are overwhelmingly organized as Delaware corporations taxed under Subchapter C of the Internal Revenue Code. The reasons are practical rather than ideological, and a lawyer should be able to explain each of them to a founder who has been told that a limited liability company is simpler or cheaper. The first reason is the investors' own structure. Most venture funds are limited partnerships whose limited partners include pension funds, university endowments, foundations, and foreign investors. A fund that invests in a pass-through entity such as an LLC taxed as a partnership passes the operating company's income through to those limited partners. For tax-exempt limited partners, that income can be unrelated business taxable income; for foreign limited partners, it can be income effectively connected with a US trade or business, which triggers US filing and withholding obligations. Fund agreements commonly restrict or prohibit investments that generate these problems. A C corporation stops the income at the corporate level and removes the issue. The second reason is the equity toolkit. Preferred stock with a liquidation preference, conversion rights, protective provisions, and anti-dilution adjustments is a creature of corporate law, and its terms have been tested in decades of Delaware litigation. Incentive stock options under Section 422 of the Code can only be granted by a corporation. The qualified small business stock exclusion under Section 1202, discussed below, is available only for stock of a domestic C corporation. An LLC can approximate some of these features through profits interests and carefully drafted operating agreements, but every investor, every acquirer, and every underwriter would have to learn the bespoke arrangement from scratch. The third reason is Delaware itself. The Delaware General Corporation Law is flexible and well understood; the Court of Chancery is a specialized court of equity whose judges decide corporate disputes without juries; and the body of case law interpreting preferred stock rights and director duties is deeper than anywhere else. The standard forms used in venture financings, including the model documents of the National Venture Capital Association discussed in Chapter 4, are drafted for Delaware. That dominance has been challenged since 2024 by Texas and Nevada, and Chapter 8 addresses the argument for leaving. For a company at formation, however, a Delaware corporation remains the default that investors expect, and a founder who chooses something else should have a specific reason and should expect to spend time explaining it. A company that begins as an LLC for good reasons, such as passing early losses through to the founders, can later convert to a corporation. The conversion is routine but not free: it requires tax analysis, and it matters for Section 1202, because the qualified small business stock holding period for stock received in the conversion generally begins at the conversion, and the stock's basis for purposes of the gross assets test and the ten-times-basis limitation is tied to the fair market value of the property contributed at that time. A founder who expects to raise institutional money within a year or two is usually better served by starting as a corporation. Founder stock, vesting, and the assignment of intellectual property At formation the company authorizes a number of shares of common stock, commonly ten million or more at a very low par value, and issues founder shares under a restricted stock purchase agreement. The low par value keeps the purchase price trivial and keeps Delaware franchise tax manageable under the assumed par value capital method. The founders pay for their shares with a small amount of cash and, critically, with an assignment to the company of all intellectual property related to the business that they created before incorporation. That second element is not a formality. Code written in a founder's spare time before the company existed belongs to the founder unless it is assigned. Investors will not fund, and acquirers will not buy, a company that does not own its core technology. Every employee and consultant should also sign an agreement assigning inventions made in the course of the work. In California, Labor Code section 2870 limits the reach of such assignments: an employer cannot claim an invention that an employee developed entirely on her own time without the employer's equipment, supplies, facilities, or trade secrets, unless the invention relates to the employer's business or anticipated research or results from work performed for the employer. Well-drafted assignment agreements carve out the protected category and include the statutory notice. Diligence teams look for signed agreements from every person who has written code or designed product, including founders who left early and contractors located abroad. Founder stock is almost always subject to vesting, even though the founders have paid for it. The standard structure gives the company a right to repurchase unvested shares at the lower of cost or fair market value if the founder leaves, with the repurchase right lapsing over four years, typically with a one-year cliff: nothing vests for twelve months, one quarter vests at the first anniversary, and the rest vests monthly over the following thirty-six months. Founders sometimes resist vesting on their own stock as an insult. The practical answer is that vesting protects each founder from the others. If one of three founders walks away after four months holding a third of the company fully vested, the remaining two are building value for a departed partner, and the next investor will price that problem into the round or refuse to invest until the stock is re-vested. Vesting agreements also address acceleration. Single-trigger acceleration vests some or all unvested shares on a change of control. Double-trigger acceleration vests them only if a change of control is followed, within a defined period, by a termination without cause or a resignation for good reason. Acquirers dislike single-trigger acceleration because it removes the retention value of the equity they are buying; investors dislike it because it can make the company harder to sell. The market settled long ago on double-trigger acceleration for founders, commonly covering all or a large portion of unvested shares, and on no acceleration or double-trigger for most employees. The question is worth settling at formation, because once institutional investors arrive, changing the terms of founder vesting becomes part of the negotiation. The Section 83(b) election Section 83 of the Internal Revenue Code governs property transferred in connection with the performance of services. When stock is subject to a substantial risk of forfeiture, such as a company repurchase right that lapses only if the founder keeps working, Section 83(a) defers the tax event to the date the risk lapses, and taxes the difference between the stock's fair market value at vesting and the amount paid as ordinary compensation income. For a founder whose stock vests monthly while the company's value rises, that deferral is a trap. A hypothetical shows how the trap works. A founder buys 5,000,000 shares of common stock at formation for $0.0001 per share, paying $500, subject to four-year vesting with a one-year cliff. She does not file an 83(b) election. At her first anniversary, 1,250,000 shares vest. By then the company has raised money on SAFEs and has obtained an independent valuation of its common stock at $0.40 per share. She recognizes $500,000 of ordinary income, less the tiny amount she paid for those shares, and owes tax on it at ordinary rates without having received any cash. The following year the company closes a Series A, and the common stock is valued at $0.90 per share. Each month roughly 104,167 more shares vest, generating about $93,750 of additional ordinary income every month. She cannot sell the shares to pay the tax because there is no market for them, and the transfer restrictions in her agreements would prohibit a sale anyway. Section 83(b) lets the founder elect to include the value of the stock in income at the time of transfer rather than at vesting. Because she paid fair market value at formation, the amount included is zero. Her holding period for capital gains begins immediately, and all future appreciation is capital gain taxed only when she sells. The election must be filed with the IRS within thirty days after the transfer of the stock. There is no extension and, in practice, no cure. Missing the deadline is one of the few formation errors that cannot be repaired by later paperwork, and diligence teams check for proof of timely filing. The filing mechanics changed recently. The IRS released Form 15620 in late 2024 as a standard form for the election, and in mid-2025 it began accepting the election electronically through an individual IRS online account. A taxpayer may file online or by mail, but should use only one method. The taxpayer must still give a copy of the election to the company, and practitioners continue to advise keeping the electronic confirmation or, for a mailed election, certified mail proof of the postmark date. The requirement to attach a copy of the election to the taxpayer's income tax return was eliminated by regulations issued in 2016. The same analysis applies to employees who receive restricted stock or who early-exercise options into restricted stock. It does not apply to ordinary options, which are not property transferred for purposes of Section 83 until exercise. Employee option grants raise a different tax question, the fair market value of the common stock under Section 409A, which Chapter 4 addresses alongside the option pool. The qualified small business stock exclusion Section 1202 of the Code allows a noncorporate taxpayer who acquires qualified small business stock at original issuance and holds it for the required period to exclude some or all of the gain on its sale from federal income tax. For founders and early investors in a successful company, it is often the single most valuable tax provision they will encounter, and its requirements need attention from the first day because several of them are tested at the moment the stock is issued. The core requirements have been stable for years. The issuer must be a domestic C corporation. The stock must be acquired at original issuance from the corporation, in exchange for money, property other than stock, or services; stock bought from another stockholder does not qualify. At the time of issuance and immediately afterward, the corporation's aggregate gross assets must not exceed a dollar ceiling. During substantially all of the taxpayer's holding period, at least 80 percent of the corporation's assets by value must be used in the active conduct of a qualified trade or business. Excluded trades include businesses in which the principal asset is the reputation or skill of employees and professional services fields such as health, law, engineering, architecture, accounting, actuarial science, performing arts, consulting, athletics, financial services, and brokerage, as well as banking, insurance, financing, leasing, investing, farming, certain extractive businesses, and hotels and restaurants. Certain redemptions by the corporation around the time of issuance can disqualify stock. The One Big Beautiful Bill Act, Public Law 119-21, signed on July 4, 2025, changed three of the economic terms for stock acquired after that date, as Table 1 sets out. It introduced a tiered holding period, so that stock held at least three years qualifies for a 50 percent exclusion, at least four years for 75 percent, and at least five years for 100 percent. It raised the per-issuer cap on excludable gain from $10 million to $15 million, indexed for inflation beginning in 2027, with the alternative limit of ten times the taxpayer's aggregate adjusted basis unchanged. And it raised the aggregate gross assets ceiling from $50 million to $75 million, also indexed beginning in 2027. The excluded gain is not an alternative minimum tax preference. Table 1. Section 1202 terms before and after the One Big Beautiful Bill Act. Term Stock acquired on or before July 4, 2025 Stock acquired after July 4, 2025 Holding period for exclusion More than 5 years (100% for stock acquired after Sept. 27, 2010) 3 years: 50%; 4 years: 75%; 5 years: 100% Per-issuer gain cap Greater of $10 million or 10x basis Greater of $15 million or 10x basis; indexed after 2026 Aggregate gross assets ceiling $50 million $75 million; indexed after 2026 Non-excluded Section 1202 gain Not applicable after 5 years Taxed at 28% maximum rate, plus 3.8% NIIT where applicable Source: Internal Revenue Code section 1202 as amended by Pub. L. 119-21 (2025). The partial exclusions are less generous than their headline percentages suggest, because the portion of Section 1202 gain that is not excluded is taxed at the 28 percent maximum rate that applies to collectibles and similar gain, not the 20 percent long-term capital gain rate. A hypothetical makes the point. A founder who received stock in August 2025 sells it in November 2028, after more than three but less than four years, for a gain of $8 million, well within the $15 million cap. Half the gain, $4 million, is excluded. The other $4 million is taxed at up to 28 percent, plus the 3.8 percent net investment income tax if it applies, for a maximum of roughly $1.27 million. The effective federal rate on the whole gain is about 15.9 percent. Had she waited until after the fifth anniversary, the entire $8 million would have been excluded. Had the stock not been qualified small business stock at all, the same gain would have been taxed at up to 23.8 percent, about $1.9 million. The three-year tier is valuable mainly for founders whose companies are sold early, and it changes the arithmetic of whether to accept an early acquisition offer. Several planning points follow for a company at formation. The gross assets test is measured when the stock is issued, so founder stock issued at formation, before any money is raised, qualifies easily, while stock issued after a large financing may not. Because the $75 million ceiling is measured by the adjusted basis of the corporation's assets, including cash, a company that raises a large round can cross it, and stock issued afterward will not qualify even if earlier stock does. The per-issuer cap applies per taxpayer, and Section 1202(h) preserves qualified status when stock is transferred by gift or at death, so founders sometimes make gifts to non-grantor trusts for family members to multiply the number of caps, a technique whose details are beyond this book and which requires careful tax advice. Section 1045 allows a taxpayer who has held qualified small business stock for more than six months to defer gain by rolling the proceeds into other qualified small business stock within sixty days. Two uncertainties deserve mention. First, it remains unsettled when the holding period begins for stock received on conversion of a SAFE or a convertible note, a question that affects early investors and is discussed in Chapter 3. Second, state conformity varies. Several states follow the federal exclusion, but California, for example, does not, so a California resident who sells qualified small business stock owes full state tax on the gain. Finally, the founders should understand that the exclusion is theirs to lose through corporate actions they may not control. A later redemption of stock, a change in business that pushes the company into an excluded field, or a failure to keep 80 percent of assets in active use can all affect qualification. The company's counsel should keep a file documenting gross assets at each issuance and the nature of the business, because an acquirer's diligence team and the founders' own tax advisers will ask for it years later. Chapter 2: Raising Money Lawfully: The Law of Private Placements Every share of stock, every SAFE, and every convertible note that a startup issues is a security. Section 5 of the Securities Act of 1933 prohibits offering or selling a security unless a registration statement is in effect or an exemption applies. No startup registers its seed round, so every startup financing depends on an exemption, and the consequences of getting the exemption wrong are severe: investors may acquire a statutory right to rescind their purchase and get their money back, state regulators may take enforcement action, and the defect surfaces in diligence at every later financing and at the exit. This chapter covers the exemptions that venture financings actually use, the accredited investor definition on which they depend, the SEC staff's 2025 guidance on verification, and the recurring mistakes that turn a routine raise into a liability. Section 4(a)(2) and Regulation D Section 4(a)(2) of the Securities Act exempts "transactions by an issuer not involving any public offering." The statute does not define a public offering, and the Supreme Court's decision in SEC v. Ralston Purina Co. (1953) framed the question as whether the offerees need the protection of registration, which in turn depends on whether they are able to fend for themselves and have access to the kind of information registration would provide. Relying on the statute alone is possible but uncertain, so nearly all venture financings rely instead on Regulation D, and specifically on Rule 506, which the SEC adopted in 1982 as a safe harbor under Section 4(a)(2). Rule 506 now comes in two versions. Rule 506(b) is the traditional private placement. The issuer may sell to an unlimited number of accredited investors and up to thirty-five purchasers who are not accredited, provided each non-accredited purchaser, alone or with a purchaser representative, has enough knowledge and experience in financial and business matters to evaluate the investment. The issuer may not use general solicitation or general advertising to market the offering. If any non-accredited investor participates, the issuer must provide the disclosure specified in Rule 502(b), which includes financial statement information that can be expensive for a startup to prepare, and must make itself available to answer questions. In practice, venture financings under Rule 506(b) are sold only to accredited investors precisely to avoid those disclosure obligations. The prohibition on general solicitation is the constraint founders most often violate without realizing it. Announcing a raise in a public social media post, pitching at a demo day open to the public, or describing the terms of an open round in a press interview can each amount to general solicitation. The SEC staff has long taken the view that communications with persons with whom the issuer, or a registered broker-dealer or investment adviser acting for it, has a pre-existing substantive relationship are not general solicitation. In amendments adopted in November 2020 and effective in March 2021, the SEC added Rule 148, which provides that communications at certain "demo day" events sponsored by colleges and universities, governments, nonprofits, angel investor groups, incubators, and accelerators are not general solicitation, provided the sponsor does not make investment recommendations or charge certain fees, online participation is limited to members or invited persons, and the issuer's communication about the offering is limited to specified basic information such as the type and amount of securities and the intended use of proceeds. A lawyer's first question to a founder about to raise money should be what she has already said publicly. Rule 506(c), adopted in 2013 to implement the Jumpstart Our Business Startups Act of 2012 and effective on September 23 of that year, permits general solicitation. The price is that every purchaser must be an accredited investor, and the issuer must take reasonable steps to verify that status. A self-certification checkbox is not enough. The rule offers non-exclusive methods, such as reviewing IRS forms that report income for the two most recent years together with a written representation about the current year, reviewing bank and brokerage statements and a credit report for net worth, or obtaining a written confirmation from a registered broker-dealer, registered investment adviser, licensed attorney, or certified public accountant who has taken reasonable steps to verify. Otherwise the issuer must make a principles-based judgment considering the nature of the purchaser, the information the issuer has, the manner of the offering, and the terms of the offering, including any minimum investment amount. For a decade the verification burden kept many issuers out of Rule 506(c). Investors disliked handing over tax returns and bank statements, and issuers worried that anything short of the listed methods would not be enough. On March 12, 2025, the staff of the SEC's Division of Corporation Finance issued a no-action letter to Latham & Watkins that substantially eased the burden for offerings with high minimum investments, and the staff issued accompanying interpretive guidance to the same effect. Under the letter, an issuer may conclude that it has taken reasonable steps to verify accredited investor status where the minimum investment amount is at least $200,000 for a natural person or at least $1,000,000 for a legal entity that is accredited by virtue of its total assets, with parallel amounts for entities accredited because all of their equity owners are accredited. The issuer must obtain a written representation from the purchaser that it is an accredited investor and that its minimum investment is not financed in whole or in part by a third party for the specific purpose of making the investment, and the issuer must have no actual knowledge of facts indicating that the purchaser is not accredited or that the investment is financed. The reasoning is intuitive: a person who can write an unfinanced check of $200,000 into a single private company is very likely to meet the income or net worth test. The guidance matters for startups in a specific way. Many seed and Series A rounds have investors whose checks exceed those amounts, and a company that wants to announce its raise publicly, speak at open events, or use online platforms can now use Rule 506(c) without collecting financial documents from those investors. It does not help with smaller angel checks, which still require either Rule 506(b) and no general solicitation, or Rule 506(c) with conventional verification. A company cannot mix the two in a single offering; it must choose. The choice has become more strategic than it was, and counsel should settle it before any public communication about the round. Two other exemptions allow startups to raise money from the general public, and founders sometimes ask about them. Regulation Crowdfunding, under Section 4(a)(6) of the Securities Act, allows an issuer to raise up to $5 million in a twelve-month period through a registered funding portal or broker-dealer, from investors of any wealth subject to investment limits for non-accredited individuals, with disclosure on Form C and ongoing annual reports. Regulation A, sometimes called a mini-IPO, allows an offering of up to $75 million in a twelve-month period under its Tier 2, with an offering circular qualified by the SEC staff, audited financial statements, and ongoing reporting. Both exemptions have produced real financings, and crowdfunding in particular suits consumer companies with enthusiastic customers. But both are more expensive and more public than a Rule 506 round, and a company that brings in hundreds or thousands of small investors creates a capitalization table that later institutional investors may regard as a complication, which is why crowdfunding vehicles that aggregate small investors into a single holder of record have become common. For a company that expects to raise institutional venture capital, Rule 506 remains the default. Who is an accredited investor The accredited investor definition in Rule 501(a) determines who may participate in most venture financings. For natural persons, the principal tests have not changed in substance since the early years of Regulation D. A person is accredited if he or she has individual net worth, or joint net worth with a spouse or spousal equivalent, exceeding $1 million, excluding the value of the primary residence, a carve-out added by the Dodd-Frank Act in 2010. Alternatively, a person is accredited if he or she had individual income exceeding $200,000, or joint income with a spouse or spousal equivalent exceeding $300,000, in each of the two most recent years and has a reasonable expectation of reaching the same level in the current year. Because these dollar amounts have never been indexed for inflation, the population that meets them has grown steadily in real terms. In August 2020 the SEC amended the definition to add tests based on knowledge rather than wealth. Holders in good standing of the Series 7, Series 65, and Series 82 licenses are accredited. So are "knowledgeable employees" of a private fund investing in that fund. The amendments added the "spousal equivalent" concept, added limited liability companies and certain family offices to the list of eligible entities, and created a catch-all for entities owning investments in excess of $5 million that were not formed for the specific purpose of acquiring the securities offered. Directors, executive officers, and general partners of the issuer have long been accredited in the issuer's own offerings, which is why a startup's own officers can invest without further analysis. Congress has been considering broader changes. On December 2, 2025, the House of Representatives passed the INVEST Act, H.R. 3383, by a vote of 302 to 123. The bill would, among other things, direct the SEC to create an examination-based pathway to accredited status, allow inflation adjustments to the wealth thresholds, add criteria based on professional licensure, education, or experience, and direct the SEC to revise Regulation D so that presentations at events sponsored by universities, nonprofits, angel groups, and accelerators are not treated as general solicitation, building on the narrower relief already available under Rule 148. At the time of writing the bill had been referred to the Senate and had not been enacted. Practitioners should watch it, because enactment would change both who may invest and how founders may pitch. Filings, state law, and the bad actor rules An issuer relying on Rule 506 must file a notice on Form D with the SEC within fifteen days after the first sale of securities in the offering, and file amendments annually for continuing offerings and on certain changes. The date of the first sale is the date the first investor is irrevocably committed, which for a rolling SAFE raise may be much earlier than the founders assume. Failure to file Form D does not by itself destroy the exemption, but a court injunction for failure to file can disqualify the issuer from future Rule 506 offerings, and the omission is a diligence finding. Securities sold under Rule 506 are "covered securities" under Section 18 of the Securities Act, as amended by the National Securities Markets Improvement Act of 1996, which preempts state registration requirements. States may still require a notice filing, usually a copy of the Form D, and a fee, and each state in which an investor resides has its own deadline. Rolling closings with investors in many states create a filing calendar that someone must own. Rule 506(d), effective in September 2013, disqualifies an offering from Rule 506 if the issuer or certain covered persons have been subject to specified "bad actor" events, such as certain criminal convictions, court injunctions, and regulatory orders relating to securities. Covered persons include the issuer's directors and executive officers, beneficial owners of 20 percent or more of its voting equity, promoters, and compensated solicitors of investors, such as placement agents. The rule contains a reasonable care exception, which is why financing documents routinely include a representation that no covered person is subject to a disqualifying event and why issuers circulate questionnaires to directors and major holders before a closing. Two further problems recur. The first is integration: whether two apparently separate offerings will be treated as one, which can destroy an exemption if, for example, a Rule 506(c) raise with general solicitation is integrated with a Rule 506(b) raise that included a non-accredited investor. The SEC's integration framework in Rule 152, adopted in amendments effective in 2021, provides a general principle and several safe harbors, including one for offerings more than thirty calendar days apart, subject to conditions. The second problem is finders. A person who is paid a success fee for introducing investors may be acting as an unregistered broker-dealer in violation of Section 15(a) of the Securities Exchange Act of 1934. Section 29(b) of that Act makes contracts made in violation of the Act voidable, and investors introduced by an unregistered broker may claim rescission rights. The SEC proposed a conditional exemption for certain finders in October 2020 but never adopted it. The safest course is to pay finders nothing contingent on the amount raised, or to use a registered broker-dealer. Equity for employees, and what happens to the stock afterward Employee equity also requires an exemption. Rule 701 exempts offers and sales of securities by non-reporting companies under written compensatory benefit plans, up to the greatest of $1 million, 15 percent of total assets, or 15 percent of the outstanding class, in any twelve-month period. If sales in a twelve-month period exceed $10 million, a threshold raised from $5 million by the Economic Growth, Regulatory Relief, and Consumer Protection Act of 2018, the company must provide employees with additional disclosure, including financial statements and risk factors, a reasonable period before the sale. For an option, the sale occurs at grant and the disclosure must be delivered before exercise. Late-stage private companies with large option programs cross this threshold and must plan for the disclosure, which raises confidentiality concerns because it puts financial statements in the hands of every optionholder. The number of holders matters too. Under Section 12(g) of the Exchange Act, a company with total assets exceeding $10 million and a class of equity held of record by either 2,000 persons or 500 persons who are not accredited investors must register that class with the SEC and begin filing periodic reports. The JOBS Act of 2012 excluded from the count persons who received their securities under an employee compensation plan in an exempt transaction. That exclusion is why modern unicorns can remain private with thousands of employee holders, but it does not cover investors who buy in secondary transactions, a point that returns in Chapter 10. Finally, securities sold in a private placement are restricted securities. They cannot be freely resold. Rule 144 permits resales by non-affiliates of a non-reporting issuer after a one-year holding period, and Section 4(a)(7), added by the Fixing America's Surface Transportation Act of 2015, provides a resale exemption for sales to accredited investors subject to conditions. In practice, however, the company's transfer restrictions, rights of first refusal, and bylaws usually matter more than the securities laws in deciding whether a holder can sell, and those contractual limits are examined in Chapters 5 and 10. A short hypothetical shows how these rules combine. A founder raising a $1.5 million seed round on SAFEs has commitments from two venture funds for $500,000 each, from six angels for $50,000 each, and from her former manager, who is not accredited, for $25,000. If she has posted about the raise on a public professional network, she cannot use Rule 506(b), because she has engaged in general solicitation. She could use Rule 506(c), and the two funds qualify for streamlined verification under the 2025 guidance, but the six angels must be verified by conventional means and her former manager cannot invest at all. If she has not solicited publicly, she can use Rule 506(b), but including her former manager requires her to deliver Rule 502(b) disclosure to him and to be satisfied of his sophistication. Most counsel would advise her to decline the $25,000. In either case, she must file Form D within fifteen days of the first binding commitment, make state notice filings where each investor lives, and confirm that no director or 20 percent holder has a disqualifying history. Chapter 3: Before the Priced Round: SAFEs, Convertible Notes, and Stacked Dilution Most startups raise their first outside money without setting a price for their stock. They issue instruments that convert into stock later, at a price set by a future priced round, usually with a ceiling on the conversion price, a discount to it, or both. The appeal is speed and cost: a round can close in days on standard documents, without negotiating a certificate of incorporation, a board structure, or a set of investor rights. The cost is that the dilution is real but invisible until the priced round arrives, and founders who raise several tranches on different terms are often surprised by how much of the company they have already sold. This chapter explains the two dominant instruments, the simple agreement for future equity and the convertible note, and works through the arithmetic on a hypothetical company. The post-money SAFE Y Combinator introduced the simple agreement for future equity, or SAFE, in 2013 as an alternative to convertible notes. A SAFE is not debt. It carries no interest and has no maturity date. It is a contractual right to receive shares in a future equity financing, or cash or shares on a sale of the company, in exchange for money paid now. The original version measured the investor's conversion price against the company's capitalization before the new money came in, a "pre-money" approach that made each SAFE holder's eventual ownership depend on how much other SAFE money the company raised. In September 2018 Y Combinator released the "post-money" SAFE, which is now the standard form. Y Combinator publishes US versions with a valuation cap and no discount, with a discount and no cap, and with neither cap nor discount but a most favored nation provision, together with an optional pro rata side letter; it also publishes versions for companies organized in Canada, the Cayman Islands, and Singapore. The defining feature of the post-money SAFE is that the holder's ownership is fixed at signing, measured after all SAFE money is counted. A $1 million SAFE with a $10 million post-money valuation cap entitles the holder, if the cap applies, to 10 percent of the company's capitalization immediately before the priced round. The conversion price, called the Safe Price, equals the post-money valuation cap divided by the "Company Capitalization." That defined term is where the drafting matters. It includes all outstanding capital stock, all outstanding options and promised options, the unissued option pool, and all converting securities, including the SAFEs themselves. It excludes any increase to the option pool made in connection with the priced round, except to the extent needed to cover promised options that exceed the existing pool. The consequence is that SAFE holders are diluted, along with the founders, by both the new money in the priced round and any pool increase the new investors demand. Three other terms are essential. On a liquidity event, such as a sale of the company before the priced round, the holder receives the greater of its purchase amount or the amount it would receive if its SAFE converted at the cap. On a dissolution, the holder is entitled to its purchase amount, ahead of common stock and alongside other SAFE holders and similar instruments. And when the SAFE converts in a priced round, the holder receives a special series of preferred stock, often called shadow preferred, with the same rights as the new series except that its liquidation preference and conversion price are based on the Safe Price rather than the new-money price. That last feature prevents a problem that plagued early convertible instruments: a holder converting at a deep discount would otherwise receive a liquidation preference far larger than the money it actually invested. A worked example: two SAFEs and the Series A Consider a hypothetical company, which we will call Harbor Robotics, Inc. At formation its two founders hold 5,000,000 and 3,000,000 shares of common stock, and the company has reserved an option pool of 1,000,000 shares, none yet granted. Its fully diluted capitalization is 9,000,000 shares. The company raises $1,000,000 from a seed fund on a post-money SAFE with a $10 million valuation cap. Six months later, with traction, it raises another $1,000,000 from a second fund on a post-money SAFE with a $16 million cap. The first SAFE holder has bought 10 percent of the pre-Series A capitalization ($1 million divided by $10 million). The second has bought 6.25 percent ($1 million divided by $16 million). Together they own 16.25 percent, and every other holder, which here means the founders and the pool, bears that dilution in proportion. Neither SAFE holder is diluted by the other. That is the point of the post-money form, and it is also the point founders miss: each additional SAFE comes entirely out of the existing holders, not out of earlier SAFE holders. To convert the SAFEs, the company must solve for the Company Capitalization, because the SAFE shares are themselves part of it. If the existing holders' 9,000,000 shares make up 83.75 percent of the total, the total is 9,000,000 divided by 0.8375, or 10,746,269 shares. The first SAFE converts into 10 percent of that, 1,074,627 shares, at a Safe Price of about $0.9306 per share ($10 million divided by 10,746,269). The second converts into 671,642 shares at about $1.4889. Before the Series A, the larger founder has gone from 55.6 percent of the company at formation to 46.5 percent. Now the Series A arrives: $8 million of new money at a $32 million pre-money valuation, for a $40 million post-money valuation, with a condition that the unissued option pool equal 10 percent of the post-closing fully diluted capitalization. Chapter 4 walks through that round, including the pool increase, and shows the full capitalization table. For present purposes the essential number is the Series A price: about $2.8729 per share. The first SAFE holder paid $0.9306 per share for the same economic position, about 32 percent of the Series A price. The second paid about 52 percent of it. Each will receive shadow preferred with a liquidation preference equal to its $1 million investment, not the $3.09 million or $1.93 million its shares would carry if priced at the Series A price. A SAFE with a discount instead of a cap works more simply. A $1 million SAFE with a 20 percent discount would convert at 80 percent of the Series A price, about $2.2983 per share, into roughly 435,100 shares. At a high Series A valuation, a discount-only SAFE is much less valuable to the investor than a capped one, which is why most priced seed investors insist on a cap. The cap is a ceiling, not a fixed price. If the Series A were priced below the Safe Price, the SAFE would convert at the Series A price itself, into more shares than the cap alone would give. Suppose Harbor Robotics struggled and its Series A was priced at $1.20 per share. The second SAFE, whose Safe Price was about $1.4889, would convert at $1.20 into about 833,333 shares instead of 671,642, and its holder's percentage would rise accordingly. The first SAFE, with its lower Safe Price of about $0.9306, would still convert at that price. In a weak priced round, then, the SAFEs convert into more of the company than the founders modeled at the caps, and the new investor's price and the SAFE conversion interact: the lower the Series A price, the more shares the SAFEs receive, which in turn affects the price the new investor will offer for its target percentage. Founders raising a priced round at a valuation near or below their SAFE caps should expect the new lead to model this precisely and to price the round with the SAFE conversion shares counted in the pre-money capitalization, which places their dilution on the founders. Now consider the same company sold before any Series A, for $30 million in cash. The first SAFE holder is entitled to the greater of its $1 million or the amount it would receive on conversion at its cap. Conversion gives it 10 percent of the capitalization, so it receives roughly $3 million, and the second SAFE holder roughly $1.875 million, before any transaction expenses or management carve-outs. If the company were instead sold for $8 million, the SAFE holders' as-converted amounts would be $800,000 and $500,000, so each would take its $1 million purchase amount instead. The SAFE's downside protection on a sale is therefore a return of capital, ranking ahead of common, which is a modest but real form of liquidation preference. Convertible notes Before SAFEs, convertible promissory notes were the standard seed instrument, and they remain common, particularly with investors who want the formal protections of debt. A convertible note is a loan that accrues interest, commonly at a single-digit annual rate, and has a maturity date, commonly eighteen to twenty-four months. It converts automatically into the securities issued in a "qualified financing" above a stated size, at a price equal to the lower of a discount to the new price or the price implied by a valuation cap. Accrued interest usually converts along with principal. On a sale of the company before conversion, the note typically entitles the holder to repayment of principal and interest, often with a premium, or conversion into common stock at the cap, whichever is greater. The differences from a SAFE are not cosmetic. First, a note is debt on the company's balance sheet, which can matter when the company is insolvent or near it and directors must consider creditors' interests. Second, the maturity date gives the holders leverage: if no qualified financing occurs before maturity, the holders can demand repayment, which the company usually cannot make, and the resulting negotiation over extension, conversion at the cap, or a new instrument shifts value to the noteholders. Third, notes are typically drafted with a pre-money cap, so the ownership each noteholder acquires depends on the total amount of notes and other converting instruments outstanding. Fourth, interest converts too: a $1 million note at 6 percent simple interest that converts after two years converts $1.12 million. None of these features is bad for founders in itself, but each must be modeled. A hypothetical shows how those features compound. Suppose that instead of its two SAFEs, Harbor Robotics had issued a single $2 million convertible note bearing 6 percent simple interest, with a 20 percent discount and an $8 million pre-money valuation cap, and that the Series A closes two years later. Accrued interest is $240,000, so $2.24 million converts. The cap price is $8 million divided by the pre-money capitalization, which under a typical note means the 9,000,000 shares outstanding and reserved before the conversion, or about $0.8889 per share. The discount price is 80 percent of the Series A price. At the Series A price of roughly $2.87, the discount price is about $2.30, so the cap governs, and the note converts into about 2,520,000 shares. The noteholder's position before the new money is about 21.9 percent (2,520,000 of 11,520,000 shares), compared with the 16.25 percent the two SAFEs bought for the same $2 million. The difference comes from three sources: the interest, the lower cap, and the pre-money measurement, which gives the note holder its percentage before its own shares are counted. Each of those terms is negotiable, but a founder comparing a note offer with a SAFE offer must put them on the same footing, and the only way to do that is to model conversion at a realistic Series A price. The same example shows why discounts and caps interact. If the Series A price had instead been $1.00 per share, the discount price of $0.80 would have been lower than the cap price of $0.8889, and the note would have converted at the discount price. Investors hold both terms precisely so that they are protected at both high and low Series A valuations. Founders sometimes accept a discount as a small concession without noticing that, in a weak Series A, it can matter more than the cap. Many practitioners now prefer SAFEs for companies that expect a priced round within a reasonable time, and notes for bridge financings between priced rounds, where existing investors want a debt claim and a maturity date to discipline the company. A bridge note in a struggling company is often the first document in a sequence that ends in a down round, and Chapter 7 returns to that sequence. Practical points and unsettled questions Several practical lessons follow from the arithmetic. First, every SAFE and note should appear on the company's capitalization table in a fully converted form from the day it is signed, using the cap as the conversion price, so that the founders see the dilution they have already accepted. Many founders track their ownership on a share count that includes the option pool but ignores converting instruments; the Harbor Robotics founders who believe they own 88.9 percent of the company before the Series A in fact own 74.4 percent on a SAFE-converted basis. Second, the most favored nation provision in an uncapped SAFE, and side letters granting it to capped holders, can reprice earlier instruments automatically when the company later issues on better terms. A founder who gives an early friend an uncapped MFN SAFE and later issues a SAFE with a $6 million cap has, in effect, given the friend that cap too. Third, pro rata side letters give SAFE holders the right to buy a share of the priced round. When several SAFE holders hold pro rata rights, the lead investor in the Series A may find that its desired allocation is not available, and the negotiation over waivers is a routine part of closing. The post-money SAFE's side letter is drafted so that pro rata rights apply in the round in which the SAFE converts, and counsel should confirm how the rights are measured. Fourth, the SAFE confers no voting rights, no information rights, and no board seat. That is part of its appeal to founders. But it also means SAFE holders have no contractual say in what happens before conversion, including a sale at a price that returns only their purchase amount. Two tax questions remain unsettled. The first is how a SAFE is characterized for federal income tax purposes before conversion. Commentators have argued for treating it as equity, as a prepaid forward contract, or in some cases as debt, and there is no specific IRS guidance. The second, which follows from the first, is when the Section 1202 holding period begins for stock received on conversion. If the SAFE is equity for tax purposes, the holding period arguably begins when the SAFE is purchased; if it is a forward contract, arguably only on conversion. For a convertible note, the holding period is generally understood to begin at conversion, because debt is not stock. Under the tiered holding periods introduced in 2025, a year of difference in the start date can move the exclusion from 50 percent to 75 percent or from 75 percent to 100 percent, so early investors and founders should take advice on the question, and companies should keep careful records of SAFE issuance dates and gross assets at issuance. Finally, both instruments are securities. Every SAFE and note round requires an exemption under Regulation D or otherwise, a Form D filing within fifteen days of the first sale, and state notice filings, as Chapter 2 explained. The ease of signing a SAFE online makes it tempting to treat these obligations as optional. They are not. Hashtags: #VentureCapitalAndStartupLaw #VentureFinancing #StartupLaw #TermSheets #EquityFinancing #DelawareCorporations #FounderStock #Section83bElection #QualifiedSmallBusinessStock #PrivatePlacements #RegulationD #Rule506 #AccreditedInvestors #PostMoneySAFE #ConvertibleNotes #CapitalizationTables #OptionPools #BoardControl #LiquidationPreferences #AntiDilutionProtection #DownRounds #FiduciaryDuties #DragAlongRights #ExitStrategies #FutureOfVentureCapitalLaw

Download the Book (PDF): Introduction A startup that loses money every month, owns little that a pawnbroker would accept, and cannot say with confidence what it will be worth in two years should not, by the ordinary logic of credit, be able to borrow. Banks lend against cash flow or against collateral, and the typical venture-backed company has neither. Yet lending to such companies has become one of the fastest-growing corners of finance. PitchBook and the lender Runway Growth Capital counted a record $68.8 billion of venture debt in the United States in 2025, spread across roughly a thousand transactions. A careful academic study by Jesse Davis, Adair Morse and Xinxin Wang found that early-stage companies use venture debt in about one financing round in three. What was once a specialist product sold by a handful of California banks is now part of the standard toolkit of any company that has raised institutional equity. The attraction is easy to state. Every dollar of equity a startup raises is paid for with a piece of the company, and the founders, employees and early investors who hold the rest are diluted accordingly. A loan, by contrast, is paid for in interest, fees and, usually, a small bundle of warrants. If the company is going to be worth a great deal more in eighteen months than it is today, then borrowing now and selling equity later, at the higher price, leaves the existing owners with a larger share of a larger business. That is the pitch every venture lender makes, and in the right circumstances it is true. The difficulty lies in the phrase "in the right circumstances". A loan has to be repaid whether or not the company hits its targets. It comes with covenants that can hand control to the lender at exactly the moment the company is weakest. Its headline interest rate understates its true cost, sometimes by a wide margin, once upfront fees, end-of-term payments and the value of the warrants are included. And the thing that makes lending to loss-making companies possible at all, the expectation that venture investors will keep funding the borrower, is the very thing that disappears in a downturn. When Silicon Valley Bank failed in March 2023, the largest venture lender in the world vanished in a weekend, and thousands of borrowers discovered how much of their financing depended on a single relationship. The argument of this book This book argues one thing. Venture debt lowers a young company's cost of capital when it is used to buy time between two equity events whose arrival is already highly likely, and it becomes the most expensive money the company holds when it is used as a substitute for equity the company cannot raise. Everything about the instrument follows from this. Lenders price and structure their loans on the assumption that the next equity round, or a sale of the company, will repay them. The warrants they take are a small claim on the upside they are helping to create. The covenants they write are tripwires designed to get them out before the equity investors lose interest. A borrower who understands that the loan is, in substance, an advance against a future financing event can use it well. A borrower who treats it as free money, or as a lifeline when equity has dried up, is borrowing against an event that may not come, on terms written by someone whose first duty is to be repaid. Put another way, venture debt is a bet on timing more than a bet on the business. The question a founder or board should ask is less "can we afford the interest?" than "how confident are we that, by the time this loan starts to bite, we will be able to raise equity at a higher price or sell the company?" If the answer is "very", the loan is cheap. If the answer is "we hope so", the loan is expensive in ways the term sheet does not show. What the book covers The chapters build the argument in order. The first explains why anyone lends to companies that lose money, and what the lender is really underwriting when it does. The second takes a venture loan apart, clause by clause, and works through what a typical loan actually costs once every fee is counted. The third deals with warrants: how coverage is calculated, what warrants are worth, and how much dilution they really cause. The fourth turns to covenants and the control they confer, including the material adverse change clause that many founders sign without reading closely, and the lessons of Silicon Valley Bank's failure. The fifth chapter is the analytical centre of the book. It compares, with worked numbers, a company that bridges to its next round with debt against one that bridges with equity, and shows both where the debt route wins and the specific conditions under which its advantage evaporates. The sixth moves up the size scale to growth capital and pre-IPO financing: the convertible notes, leveraged loans and structured facilities that late-stage companies such as Spotify, Uber and Airbnb have used, and the enormous GPU-backed loans of the artificial intelligence boom. The seventh surveys the lenders themselves, from banks to publicly traded business development companies to Europe's public development bank, because who lends to a company shapes how the loan behaves in a crisis. The eighth sets out a practical discipline for using venture debt: when to borrow, how much, from whom, and what to do when things go wrong. The conclusion draws out what follows for founders, boards and investors, and what remains unsettled. A note on numbers and scope Venture debt is a private market, and much of what happens in it is never disclosed. The aggregate figures published by data providers depend heavily on what they count. PitchBook's totals, for example, now include very large facilities secured on data-centre hardware for artificial intelligence companies, which sit a long way from the $3 million to $10 million term loans that make up most transactions. Where this book cites market figures it names the source and, where appropriate, flags what the figure includes. The worked examples are illustrative: the companies in them are invented so that every number can be followed, but the terms they use sit within the ranges that lenders and borrowers commonly report, and every calculation has been checked. The book concentrates on debt provided to companies that are already backed by professional equity investors, which is what the industry means by venture debt, and on the larger growth and pre-IPO financings that are its natural extension. It touches on revenue-based financing and other cash-flow lending where they help to mark the boundaries, but it does not attempt to cover small-business lending in general, nor the leveraged buyout market. The focus is the United States, where the market is largest and best documented, with attention to Europe where its institutions differ in instructive ways. The reader this book has in mind is a founder or finance lead weighing a term sheet, a board member asked to approve one, an early employee trying to understand what the company's debt means for their options, or an investor who wants to understand how lenders think. None of them needs a background in credit analysis. All of them need to understand that the cheapest-looking capital on offer is only cheap under specific conditions, and that knowing whether those conditions hold is the whole of the skill. Chapter 1: Why Anyone Lends to a Company That Loses Money Conventional credit analysis rests on two questions. Can the borrower pay interest and principal out of the cash its business generates? And if it cannot, is there something the lender can seize and sell to recover the money? A mortgage lender looks at the borrower's salary and then at the house. A bank lending to a manufacturer looks at operating cash flow, then at receivables, inventory and machinery. The ratios differ from industry to industry, but the logic is the same everywhere: repayment comes from earnings, and collateral covers the shortfall when earnings disappoint. A venture-backed startup fails both tests. Its operating cash flow is negative, often deeply so, and deliberately: the whole point of raising venture capital is to spend ahead of revenue in pursuit of a large market. Its tangible assets are usually trivial. A software company's balance sheet is mostly cash, laptops and a lease. A biotechnology company may own laboratory equipment, but equipment designed for a narrow scientific purpose fetches little at auction. On the standard criteria, a lender should decline, or should demand a rate so high that no sane borrower would accept it. Yet lenders do lend, in large volumes and at rates that, while well above what a profitable public company pays, are far below what the raw risk of a single startup would seem to require. The explanation is that venture lenders are not underwriting the startup in the way a bank underwrites a manufacturer. They are underwriting something else, and understanding what that something is explains almost everything about how venture loans are priced, structured and enforced. The real source of repayment The law professor Darian Ibrahim put the puzzle crisply in a 2010 article, "Debt as Venture Capital". Startups, he noted, lack every traditional marker of creditworthiness: operating history, positive cash flow, hard collateral and, in most cases, personal guarantees from founders. Venture lenders nonetheless extend billions of dollars a year to them. His answer was that two things substitute for the missing criteria: the company's backing by established venture capital firms, and its intellectual property. The first substitute is the more important. When a venture lender makes a loan to a company that has just raised, say, a $15 million Series A from a well-known fund, the lender's primary expectation is that it will be repaid from the proceeds of the company's next equity round, or from the proceeds of a sale of the company, and not from the company's own earnings. The venture investors' presence signals two things. It signals quality, because a professional investor has examined the business and committed real money to it. And it signals continued support, because venture funds typically reserve capital to invest again in their portfolio companies, and a fund that has put $15 million into a company has strong reasons not to let it fail for want of the next few million. This is why venture lenders care so much about who the equity investors are. A company backed by a fund with a long record of supporting its companies through difficult periods is a better credit than an otherwise identical company backed by a first-time fund or by a scattering of angels, because the probability that someone will write the next cheque is higher. Lenders keep detailed histories of how different funds behave when a portfolio company struggles. Some funds are known for bridging their companies through bad quarters; others are known for walking away. The lender's view of the equity sponsor is, in practice, a large part of its view of the credit. The economists Yael Hochberg, Carlos Serrano and Rosemarie Ziedonis tested this idea in a study published in the Journal of Financial Economics in 2018. Examining startups in three innovation-intensive sectors, they documented widespread lending to companies at early stages of development, which they described as a "surprisingly active but opaque" market. They then used the collapse in venture fundraising around 2000 to ask what happens when the venture investors' ability to support their companies weakens. Lending fell. When venture funds were short of capital, and so less able to credibly promise follow-on investment, lenders became less willing to lend to the companies those funds backed. The lender's collateral, in effect, was the venture fund's commitment, and when the commitment became less credible, the collateral lost value. Intellectual property and enterprise value The second substitute, intellectual property, matters most when things go wrong. A venture lender typically takes a security interest over all of a company's assets, including its patents, software and other intellectual property; in some structures the company instead agrees not to pledge its intellectual property to anyone else, a so-called negative pledge, which preserves the lender's position without a formal lien. If the company fails, the lender has a claim on whatever can be sold. Hochberg, Serrano and Ziedonis found that this matters in a measurable way. Using changes in the liquidity of the market for patents, they showed that startups whose patents were more readily tradeable, because there was an active market in which such patents could be sold to other companies, were more likely to obtain loans. Patents that could be redeployed to another owner were better collateral than patents that only made sense inside the company that filed them. The finding confirms what lenders say in practice: the question is not only whether the company owns intellectual property, but whether anyone else would pay for it. In most failures, though, the most valuable thing a lender can recover is neither a patent nor a server, but the business as a going concern. A struggling startup with a real product, real customers and a capable team can often be sold, even if at a price far below what its investors paid. Because the lender sits ahead of every equity holder in the capital structure, it is repaid first out of any sale. A company that has raised $40 million of equity and borrowed $8 million has a large cushion beneath the lender: the company must lose more than four-fifths of the value its equity investors paid for before the lender's principal is at risk, assuming a buyer can be found. Lenders often describe their exposure in exactly these terms, as a "loan-to-value" ratio in which the value is the company's enterprise value as implied by its most recent equity round. A loan equal to a small fraction of that value is, in their eyes, well covered. That reasoning has an obvious weakness. The enterprise value implied by a venture round is a price paid by optimistic investors for a minority stake with preferential rights, and it can fall sharply and suddenly. But a low loan-to-value ratio combined with a strong equity sponsor and some saleable intellectual property gives the lender three separate routes to repayment: the next round, a sale of the company, or a liquidation of its assets. Each is uncertain; together they are enough to support a business. How the market was built Venture lending grew out of equipment leasing. In the late 1970s and early 1980s, specialist lessors in California began financing the computers, laboratory equipment and fabrication tools that young technology companies needed, secured on the equipment itself. Because the equipment had limited resale value, lessors began asking for warrants to compensate them for the risk, and over time they discovered that the warrants and the quality of the venture backing mattered more to their returns than the equipment did. The collateral became an afterthought, and the loan became, in substance, a bet on the company's ability to raise more money. Silicon Valley Bank, founded in 1983, turned this into a banking business. Its model combined lending with deposits: it offered loans to venture-backed companies, and in return required them to keep their cash, often including the proceeds of their equity rounds, at the bank. The deposits gave SVB cheap funding and a real-time view of each borrower's cash position, which was itself a form of credit protection, since a lender that can see a borrower's balance fall week by week has plenty of warning. The model worked for four decades. By the time of its failure in 2023, PitchBook estimated that SVB controlled somewhere between 60 and 70 percent of the venture debt market and held deposits from nearly half of US venture-backed technology and life-science companies. Alongside the banks grew a second tier of non-bank lenders: specialist funds and, from the mid-2000s, publicly traded business development companies such as Hercules Capital, a structure explained in Chapter 7. These lenders took more risk than the banks, lent larger amounts to later-stage companies, required fewer deposit relationships, and charged more. The market that results is layered: banks for smaller, cheaper, more tightly controlled loans to companies they also bank; non-bank lenders for larger, costlier, more flexible facilities; and, at the top end, private credit funds and investment banks arranging nine- and ten-figure financings for companies approaching an initial public offering. Why the equity investors want it A lender's willingness is only half the explanation. The other half is that the venture investors themselves generally welcome their companies borrowing, and often introduce them to lenders. At first sight this is odd, since a loan sits ahead of the venture investors' preferred shares and reduces what they recover if the company fails. The explanation lies in the arithmetic of venture returns. A venture fund makes its money from a small number of very large outcomes. For those outcomes, the fund's percentage ownership at exit matters enormously, and every round of equity the company raises between now and then reduces it. A modest loan that lets the company reach its next milestone before raising equity, and so raise at a higher price, preserves the fund's ownership in exactly the cases that determine its returns. The cost falls mainly in the cases where the company fails, in which the fund's stake was going to be worth little anyway. Ibrahim argued that venture debt, once a company has an institutional equity sponsor, is consistent with standard corporate finance theory for precisely this reason: it improves the payoff to the equity holders in the scenarios that matter to them. Tereza Tykvová, in a 2017 paper in the Journal of Financial and Quantitative Analysis, modelled the choice a venture investor faces when its company needs later-stage financing: bring in another venture fund, or bring in a venture lender. Her model and evidence suggest that investors who hold high-quality companies tend to turn to lenders, partly because doing so signals their confidence, while they prefer new equity partners when the company's upside is very large, when uncertainty is high or when a later investor can add real value. The practical translation is useful. Venture debt tends to be used by companies whose insiders believe the next round will come easily, and it is least suitable where the future is genuinely uncertain. This alignment of interests between equity investors and lenders is the foundation of the market, and it is also its point of fragility. The lender relies on the venture investors to keep the company funded. The venture investors rely on the lender to provide cheap, non-dilutive capital. As long as both believe the company is heading for a successful next round, the arrangement works for everyone, including the founders. When that belief weakens, their interests diverge sharply: the venture investors may prefer to stop funding a struggling company, and the lender may prefer to be repaid immediately out of whatever cash remains. Much of the legal machinery of a venture loan, which later chapters examine, exists to manage that moment of divergence. The lender's arithmetic The economics of a venture lending book make sense of the prices borrowers see. Consider a lender with fifty loans of $5 million each, a $250 million portfolio. Suppose that, once the upfront fees, final payments and prepayment charges described in the next chapter are included, its loans earn an effective return of about 14 percent a year on the money outstanding, a figure consistent with the terms typically offered by non-bank lenders. If losses run at the 4.3 percent a year that González-Uribe and Mann found on average in their sample of business development company loans, the lender's return after losses is about 9.7 percent, before its own funding and operating costs. That is a respectable return for a senior lender, but it leaves little room for error. A few more failures than expected, or a year in which the venture market turns and write-offs double, and the margin is gone. This is why the warrants matter to lenders more than their small size suggests, and why lenders are so selective about equity sponsors. A lender cannot charge 25 percent interest to cover a higher loss rate, because at that price its borrowers would burn through their cash faster and fail more often, and the best borrowers would go elsewhere. The only way to keep losses near the level its pricing can absorb is to lend to companies whose investors are very likely to keep funding them. The warrants then provide the margin of safety: one borrower in fifty that becomes a large public company can produce a warrant gain worth several million dollars, enough to offset a meaningful slice of a year's losses. The same arithmetic explains why lenders behave as they do when a borrower struggles. A lender earning 14 percent on performing loans and losing most of the principal on failed ones has an overwhelming interest in turning a potential failure into a partial recovery. Recovering $3 million of a $5 million loan through an early sale of the company, instead of nothing through a liquidation six months later, is worth more to the lender than a year's interest on the entire loan. Every provision that lets the lender act early, from monthly reporting to the material adverse change clause, is designed to capture that difference. What the lender is really buying It helps to state the lender's position plainly. A venture lender provides a sum of money, typically equal to a quarter to a half of the company's most recent equity round, on the expectation that within two to four years the company will either raise more equity or be sold. It charges an interest rate well above bank lending rates, collects fees at the start and end, and takes warrants that pay off if the company becomes very valuable. It protects itself with a security interest over the company's assets, with covenants that let it act early if the company deteriorates, and, above all, by choosing borrowers whose equity investors are likely to keep supporting them. Its returns come mostly from interest and fees on the many loans that are repaid on time or early. Its losses come from the minority of companies that fail before they can raise again. The warrants are a sweetener that, across a portfolio, can offset a meaningful share of those losses when one or two borrowers become very large. The economists Juanita González-Uribe and William Mann, examining the loan books of five US business development companies over 2005 to 2022, found an average annual write-off rate of about 4.3 percent, with the median loan paid off roughly two years before its scheduled maturity, typically around a new equity round or an acquisition. That pattern is the signature of a lender who is financing the gap between events rather than the business itself. For the borrower, the implication is direct. A venture lender is prepared to lend to a loss-making company because it is effectively lending against that company's next equity round. The loan is cheap relative to equity precisely because the lender expects that round to happen. If the borrower shares that expectation and is right, the loan is a good deal. If the borrower is using the loan because it doubts the next round will come, then it is borrowing on terms set by someone who expected the opposite, and the terms will reflect that the moment the doubt becomes visible. Chapter 2: The Anatomy of a Venture Loan Term sheets for venture loans are short, usually two to four pages, and most of them look alike. That similarity is deceptive. A handful of numbers on the first page determine the loan's real cost, and a handful of phrases further down determine who controls the company if things go badly. Founders who have negotiated equity rounds often assume the debt term sheet is the simpler document. In one sense it is: there are fewer moving parts than in a preferred-stock financing. In another sense it is harder, because the economic effect of each term depends on when the loan is repaid, and the borrower does not know that when it signs. This chapter takes the standard growth-stage term loan apart and puts a price on it. The warrants get a chapter of their own, as do the covenants; here the focus is on the cash terms and on how they combine. The main terms A venture loan begins with a commitment: the maximum amount the lender agrees to provide. The commitment is frequently split into tranches. A company might have a $10 million facility, of which $5 million is available at closing and the remaining $5 million becomes available only if the company reaches a milestone, such as a revenue level or the closing of a further equity round, within a set period. Tranching protects the lender, which only has to advance the full sum if the company is performing, and it can look attractive to the borrower, since it pays interest only on what it draws. But a tranche the borrower cannot draw is not capital it can plan around. A company that budgets on the assumption of the full $10 million and then misses the milestone that unlocks the second half discovers that half of its debt financing was contingent on the very success it was meant to fund. The draw period is the window during which the borrower can take down the committed funds, commonly six to eighteen months. Some lenders charge an unused-line fee on undrawn amounts; many require a minimum draw at closing. The interest rate on venture loans in the United States is usually floating, set as a margin above the prime rate, with a floor that stops it falling below a stated level if rates are cut. Bank lenders, which fund themselves with deposits, have historically charged the lowest margins, in some cases only a point or so above prime. Non-bank lenders, which fund themselves with more expensive capital, charge more; a practitioner guide published by the accounting firm Kruze Consulting puts typical non-bank rates in the region of 11 to 12 percent, and the pricing of larger and riskier loans runs higher still. The floor matters more than it appears: when rates were near zero in 2020 and 2021, floors set at several percentage points above prime were binding, and borrowers paid the floor rather than the headline formula. The interest-only period is the stretch at the start of the loan during which the borrower pays interest but no principal. Twelve months is common for a growth-stage loan, sometimes extendable to eighteen or more if the company hits a milestone. After the interest-only period comes amortization, during which the principal is repaid in equal monthly instalments, commonly over twenty-four to thirty-six months. The combination determines how much of the loan the company actually keeps in its bank account over time, which, as Chapter 5 shows, is the number that matters for runway. Next come the fees. There is usually a commitment or facility fee at closing, often between half a percent and one percent of the commitment. There is almost always a final payment, sometimes called an end-of-term charge, of a few percent of the amount drawn, due when the loan is repaid, whether at maturity or earlier. And there is usually a prepayment fee, payable if the borrower repays before maturity, which typically steps down over time: something like 3 percent in the first year, 2 percent in the second and 1 percent thereafter. Finally, there are the warrants, which give the lender the right to buy a small number of the company's shares at a fixed price, and the covenants and events of default, which govern what the borrower must and must not do. Both are discussed at length later. Table 1 gathers these terms, with the ranges that practitioners commonly report for a US growth-stage loan. They are indicative: terms vary with the lender, the borrower's stage and the state of the market. Table 1. Principal economic terms of a US growth-stage venture term loan (indicative ranges reported by practitioners). Term What it is Common range Main effect Commitment Maximum amount available, often in tranches 20-50% of last equity round Sets how much runway can be bought Interest rate Floating margin over prime, with a floor Near prime for banks; roughly 10-13% for non-banks Monthly cash cost Interest-only period Months with no principal repayment 6-18 months How long the full loan stays in the bank Amortization Equal monthly principal repayments 24-36 months How fast the loan drains cash Facility and final fees Charged at closing and at repayment 0.5-1% upfront; 3-8% at end Raises effective cost, most sharply if repaid early Prepayment fee Charged for early repayment Steps down from about 3% to 1% Penalises refinancing Pricing a loan properly The only honest way to compare venture loans, or to compare a loan with equity, is to write down every cash flow between the borrower and the lender and calculate the rate of return those flows imply. The headline interest rate does not do this, because it ignores the fees and ignores the timing of repayment. Consider a representative loan. A company borrows $10 million at a fixed rate of 10.5 percent a year. It pays a 1 percent facility fee at closing, so it receives $9.9 million. The loan is interest-only for twelve months, during which the company pays $87,500 a month in interest. It then amortizes over thirty-six months, with principal payments of about $277,800 a month plus interest on the declining balance. At maturity, forty-eight months after closing, the company pays a final fee of 5 percent of the amount drawn, $500,000. Over the full four years the company pays about $2.67 million in interest, plus the $100,000 upfront fee and the $500,000 final payment: roughly $3.27 million in total cost on $10 million borrowed. Measured properly, as the internal rate of return on the lender's cash flows, the loan costs the borrower a little over 1 percent a month, which compounds to about 13.1 percent a year. The fees have added more than two and a half percentage points to the headline rate. That is the cost only if the loan runs to maturity, and venture loans rarely do. Suppose, more realistically, that the company raises its next equity round two years after closing and repays the loan then. It has paid twelve months of interest-only payments and twelve months of amortization, a little under $1.94 million in interest in total. At repayment it owes the remaining principal of about $6.67 million, a 2 percent prepayment fee on that balance of about $133,000, and the full $500,000 final payment, since the final payment is typically due on any repayment. The internal rate of return now rises to about 15.0 percent a year. If the company raises sooner and repays after only twelve months, the cost rises further. It has paid $1.05 million of interest, and at repayment owes the full $10 million, a 3 percent prepayment fee of $300,000 and the $500,000 final payment. The effective annual cost is now about 20.7 percent. The pattern is the essential point. Fixed fees are spread over the life of the loan, so the shorter the life, the larger their weight in the annual cost. A loan that looks like 10.5 percent money on the term sheet costs roughly 13 percent if held for four years, 15 percent if repaid after two, and more than 20 percent if repaid after one. And these figures ignore the warrants, which, as the next chapter shows, add several percentage points more. Since the whole purpose of a growth loan is usually to bridge to an equity round that will arrive within twelve to twenty-four months, the realistic cost of the loan is the high end of this range, not the low end. None of this makes the loan a bad deal. Twenty percent a year is still far cheaper than equity in a company that expects to multiply in value, as Chapter 5 demonstrates. But a founder who compares a 10.5 percent headline rate with the cost of equity is comparing the wrong numbers, and a board that approves a loan without seeing the effective cost under realistic repayment scenarios has not really priced it. Two offers compared The value of this method shows most clearly when a company has two offers that differ on almost every term. Suppose the same company, fresh from a $15 million Series A, burning $1.25 million a month and with twelve months of runway, receives two term sheets. The first, from a bank, offers $5 million at 7.75 percent, one point over a prime rate taken here as 6.75 percent, its level after the Federal Reserve's December 2025 cut, with a 0.5 percent facility fee, six months of interest-only payments followed by thirty months of amortization, a 3 percent final payment and warrants over 25 basis points of the company's shares. It requires the company to keep its operating accounts at the bank and to meet a minimum-revenue covenant set against its budget. The second, from a non-bank lender, offers $10 million at 11.5 percent, with a 1 percent fee, twelve months of interest-only payments followed by thirty months of amortization, a 6 percent final payment and warrants with 8 percent coverage at the Series A price. It has no financial covenants and no deposit requirement. On headline rate the bank wins easily. If both loans are repaid after eighteen months, when the company expects to raise its Series B, the bank loan's effective annual cost, counting all fees, is about 11.7 percent; adding the Black-Scholes value of its small warrant, about 13.6 percent. The fund loan costs about 18.3 percent in cash terms and about 23.6 percent including its much larger warrant. On price alone, the bank offer is far cheaper. But the two loans buy very different amounts of time. Run through the company's monthly cash position, the bank loan extends runway from twelve months to about fourteen and a half; the fund loan extends it to about seventeen. If the Series B milestone needs sixteen months, only the fund loan gets the company there. The bank loan, for all its low cost, would leave the company raising before its milestone, at a lower price, with the bank's revenue covenant hanging over the negotiation. The right choice depends on which is scarcer for this company, money or time, and the headline rate says nothing about either. A company in this position has other possibilities: negotiating a longer interest-only period from the bank, taking a smaller amount from the fund, or combining a small bank facility with a modest equity extension. The point is that each option should be evaluated on the same two measures, effective cost under the expected repayment date and months of runway added, and set against the milestone the company is trying to reach. Other shapes of venture debt The growth-stage term loan is the archetype, but several other structures are common, and each suits a different purpose. Accounts receivable and revenue lines are revolving facilities sized against the company's receivables or, for subscription businesses, its recurring revenue. A software company might be able to borrow, say, a multiple of its monthly recurring revenue, with the available amount rising as revenue grows. These facilities are cheaper than term loans and carry lighter warrant requirements or none, because the lender has a more direct claim on cash the business actually generates. They suit companies with predictable revenue and working-capital needs; they are a poor fit for companies that are still pre-revenue. Equipment financing is the oldest form of venture lending and remains useful for companies with genuine capital expenditure: laboratories, manufacturing lines, data-centre hardware. Because the loan is tied to identifiable assets, it can be cheaper and less restrictive than a general term loan, though the lender will still care about the equity sponsors. At the very large end of the market, as Chapter 6 describes, equipment financing secured on graphics processors has become one of the main ways artificial intelligence infrastructure companies fund themselves. Convertible notes occupy the border between debt and equity. A convertible note is a loan that, instead of being repaid in cash, converts into shares at the next equity round, usually at a discount to the round price or subject to a maximum valuation. In early-stage companies, convertible notes are mostly a convenient way of raising small amounts of equity without negotiating a valuation; they are rarely expected to be repaid, and the lenders are usually the company's existing investors. Larger convertible instruments raised before an initial public offering behave differently, and Chapter 6 examines a notable example. Revenue-based financing gives the lender a fixed percentage of the company's revenue until a set multiple of the original advance, commonly somewhere between 1.1 and 1.5 times, has been repaid. It requires no warrants and does not depend on a venture sponsor, which makes it attractive to companies that do not want, or cannot get, institutional equity. But because repayment comes directly out of revenue, its effective annual cost depends on how quickly the company grows: the faster revenue rises, the sooner the fixed multiple is paid and the higher the annualized rate. The rapid rise and painful recapitalization of Clearco, a Canadian revenue-based lender to e-commerce companies once valued at about $2 billion, which cut some 30 percent of its staff in early 2023 before restructuring with new equity from existing investors and a $100 million asset-backed facility, showed that the model's economics are more fragile than its early boosters claimed, both for the lender and for borrowers who relied on its availability. Reading the term sheet in the right order The practical lesson of this chapter is that the terms of a venture loan should be read, and negotiated, in order of their effect on the borrower under realistic scenarios, not in the order they appear on the page. First, how much money will the company actually have, and when? That depends on the tranches and their conditions, the draw period, the interest-only period and the amortization schedule. A larger commitment with a short interest-only period can provide less usable runway than a smaller one with a long interest-only period. Second, what does the loan cost if it is repaid when the company expects to raise its next round? That depends on the rate, the fees and the prepayment schedule, and should be calculated as an effective annual rate under two or three plausible repayment dates. Third, what does the lender receive if the company does very well? That is the warrant coverage and its terms. Fourth, and most important, what can the lender do if the company does badly? That is the covenant package and the events of default, including the material adverse change clause. Founders tend to negotiate hardest on the interest rate, which is the most visible number and often the least important. A quarter of a point on the rate of a $10 million loan held for two years on the schedule above is worth roughly $46,000. A six-month extension of the interest-only period, or the removal of a tripwire that would let the lender call the loan if the company missed its plan by a few percent, can be worth the difference between surviving a bad year and not. Chapter 3: Warrants and the Lender's Share of the Upside A lender that advances money to a company with a meaningful chance of failing needs to be paid for that risk. It can charge a higher interest rate, but there are limits: a rate high enough to compensate fully for the chance of losing principal would drain the borrower's cash so fast that it would make failure more likely. The venture lending industry's solution, inherited from the equipment lessors of the 1970s and 1980s, is to take a modest amount of interest-rate risk premium and to supplement it with a small claim on the company's equity. That claim takes the form of warrants. For the borrower, warrants are the part of the loan that looks free. They cost no cash, they usually represent a fraction of a percent to a percent or two of the company, and they only have value if the company does well, in which case the founders will be too pleased to care. That reasoning is half right. Warrants are cheap relative to equity, but they are not free, and their cost is concentrated in exactly the outcomes the founders are working towards. A clear understanding of how coverage is calculated, what warrants are worth and which of their terms matter is part of pricing a loan honestly. How coverage is calculated A warrant is an option: the right, but not the obligation, to buy a specified number of shares at a fixed price, the exercise price or strike, at any time before an expiry date. Venture lenders typically receive warrants to buy the same class of preferred shares that the company issued in its most recent equity round, at the price per share of that round, with a life of seven to ten years. The number of shares is set by one of two conventions. In the older and still common approach, the lender negotiates warrant coverage expressed as a percentage of the loan amount. The coverage percentage multiplied by the loan amount gives a dollar figure, and dividing that figure by the exercise price gives the number of shares. A $10 million loan with 8 percent coverage and an exercise price of $1.50 per share gives the lender warrants over $800,000 divided by $1.50, or about 533,000 shares. In the second approach, favoured by many bank lenders, the warrant is expressed directly as a percentage of the company's fully diluted capitalization, often in basis points (hundredths of a percent). A sample term sheet circulated by Kruze Consulting shows warrants over 25 basis points of the fully diluted shares, half vesting at closing and half when the loan is first drawn, and the same guide describes 20 to 35 basis points as standard for that kind of lender. For a company with 40 million fully diluted shares, 25 basis points is 100,000 shares. The two conventions can produce very different outcomes, and it is worth translating one into the other before comparing offers. In the first example, the company had 40 million shares outstanding after its Series A. Issuing 533,000 warrant shares increases that to about 40.5 million, so the lender's warrants represent roughly 1.3 percent of the company, or about 130 basis points. That is several times the bank-style figure. The difference reflects, in part, the different risk each lender takes: non-bank lenders lending larger amounts with fewer controls typically ask for more equity participation. But a borrower comparing "8 percent coverage" from one lender with "30 basis points" from another cannot tell which is more expensive without doing the conversion. Several details affect how many shares the lender ultimately gets. Coverage calculated on the full commitment, rather than on the amount actually drawn, gives the lender warrants over money it may never lend. Many borrowers negotiate for warrants to vest tranche by tranche, so that an undrawn tranche carries no warrants. Some warrants carry a feature that lets the lender choose between the price of the last round and the price of the next round, whichever is lower, which protects the lender if the next round is a down round. Others include an adjustment mechanism under which the number of shares increases if the company issues shares below the warrant price. Each of these shifts value from the existing shareholders to the lender, and each is negotiable. What warrants are worth Because a warrant is an option, it can be valued with option-pricing methods. The standard tool is the model published by Fischer Black and Myron Scholes in 1973, which values an option from five inputs: the current share price, the exercise price, the time to expiry, the risk-free interest rate and the volatility of the share price. For a public company these inputs can be measured. For a startup, the share price is the latest round price and the volatility is borrowed from comparable public companies, so the result is an estimate rather than a market price. It is still a useful discipline. Take the warrants from the first example: a share price and exercise price of $1.50, a ten-year life, a risk-free rate of 4 percent and a volatility of 60 percent, which is in the range often used for early-stage technology companies. The Black-Scholes model values each warrant at about $1.08, or roughly 72 percent of the share price. The lender's 533,000 warrants are therefore worth, on this measure, about $578,000 at the moment the loan is made. Using a shorter seven-year life and a lower 50 percent volatility, the value falls to roughly 56 percent of the share price, or about $450,000. These figures are high because long-dated options on volatile assets are valuable: the holder keeps all of the upside and none of the downside. Several features of private-company warrants make the theoretical value an overstatement of what a lender can actually expect. The shares cannot be sold until the company is acquired or goes public. The warrants are often over preferred shares whose value in a sale depends on a stack of liquidation preferences. And many warrants terminate in an acquisition unless exercised, which the lender will only do if the acquisition price exceeds the exercise price. Accountants valuing such warrants make adjustments for these features, and under US accounting standards warrants over preferred shares that could be redeemed are often recorded as liabilities and remeasured at fair value every reporting period, which means a rising company valuation shows up as a non-cash loss in the borrower's income statement. Even discounted, the warrants are a real cost, and they belong in the effective-cost calculation from the previous chapter. Treating the Black-Scholes value of $578,000 as an additional upfront cost to the borrower raises the effective annual cost of the four-year loan from about 13.1 percent to about 16.2 percent, and the effective cost of the loan repaid after two years from about 15.0 percent to about 19.2 percent. A reasonable summary is that, for a growth-stage loan of this kind, the warrants add something like three to four percentage points to the annual cost of borrowing, on top of the fees. Where the warrant cost falls The distribution of the warrant's value across outcomes matters more than its average. Consider what the 533,000 warrants are worth at the moment of an eventual sale, using the company that Chapter 5 follows through a Series B round and ignoring, for simplicity, the effect of liquidation preferences. After that round the company has about 51.9 million shares. If it is sold for $600 million, each share is worth about $11.57, and the lender's warrants, exercisable at $1.50, are worth about $5.4 million: more than half the principal of the original loan. If the company is sold for $300 million, the warrants are worth about $2.3 million. At $150 million, about $740,000. At $60 million, when each share is worth less than the $1.50 exercise price, they are worthless. From the lender's side, this is the point of the structure. A lender with a portfolio of fifty loans expects several to fail, costing it much of their principal, and hopes for one or two very large outcomes whose warrants make up for those losses. From the borrower's side, the warrant cost is concentrated in success. In the $600 million outcome, the founders and employees would have preferred to pay the lender an extra $5.4 million in cash, if the choice had been available at the start, rather than surrender that value in equity. In the outcomes where the company struggles, the warrants cost nothing at all. This asymmetry cuts in the borrower's favour more often than not. A company that is confident enough of its prospects to borrow on the assumption that it will raise again at a higher price is, by that confidence, expecting the warrants to be worth something. But the warrant cost is small next to the dilution the company avoids. In the Chapter 5 example, the same $10 million raised as equity at $1.50 a share would have required issuing 6.67 million shares, about 14.3 percent of the enlarged company, compared with 1.3 percent for the warrants. At a $600 million sale, those equity shares would have been worth tens of millions of dollars. The warrant is the price of borrowing instead, and measured against that alternative it is modest. Terms that matter in a down round The fine print of a warrant matters most when the company's next round is priced below the last one, which is also when the founders are least able to argue about it. Return to the company with 40.53 million shares after issuing the lender's warrants, and suppose its Series B is priced at a pre-money valuation of $40 million, a steep down round. The new price per share is about $0.99. Under a plain warrant, the lender holds 533,000 warrants at $1.50 that are now out of the money. They may recover if the company does well, but they have lost most of their value. Under a warrant that allows the lender to choose the lower of the last round price and the next round price, and that recalculates the number of shares from the original dollar coverage, the outcome is very different. The lender's $800,000 of coverage is now divided by $0.99 rather than $1.50, giving it about 811,000 shares at the lower price instead of 533,000 at the higher one. The number of warrant shares has risen by about half, and every one of them is exercisable at a third less. The value transferred to the lender comes from the common holders and the earlier investors, at exactly the moment the down round is already diluting them heavily. Two other terms deserve attention. Net exercise, sometimes called cashless exercise, allows the lender to exercise its warrants without paying the exercise price in cash, receiving instead a smaller number of shares equal in value to the warrants' intrinsic value. It is standard and harmless to the borrower, since it simply avoids the lender having to find cash. Treatment on acquisition is more consequential. Some warrants survive a sale of the company and become warrants over the buyer's shares; others are automatically exercised or cancelled. A buyer negotiating to acquire the company will want to know that no warrants will survive, and an awkward warrant can complicate a sale or reduce its price. The simplest structure, from the borrower's point of view, is one in which the warrant is automatically net-exercised immediately before an acquisition if it is in the money, and expires if it is not. Does the warrant make the lender a partner? Lenders sometimes present their warrants as evidence that they share the founders' interest in the company's success, and in a narrow sense they do. A lender holding warrants over 1.3 percent of a company gains from every increase in its value. But the size of that interest relative to the loan determines how the lender will behave when interests diverge, and the numbers are lopsided. In the example above, the lender has $10 million of principal at risk and a warrant worth, on a generous theoretical valuation, about $578,000 at issuance. When the company is healthy, the two interests point the same way: the lender wants the company to grow, both so that the loan is repaid and so that the warrant gains value. When the company is in trouble, they diverge. A decision that raises the chance of a large outcome at the cost of a higher chance of failure, such as spending the remaining cash on one more attempt at a product launch, may be sensible for shareholders but is unattractive to a lender whose principal is worth far more than its warrant. A decision to sell the company early for enough to repay the loan in full, even at a price that leaves little for the shareholders, may be attractive to the lender for the same reason. Economists describe this as the classic conflict between creditors and shareholders over risk-taking, and the warrant softens it without removing it. The equity component is large enough to make the lender care about the upside when things are going well, and far too small to make it behave like a shareholder when things are going badly. Borrowers should therefore take the lender's warrant for what it is, a form of payment, and should not treat it as a reason to expect equity-like patience in a downturn. The patience, if it comes, will come from the lender's reputation, its relationships with the company's investors and its judgement about what it can recover, which is why the lender's history and funding, discussed in Chapter 7, matter as much as the warrant terms. Negotiating the warrant Some lenders offer alternatives to warrants, and borrowers should understand them. A success fee replaces the warrant with a cash payment due if the company is sold or goes public, often expressed as a multiple of the warrant-equivalent value or a fixed percentage of the loan. It avoids issuing equity and removes the accounting complications, but it is a cash claim that must be paid out of sale proceeds, typically ahead of the shareholders. An equity participation right gives the lender the right, but not the obligation, to invest a stated amount in the company's next equity round on the same terms as the other investors. It costs the borrower nothing directly, although it takes up allocation that existing investors might want. The most useful negotiating points are usually structural rather than headline. They include tying coverage to amounts drawn rather than committed; setting the exercise price at the most recent round price, without a look-forward to a lower future price; removing anti-dilution adjustments beyond those given to ordinary preferred shareholders; and ensuring that the warrants either are exercised or expire on an acquisition, so that the buyer does not inherit an awkward outstanding claim. On the coverage percentage itself, the room to negotiate depends on competition: in markets where several lenders are bidding for the same borrower, coverage falls; in markets like the one that followed Silicon Valley Bank's collapse, when lenders were scarce, it rises. The larger point of this chapter is that the warrant is the clearest expression of what a venture lender is. It is a creditor that also holds a thin slice of equity, and it is paid on two different schedules: interest and fees reward it for providing money on time, warrants reward it for having backed a company that turns out to be valuable. A borrower's cost of capital from venture debt is therefore the sum of a cash cost and an equity cost. The cash cost is highest when the loan is short; the equity cost is highest when the company succeeds. Both point in the same direction: venture debt is dearest in exactly the case it is designed for, a company that raises its next round quickly at a much higher valuation, and it is still cheaper than equity in that case. That apparent paradox is resolved in Chapter 5. Hashtags: #VentureDebtAndGrowthCapital #VentureDebt #GrowthCapital #NonDilutiveFinancing #EquityDilution #StartupFinancing #CostOfCapital #EquityBridge #VentureLending #GrowthStageFinancing #VentureBackedCompanies #LoanToValue #IntellectualPropertyCollateral #EquitySponsors #RunwayExtension #InterestOnlyPeriod #LoanAmortization #EffectiveBorrowingCost #VentureDebtWarrants #WarrantCoverage #DebtCovenants #MaterialAdverseChange #PreIPOFinancing #CapitalStructure #FutureOfGrowthFinancing

Download the Book (PDF): Introduction Most people who fall ill from something they ate never find out what it was. They spend a day or two with cramps, fever and diarrhoea, blame the last restaurant meal or a dubious sandwich, and recover. A smaller number end up in hospital. A few, usually the very young, the very old, the pregnant and the immunocompromised, die. Behind a large share of those illnesses stands an animal: a broiler chicken carrying Campylobacter in its gut, a laying hen whose ovaries harbour Salmonella Enteritidis, a steer whose hide is caked with faeces containing Shiga toxin-producing Escherichia coli, a pig with Yersinia in its tonsils. The animal was usually not sick. That is the first and most important fact in this field. The pathogens that cause the most human foodborne illness are, for the most part, quiet passengers in healthy livestock. Veterinary public health is the discipline that sits at this junction. It is the application of veterinary science to the protection of human health, and its territory runs from the farm through the slaughterhouse and processing plant to the laboratory where a stool sample from a sick child is sequenced. It includes meat inspection, the oldest organised food safety activity in most countries. It includes the surveillance and control of zoonoses, the diseases that pass between animals and people. And it includes a problem that has become impossible to ignore over the past three decades: the use of antibiotics in food animals and the resistant bacteria that use selects for. This book is about how those three activities fit together, and about one idea that connects them. The idea is simple to state and surprisingly hard to act on: a pathogen is most cheaply and reliably controlled at the point where it enters the food chain, and every step further downstream is a less efficient, less certain line of defence. The carcass on the inspection line is not where Salmonella is won or lost. By the time a bird reaches the evisceration machine, its intestinal contents and feathers already carry whatever it picked up on the farm, and the plant's job is reduced to limiting the damage. The same logic applies to antimicrobial resistance: once resistant strains are established in a herd, and in the environment around it, no amount of downstream testing will remove them. They must be prevented by changing how animals are raised and how drugs are used. Stated that way, the idea sounds obvious. Its implications are not. It means that the most visible part of food safety regulation, the inspector at the slaughter line, is often the least effective part at preventing the illnesses that matter most today. It means that surveillance systems built to catch outbreaks are valuable in proportion to how well they trace illness back upstream, not merely in proportion to how fast they recall a product. And it means that the hardest questions in the field are not technical at all. They are questions about who should bear the cost of controls applied on farms, where the benefit is felt by consumers far away and diffusely, and about what a regulator is legally permitted to demand of a producer whose product is, by the old legal definitions, perfectly wholesome. Why this matters now Several developments of the past few years make the argument timely. The first is the collapse of the most ambitious attempt in a generation to change how the United States regulates Salmonella in poultry. In August 2024 the Food Safety and Inspection Service (FSIS) of the US Department of Agriculture proposed a framework that would, for the first time, have treated certain levels and types of Salmonella on raw chicken and turkey as adulterants, making product that exceeded them unfit for sale. The proposal was built explicitly on the upstream logic: it asked plants to demonstrate control over incoming flocks. In April 2025, after receiving thousands of comments, the agency withdrew it. The debate that followed, and the more modest pilot programme FSIS announced in mid-2026, show precisely where the difficulties lie. The second is the maturing of whole-genome sequencing as the standard tool of foodborne disease surveillance. Public health laboratories in North America, Europe and increasingly elsewhere now routinely sequence the entire genome of bacterial isolates from sick patients and compare them with isolates from food, animals and processing environments. This has changed what outbreak investigators can see. Illnesses scattered across states and months, which would once have looked unrelated, are now linked with a precision that points investigators toward particular farms, feed mills or hatcheries. The tool that finds outbreaks has become a tool that maps the upstream reservoirs of disease. The third is the uneven progress on antimicrobial use in livestock. The European Union made prophylactic group treatment with antibiotics illegal in 2022 under its veterinary medicines regulation, and sales of veterinary antimicrobials in reporting European countries had roughly halved over the preceding decade. In the United States, by contrast, the Food and Drug Administration's most recent annual summary, released in December 2025, reported a 16 percent rise in sales of medically important antimicrobials for food-producing animals between 2023 and 2024, although sales remained well below their 2015 peak. The contrast between those two trajectories is itself instructive about what works. The fourth is the reminder, delivered by highly pathogenic avian influenza H5N1 in American dairy cattle from 2024 onward, that the animal-human interface can produce surprises. A virus that had spent decades as a poultry and wild bird problem turned up in cows, in their milk, in farm cats that drank raw milk, and in dairy and poultry workers. It is not a foodborne pathogen in the classic sense, but the way it was detected, the way it spread through the movement of animals and equipment, and the role that pasteurisation played in protecting consumers illustrate the same principles that govern Salmonella. What this book covers and what it leaves out The book moves in a line from the animal to the patient and back again. The first chapter sets out what veterinary public health is, how zoonotic foodborne disease actually arises, and why the dominant hazards have shifted from the visible diseases that meat inspection was designed to catch to the invisible carriage that it cannot. The next two chapters deal with meat inspection: first its history, from the organoleptic methods of the early twentieth century to the introduction of Hazard Analysis and Critical Control Point systems in the 1990s, and then how a modern slaughterhouse is actually inspected, with a comparison of American and European approaches and the move toward risk-based inspection. Three chapters then take Salmonella as the central case. One explains the organism itself: why it is so good at living in animals without making them ill, how it moves through different production systems, and what on-farm control programmes in countries such as Sweden and Denmark have achieved. One explains how outbreaks are detected and traced, from serotyping through pulsed-field gel electrophoresis to whole-genome sequencing, and what the investigators do once a cluster appears. One examines the regulatory struggle over Salmonella in American poultry, including the rise and fall of the 2024 framework. Two chapters address antibiotics. The first explains why antibiotics are used in livestock, how resistance develops and moves from animals to people, and what the evidence does and does not show. The second examines the policies that have actually reduced use, in Europe and the United States, and the farming changes that make lower use possible without sacrificing animal welfare. The final chapter uses H5N1 in dairy cattle as a case study in how the same principles apply to an emerging zoonosis. Some subjects are deliberately left out or touched only lightly. Plant-origin outbreaks, such as E. coli on leafy greens, appear only where they connect to animal reservoirs, which they frequently do through contaminated water and manure. Chemical residues, pesticides and veterinary drug residues receive brief treatment rather than a chapter of their own. Parasites such as Trichinella and Taenia are discussed mainly for what they reveal about the origins of meat inspection. Fish and shellfish safety, which has its own regulatory systems, is outside the scope. The emphasis is on the United States and the European Union because their systems are well documented and because the contrasts between them are illuminating, but the principles apply wherever animals are raised for food. A note on evidence Food safety is a field in which numbers are both essential and easily misused. Estimates of the burden of foodborne illness depend on models that correct for underreporting, and different agencies use different methods. Sales figures for antimicrobials measure what was sold, not what was given to animals, still less what was necessary. Outbreak case counts include only people who sought care, gave a sample, and had it tested. Throughout this book, figures are given with their source and with an indication of what they do and do not measure. Where the evidence is contested, the book says so rather than pretending to a consensus that does not exist. The reader does not need a background in microbiology or veterinary medicine. Technical terms are explained as they arise. What the reader does need is patience with an argument that runs against intuition: that the most reassuring parts of the food safety system are not always the most protective ones, and that protecting human health often means spending money and effort on animals that look perfectly well. Chapter 1: The Animal Behind the Meal In 1860 a young servant woman in Dresden fell ill with fever, muscle pain and swelling after Christmas. She died in hospital, and the pathologist Friedrich Zenker found her muscles full of tiny coiled larvae. He traced them to a pig slaughtered on the farm where she worked; the ham and sausages made from it contained the same worms, and others who had eaten them were ill. Zenker had shown, along with Rudolf Virchow and others working in the same years, that Trichinella spiralis passed from pigs to people through meat, and that it could kill. Within a few decades, Prussian cities employed armies of microscopists to examine pork samples for larvae before sale. It was one of the first systematic programmes in which the state examined animal products in order to protect human health, and it established a pattern that lasted a century: find the visibly diseased animal or carcass, remove it, and the public is safe. That pattern worked well against the hazards of its time. Trichinella larvae can be seen in muscle. Tuberculous lesions can be seen in lymph nodes and lungs. The cysts of the pork and beef tapeworms, the bladder-like stages of Taenia solium and Taenia saginata, can be seen when an inspector cuts the heart, the tongue or the chewing muscles. Anthrax, glanders and many other animal diseases produce visible signs in the living animal or its carcass. The inspector's trained eye, nose and hands were matched to the dangers. The foodborne diseases that cause most human illness in industrialised countries today do not work that way. Campylobacter jejuni colonises the intestines of chickens in enormous numbers without making the birds noticeably unwell. Salmonella serotypes that make people seriously ill can live in the gut, lymph nodes and even the ovaries of apparently healthy poultry, pigs and cattle. Shiga toxin-producing E. coli O157:H7 lives in the lower intestine of cattle, which lack the receptors that would let the toxin harm them. Listeria monocytogenes is common in soil, silage and the environments of processing plants. Toxoplasma gondii forms tissue cysts in the muscle of sheep, pigs and other animals that are invisible to the naked eye. None of these organisms reveals itself to an inspector looking at a carcass. That single shift, from visible disease to invisible carriage, is the fact on which modern veterinary public health turns. What veterinary public health is The term covers the contribution of veterinary science to human health, and its components are broader than food. It includes the control of rabies in dogs and wildlife, the surveillance of mosquito-borne viruses that infect horses and birds before people, the management of animal waste, and the safety of the companion animal trade. The World Health Organization has used the term since the mid-twentieth century, and it sits naturally within what is now called One Health: the recognition that human, animal and environmental health are connected and cannot be managed in isolation. Since 2022, the World Health Organization, the World Organisation for Animal Health (WOAH, formerly the OIE), the Food and Agriculture Organization of the United Nations and the United Nations Environment Programme have coordinated their work on this theme as a formal quadripartite. For most people the most important part of veterinary public health is food. The meat, milk and eggs they buy have passed through a chain of controls, and many of those controls are carried out or supervised by veterinarians. In the European Union, official veterinarians supervise slaughterhouses, and certain post-mortem decisions can be made only by them. In the United States, FSIS employs veterinarians as public health veterinarians in plants and at supervisory levels, alongside a much larger number of food inspectors who carry out routine line inspection. In both systems, veterinary knowledge is essential because the question being asked is fundamentally about animal biology: what does this animal carry, and how might it reach a person? How zoonotic foodborne disease arises A useful way to understand any foodborne zoonosis is to trace the pathogen's route from its reservoir to a human gut. Each route has three parts: a reservoir in which the organism survives and multiplies, a set of pathways by which it contaminates food, and a set of conditions that let it reach an infectious dose by the time someone eats it. For most of the major bacterial pathogens, the reservoir is the intestinal tract of the food animal. Campylobacter is the clearest example. Broiler flocks are commonly free of it for their first two to three weeks of life, but once a single bird is colonised, the organism spreads through the flock rapidly, reaching very high counts in the caeca, the blind-ended pouches of the chicken's lower intestine. At slaughter, the birds' feathers and skin are contaminated with faeces from transport crates, and evisceration machines inevitably rupture some intestines. The result is that a large fraction of raw chicken carcasses at retail carry Campylobacter. The organism does not grow on meat at refrigeration temperatures, but it does not need to: the infectious dose for humans can be as low as a few hundred cells. Across the European Union, campylobacteriosis has been the most commonly reported zoonotic disease in humans for well over a decade, with confirmed cases numbering well over one hundred thousand each year according to the annual One Health zoonoses reports of the European Food Safety Authority (EFSA) and the European Centre for Disease Prevention and Control (ECDC). The true number is far higher, because most cases are never tested. The second step, contamination of food, happens by several pathways. Some are direct: faecal material from the gut or hide reaching the carcass surface during slaughter. Some are internal: Salmonella in the lymph nodes of cattle, which end up in ground beef because the fat trimmings used to make it contain lymph nodes that cannot practically be removed. Some are vertical: Salmonella Enteritidis infecting the reproductive tract of hens and being deposited inside the egg before the shell forms. And some are indirect: manure applied to fields, irrigation water contaminated by runoff from feedlots, or wild animals moving between livestock and crops. Several of the largest outbreaks of E. coli O157:H7 in the past two decades were caused by leafy greens rather than beef, but investigators repeatedly found that the source of the contamination was cattle kept on or near the growing areas. The third step, reaching an infectious dose, depends on what happens between the processing plant and the plate. Bacteria that survive on raw meat can be killed by cooking, which is why the vast majority of contaminated chicken never makes anyone ill. But they can also cross-contaminate: a cutting board used for raw chicken and then for salad, hands that handle raw meat and then a sandwich. Some organisms, Salmonella and Listeria among them, can multiply in food if it is held at the wrong temperature. Listeria can grow even in the refrigerator, which is why it is especially dangerous in ready-to-eat foods such as deli meats and soft cheeses that are not cooked before being eaten. The pathogen that makes someone ill is therefore the product of a chain of events, most of which happened far from the person who ate the food and long before. This is why the controlling idea of this book matters. Every link in that chain is a possible control point, but the links differ greatly in how much they can achieve. Cooking kills most bacteria, but it depends on the behaviour of hundreds of millions of individual consumers and cooks, and it does nothing for cross-contamination. Interventions at the processing plant, such as antimicrobial washes and chilling, reduce the load but cannot sterilise raw meat. The farm is where the load originates, and a flock or herd that does not carry a pathogen cannot pass it on. The burden of disease Estimating how many people are made ill by food is harder than it sounds, because only a small fraction of cases is ever diagnosed. For every laboratory-confirmed case of salmonellosis reported to public health authorities, many more people were ill and never saw a doctor, or saw one who did not order a stool test, or had a test that was not reported. Agencies correct for this using multipliers derived from studies of each step in the reporting pyramid. The US Centers for Disease Control and Prevention (CDC) estimates that Salmonella causes about 1.35 million infections, 26,500 hospitalisations and 420 deaths in the United States each year, with food the source of most of these illnesses. CDC's broader estimates, first published in 2011, put the total burden of domestically acquired foodborne illness from all causes at roughly 48 million illnesses, 128,000 hospitalisations and 3,000 deaths a year. At the global level, the World Health Organization's Foodborne Disease Burden Epidemiology Reference Group estimated in 2015 that foodborne hazards caused about 600 million illnesses and 420,000 deaths worldwide in 2010, with the heaviest burden falling on children under five and on low-income regions of Africa and South-East Asia. These are estimates with wide uncertainty bounds, and they are periodically revised. But their scale makes the point. Foodborne disease is not a rare event visible mainly through dramatic outbreaks; it is a large, persistent, mostly invisible burden, and a substantial part of it originates in animals. It is also important to know which foods are responsible. This is the problem of source attribution, and it is harder than it seems. An outbreak with a known source contributes clear evidence, but the great majority of illnesses are sporadic, meaning they are not part of any recognised outbreak. For those, attribution relies on indirect methods: comparing the types of Salmonella found in patients with those found in different animal and food sources, analysing outbreak data statistically, and conducting case-control studies. In the United States, the Interagency Food Safety Analytics Collaboration, a partnership of CDC, FDA and FSIS, publishes annual estimates of which food categories cause illnesses from the major pathogens. Its reports consistently attribute a large share of Salmonella illness to a broad range of foods, with chicken, pork, eggs, and various produce categories all prominent, and a large share of Campylobacter illness to chicken. Such estimates are imprecise, but they direct regulatory attention to the right places. From disease in animals to carriage in animals The change in the nature of the hazard has not been matched by a complete change in the systems designed to deal with it. This is not because regulators are unaware of it. The mismatch has been recognised in official reports on both sides of the Atlantic since at least the 1980s. It persists because the old system has strengths that should not be discarded, because it is embedded in law, and because the alternative is harder to design and to pay for. Traditional inspection still catches things that matter. Animals that arrive at slaughter visibly sick, lame or dead are identified and removed. Carcasses with abscesses, tumours, septicaemia or extensive bruising are condemned or trimmed. Tuberculosis, though rare in most developed-country livestock, is still detected at slaughter, and the slaughterhouse remains an important surveillance point for bovine tuberculosis in countries such as the United Kingdom and Ireland where the disease persists in cattle and wildlife. The removal of specified risk materials, the brain, spinal cord and other tissues that can carry the agent of bovine spongiform encephalopathy, is enforced at slaughter. The inspector's presence also has a deterrent effect: it is more difficult to process diseased animals when a government official is watching. What traditional inspection cannot do is detect the bacteria that cause most illness. A carcass contaminated with Salmonella looks, smells and feels exactly like one that is not. The inspector cannot see Campylobacter on a chicken. Neither can the plant's own staff. The only way to know is to take samples and test them in a laboratory, and even then the results describe the lot that was sampled, not the next one. This is why the reform of meat inspection that began in the 1990s shifted the emphasis from inspecting products to verifying processes. If you cannot see the hazard in the product, you must control the process that creates or removes it, and demonstrate that control through records and testing. That shift, from inspection to prevention, is the subject of the next chapter. But even process control in the slaughterhouse has limits, and those limits are what drive attention further upstream. The case for going upstream Three examples from the past forty years show what upstream control can achieve. The first is Salmonella Enteritidis in eggs. In the 1980s, both the United Kingdom and the United States experienced a surge in human infections with this serotype, much of it traced to eggs. The organism was infecting the hens' ovaries and oviducts and being deposited inside the eggs, so that washing and grading could not remove it. In the United Kingdom, the problem became a political crisis in 1988 when a junior health minister, Edwina Currie, said that most of the country's egg production was affected; egg sales collapsed and she resigned. The response that eventually worked was upstream: vaccination of laying flocks against Salmonella, which became standard in the British egg industry under the British Lion scheme launched in 1998, combined with testing and culling of infected flocks. Human Salmonella Enteritidis infections in the United Kingdom fell substantially in the following years. In the United States, the FDA's Egg Safety Rule, which took effect for large producers in 2010, required on-farm measures including rodent control, biosecurity, testing of the poultry house environment, and refrigeration of eggs. Its adoption was accelerated by a 2010 outbreak linked to eggs from Iowa, which led to a recall of roughly half a billion eggs. The second is Sweden's Salmonella control programme. After an outbreak in 1953 linked to meat from a slaughterhouse in Alvesta, which sickened thousands of people and killed about ninety, Sweden built a programme that tested animals and feed for Salmonella and removed or treated infected flocks and herds. The approach has been sustained for decades, and Sweden, like Finland and Norway, has maintained a very low prevalence of Salmonella in its domestic livestock. When Sweden and Finland joined the European Union in 1995, they negotiated special guarantees allowing them to require that imported meat and eggs be tested for Salmonella. Their experience is often cited as proof that upstream control is achievable. The third is Denmark's pig Salmonella programme, begun in the 1990s after outbreaks linked to pork. Denmark tests meat juice from slaughtered pigs for antibodies to Salmonella, uses the results to classify herds, and imposes financial penalties and requirements for improved hygiene on herds with high levels. The programme has operated for three decades and is described in more detail in Chapter 4. None of these examples is without cost, and none eliminated Salmonella from human illness, because people also acquire it from imported food, travel, pets, reptiles and produce. But they show that the reservoir can be shrunk. What they also show is that shrinking it requires sustained, expensive, and sometimes politically uncomfortable measures on farms, where most of the cost is borne by producers and most of the benefit goes to the public at large. Antibiotics and the same logic The problem of antimicrobial resistance follows the same pattern. Bacteria in food animals are exposed to antibiotics used for treatment, prevention and, historically, growth promotion. Resistant strains are selected, and they can reach people through food, direct contact with animals, and the environment. Once a resistant strain or a resistance gene is established in a population of animals, it is very difficult to remove. The way to keep it out is to use antibiotics less and more carefully in the first place. That is not the same as saying that animals should not be treated. Sick animals need treatment, both for their own welfare and because untreated disease causes suffering and economic loss. The argument, which Chapters 7 and 8 develop, is that the routine, preventive use of antibiotics in healthy animals to compensate for crowding, poor hygiene, early weaning and stress is the use most amenable to reduction, and the use whose reduction brings the most benefit for human health. The shape of the problem The picture that emerges is of a system with several layers, each with different strengths. Farm-level controls prevent pathogens from entering the food chain. Slaughter and processing controls reduce the load on carcasses and products. Surveillance detects outbreaks and traces them back. Regulation sets the incentives that determine how much effort each actor puts in. Consumer handling is the last layer and the least reliable. Most of the public attention, and much of the legal architecture, is concentrated on the middle layers: the slaughterhouse and the recall. The argument of this book is that the outer layers, the farm on one side and the surveillance system that connects human illness back to the farm on the other, deserve a much larger share of attention than they usually receive. The chapters that follow examine each layer in turn, starting with the institution that has defined veterinary public health for more than a century: the meat inspector. Chapter 2: From Poke-and-Sniff to HACCP Modern American meat inspection was born of a novel. Upton Sinclair published The Jungle in 1906 to expose the conditions of immigrant workers in Chicago's stockyards, but readers fixed on the passages describing what went into the meat: spoiled product treated with chemicals, rats and poisoned bread shovelled into sausage hoppers, tubercular carcasses passed for sale. Sinclair's famous complaint was that he had aimed at the public's heart and hit it in the stomach. President Theodore Roosevelt, sceptical of Sinclair but alarmed by the public reaction, sent investigators to Chicago. Their report confirmed much of what the novel described. Within months Congress passed the Federal Meat Inspection Act, alongside the Pure Food and Drug Act, and the United States had a system of continuous federal inspection in plants that shipped meat across state lines. The system created in 1906 was built around the hazards of the day and the methods available to detect them. Every animal was to be examined before slaughter, and every carcass after it. The post-mortem examination relied on the inspector's senses: looking at the carcass and organs, palpating lymph nodes and tissues, making incisions in specified places to look for lesions, and smelling for signs of spoilage or disease. This approach became known in the trade as organoleptic inspection, and later, less kindly, as poke-and-sniff. It was good at what it was designed to do. It removed from the food supply carcasses showing tuberculosis, abscesses, parasitic cysts, septicaemia, and many other conditions that were common in livestock at the start of the twentieth century and that made meat both repulsive and dangerous. The same model spread to poultry, first through voluntary programmes and then through the Poultry Products Inspection Act of 1957, which made inspection mandatory for birds moving in interstate commerce. The Wholesome Meat Act of 1967 and the Wholesome Poultry Products Act of 1968 required states to operate inspection programmes at least equal to the federal one for plants selling only within their borders. By the 1970s, the American system was large, expensive and in some respects highly effective. It was also increasingly aimed at the wrong target. The problem the inspectors could not see The prevalence of the diseases that organoleptic inspection was designed to catch fell sharply over the twentieth century, as eradication programmes for tuberculosis and brucellosis succeeded and as animal husbandry improved. Meanwhile, the bacterial pathogens that were becoming the main causes of foodborne illness, Salmonella above all, were invisible to the methods in use. This was recognised early. In the early 1970s, the American Public Health Association sued the Department of Agriculture, arguing that raw meat and poultry carrying Salmonella should bear a warning label. The agency argued that Salmonella was naturally present on raw meat, that it was destroyed by proper cooking, and that consumers knew raw meat had to be handled carefully. In 1974, in American Public Health Association v. Butz, the federal appeals court in Washington agreed with the Department. The court's reasoning was that the presence of Salmonella in raw meat did not make it adulterated, because the American public knew that raw meat was not to be eaten without cooking, and properly cooked meat was safe. That decision cast a long shadow. For decades it was taken to mean that Salmonella on raw meat and poultry was not an adulterant under the inspection laws, which placed a legal ceiling on what regulators could demand. In 1985, a committee of the National Research Council of the National Academy of Sciences published an evaluation of the scientific basis of the federal meat and poultry inspection programme. It concluded that the system was adequate for detecting the diseases it was designed for, but that it did not address the microbial and chemical hazards that were now the main threats to health. It recommended that inspection should move toward a system based on hazard analysis and the identification of critical control points. A companion report on poultry inspection, published in 1987, made similar arguments. For several years, not much happened. Jack in the Box and the O157 decision The event that forced change came in January 1993. Children in Washington State began arriving at hospitals with bloody diarrhoea, and some developed haemolytic uraemic syndrome, a complication in which the Shiga toxin damages the kidneys and blood vessels. Investigators traced the illnesses to undercooked hamburgers sold at restaurants of the Jack in the Box chain. The outbreak ultimately involved more than 700 people in four western states, and four children died. E. coli O157:H7 had been identified as a human pathogen only a decade earlier, after outbreaks linked to hamburgers in Oregon and Michigan in 1982. Its reservoir was cattle, which carry it in their intestines without illness. It reached beef through faecal contamination of hides and carcasses during slaughter, and grinding spread contamination from the surface of a piece of meat throughout a batch of ground beef, where it would survive in a hamburger that was not cooked through. Its infectious dose was very low. And unlike Salmonella, it could cause catastrophic disease in young children. The Jack in the Box outbreak had an effect that years of expert reports had not. Parents of the children who died became effective advocates for reform. In 1994, Michael Taylor, the newly appointed administrator of FSIS, announced in a speech to meat industry executives that the agency considered E. coli O157:H7 in raw ground beef to be an adulterant. This was a significant legal step: it meant that ground beef found to contain the organism could not be sold, and that the agency would begin testing for it. Industry groups sued, arguing that the decision was inconsistent with the logic of Butz, since ground beef, like other raw meat, was intended to be cooked. A federal court in Texas rejected their challenge, and the policy stood. The distinction the agency drew, and which the courts accepted, rested on the particular characteristics of O157:H7: its very low infectious dose, the severity of the illness it caused, and the fact that ordinary cooking of hamburgers, which many consumers served rare or medium, could not be relied upon to destroy it. The same reasoning was later extended. In 2011, FSIS announced that six additional serogroups of Shiga toxin-producing E. coli, often referred to as the "big six" (O26, O45, O103, O111, O121 and O145), would also be treated as adulterants in raw non-intact beef and in the trimmings used to make it, with testing beginning in 2012. The question of whether the same reasoning could be applied to Salmonella in poultry would remain open for three decades, and it is the subject of Chapter 6. The HACCP revolution The more fundamental change was the adoption of Hazard Analysis and Critical Control Point systems. HACCP had its origins in the American space programme. In the early 1960s, the Pillsbury Company worked with NASA and the US Army's laboratories at Natick, Massachusetts, to produce food for astronauts that would be safe with near certainty. Testing finished products could not provide that assurance, because testing enough units to be confident destroyed most of the product. The alternative was to analyse the production process, identify the points where hazards could be introduced or controlled, set limits at those points, monitor them continuously, and keep records. The approach was presented publicly at the National Conference on Food Protection in 1971, and it was adopted in the regulation of low-acid canned foods in the United States during the 1970s. HACCP's principles are now codified internationally in guidelines of the Codex Alimentarius Commission, the joint body of the Food and Agriculture Organization and the World Health Organization that sets food standards. In simplified form, a business must conduct a hazard analysis, identify critical control points where a hazard can be prevented, eliminated or reduced to an acceptable level, establish critical limits at each, establish monitoring procedures, establish corrective actions for when monitoring shows a limit has been breached, establish verification procedures, and keep records. The system is preventive rather than reactive: its purpose is to ensure that unsafe product is not produced in the first place, rather than to find it afterwards. In July 1996, FSIS published the Pathogen Reduction/HACCP final rule, which required all federally inspected meat and poultry plants to develop and implement HACCP plans. Large plants had to comply by 1998 and the smallest by 2000. The rule also required plants to have written sanitation standard operating procedures, required slaughter plants to test carcasses for generic E. coli as an indicator of faecal contamination, and established Salmonella performance standards for various classes of product. Under those standards, FSIS would take samples from each plant, and a plant that exceeded the allowed number of positive samples in a set would be required to take corrective action. The rule fundamentally changed the relationship between inspector and industry. Under the old system, the inspector was, in effect, the quality control department of the plant, deciding which carcasses passed and which did not. Under HACCP, the plant was responsible for the safety of its products and had to demonstrate how it achieved it. The inspector's role shifted toward verification: checking that the plant was following its own plan, reviewing records, and taking samples to test whether the plan was working. This shift was controversial from the start. Critics argued that it allowed the industry to write its own rules and reduced the role of independent government oversight. Supporters argued that it placed responsibility where it belonged, on the company that controlled the process, and that it focused regulatory attention on the invisible microbial hazards that caused most illness. Both arguments had merit, and the debate continues in different forms. The limits of the performance standard The Salmonella performance standards were an important part of the new system, and they quickly ran into legal trouble that shaped everything that followed. In 1999 and 2000, a Texas company called Supreme Beef Processors repeatedly failed the Salmonella performance standard for ground beef. FSIS moved to suspend inspection at the plant, which would have shut it down. The company sued, and in 2001 the federal appeals court for the Fifth Circuit ruled in its favour. The court held that the agency could not use the Salmonella performance standard as the sole basis for concluding that a plant's products were adulterated, because Salmonella on raw beef was not itself an adulterant and because the Salmonella in the ground beef might have come from the incoming meat rather than from insanitary conditions in the plant. The company ground beef from trimmings it bought from other plants; if those trimmings arrived contaminated, the court reasoned, the grinder's own sanitation could not be blamed. That reasoning is worth dwelling on because it touches directly on the controlling idea of this book. The court recognised, in effect, that contamination originates upstream, and it concluded that the downstream plant could not therefore be held legally responsible for it through a performance standard. The agency's response was to continue using the performance standards as a tool to trigger further investigation and corrective action rather than as a basis for shutting plants down, and to publish the results in categories that created reputational pressure. The standards were revised several times, and in 2011 and 2016 new standards were introduced for chicken parts and ground poultry, along with standards for Campylobacter. But the fundamental legal problem remained: the agency could ask for control but, as long as Salmonella was not an adulterant in raw poultry, it had limited power to demand it. The European parallel Europe followed a broadly similar path, but through a different legal architecture. National meat inspection systems developed from the nineteenth century onward, strongly influenced by the German tradition of veterinary public health that Virchow and his successors had built. The European Economic Community began harmonising meat hygiene rules in the 1960s, largely to allow trade in fresh meat between member states. The major reform came with the so-called hygiene package of 2004, a set of regulations that consolidated and replaced the many sectoral directives. Regulation (EC) No 852/2004 set general hygiene rules for all food businesses and required them to implement procedures based on HACCP principles. Regulation (EC) No 853/2004 set specific hygiene rules for food of animal origin, including slaughterhouses. Regulation (EC) No 854/2004 set rules for the official controls of products of animal origin, including meat inspection. The official controls regulation was later replaced by Regulation (EU) 2017/625, which applies across the food chain, with detailed rules for meat inspection set out in Implementing Regulation (EU) 2019/627. In parallel, the European Union developed a legal framework for controlling Salmonella on farms. Regulation (EC) No 2160/2003 required member states to establish national control programmes for Salmonella in specified animal populations, beginning with breeding flocks of chickens and moving on to laying hens, broilers and turkeys, with reduction targets set at EU level and focused on the serotypes most significant for public health. Regulation (EC) No 2073/2005 established microbiological criteria for foods, including criteria for Salmonella in various meat products. Unlike in the United States, the European framework thus placed binding obligations at the farm level, and it set targets for flocks rather than only for products. Trade and equivalence One further force shaped both systems: international trade. The World Trade Organization's Agreement on the Application of Sanitary and Phytosanitary Measures, which came into force in 1995, allows countries to set their own levels of protection for food safety and animal health, but requires that measures be based on science and not used as disguised trade barriers. It names the Codex Alimentarius Commission and the World Organisation for Animal Health as the reference bodies for international standards. The practical consequence is that exporting countries must show that their systems achieve an equivalent level of protection to that of the importing country, even if they use different methods. FSIS maintains a list of countries whose meat and poultry inspection systems it has found equivalent, audits them, and reinspects imported product at the border; the European Union audits third countries through its own inspection service and lists approved establishments. Equivalence determinations have been the site of long disputes, of which the most famous is the European Union's refusal to accept American poultry treated with chlorine and other antimicrobial washes, on the grounds that such treatments could substitute for hygiene earlier in the chain. That dispute is, at root, about the same question this book keeps returning to: whether it is acceptable to clean up at the end of the line what could have been prevented at the beginning. That difference, between a system that regulates primarily at the slaughterhouse and one that reaches back into the flock, is one of the most instructive contrasts in the field. The European approach has not eliminated Salmonella, and it has its own weaknesses, but its steady reduction of the most dangerous serotypes in poultry flocks is evidence that the upstream approach can work at scale. What HACCP changed and what it did not HACCP was a genuine advance. It moved food safety from inspecting products to controlling processes, and it did so in a way that could be verified. Measures of carcass contamination in American plants fell after its introduction, and the industry adopted a range of interventions, such as organic acid sprays, hot water washes, steam pasteurisation of beef carcasses and antimicrobial chilling of poultry, that have reduced surface contamination. What HACCP did not change is the basic geography of the problem. A HACCP plan in a slaughterhouse controls what happens inside the slaughterhouse. It cannot control what the animals bring with them. A poultry plant can have excellent process control and still produce contaminated product if the incoming flocks are heavily colonised with Salmonella, because no combination of washes and chills can remove all of it from the skin and cavities of a bird. The plant can reduce the load, sometimes substantially, but it starts from wherever the farm left it. The implications of this, and the attempts by regulators on both sides of the Atlantic to extend control upstream, are the thread that runs through the rest of this book. Before following it, however, the next chapter looks more closely at how a modern slaughterhouse is actually inspected today, and at the reforms of the past two decades that have changed what inspectors do. Chapter 3: Inside the Modern Slaughterhouse A large American broiler plant can slaughter more than two birds every second, and a large pork plant more than a thousand pigs an hour. At those speeds, the image most people have of meat inspection, a veterinarian carefully examining each animal, collapses. Something still happens to every carcass, because American law requires it. But what an inspector can do in a fraction of a second, and what an inspection system should be asking inspectors to do, are questions that have preoccupied regulators for thirty years. This chapter describes how a slaughterhouse is inspected today, how the American and European systems differ, and where both are heading. Before slaughter: the living animal Inspection begins before the animal dies. Ante-mortem inspection is the examination of live animals on arrival at the slaughterhouse and before slaughter. Its purposes are to identify animals that are too sick to be slaughtered for food, to separate those that need closer examination after slaughter, to detect notifiable diseases such as foot-and-mouth disease, and to check animal welfare during transport and handling. In the United States, animals that show signs of certain conditions at ante-mortem inspection are classed as "US suspect" and slaughtered separately so their carcasses can receive a more detailed post-mortem examination; others are condemned outright. Cattle that cannot stand or walk on arrival, known as non-ambulatory disabled or "downer" cattle, must be condemned and humanely euthanised. That rule was tightened after 2008, when undercover video from a slaughterhouse in Chino, California, operated by the Hallmark/Westland Meat Packing Company, showed workers using forklifts, electric prods and water hoses to force downer cattle to their feet so they would pass inspection. The ensuing recall, of about 143 million pounds of beef, was the largest in American history, much of it already eaten, and it included beef supplied to the national school lunch programme. The concern was partly animal welfare and partly the elevated risk that downer cattle might carry bovine spongiform encephalopathy, which is more common in animals that cannot stand. European rules place more weight on information from the farm. Under EU law, slaughterhouse operators must receive and assess food chain information for each batch of animals before they arrive: information about the holding of origin, any veterinary medicines administered and their withdrawal periods, diseases that may affect meat safety, the results of relevant laboratory tests, and production data such as mortality rates that might indicate disease. The official veterinarian uses this information to plan ante-mortem and post-mortem inspection and to decide whether a batch needs particular attention. The principle is an explicitly upstream one: a slaughterhouse should know what it is receiving before the animals come off the truck. In practice, the quality of food chain information has varied widely. For poultry, where flocks are raised by a small number of integrated companies with good record systems, it can be detailed and useful. For cattle and pigs, especially from smaller farms or animals traded through markets, it may amount to a form declaring that nothing is wrong. But the legal principle is important, and it underpins the European move to risk-based inspection described later in this chapter. After slaughter: the carcass and organs Post-mortem inspection is the examination of the carcass and its organs after slaughter. For cattle and pigs, traditional post-mortem inspection involves looking at the carcass surfaces and body cavities, examining the head, the lymph nodes of the head, the tongue, the lungs, the heart, the liver and the other organs, and incising particular lymph nodes and tissues to look for lesions. In cattle, the chewing muscles and heart are incised to look for the cysts of the beef tapeworm, Taenia saginata. The carcass, its head and its organs must be identified so that they can be matched if any is found to be abnormal. For poultry, the process is necessarily quicker. Birds move past inspection stations on a line of shackles, and the inspector looks at the inside and outside of the carcass and the organs for signs of disease such as septicaemia, airsacculitis, tumours and inflammatory processes, as well as for contamination with faeces or ingesta. Birds with systemic disease are condemned; those with localised problems can be trimmed or reprocessed. Beyond disease, post-mortem inspection enforces rules about specified risk materials in cattle, the tissues that can carry the prion agent of bovine spongiform encephalopathy. Since the BSE crisis of the 1990s, the brain, spinal cord, parts of the vertebral column, tonsils, and parts of the small intestine of cattle above certain ages must be removed and destroyed rather than entering the food chain. BSE is now extremely rare, and the international status of most countries has improved to negligible risk, but the removal rules remain a critical part of slaughter hygiene. Carcasses in many systems are also sampled for residues of veterinary drugs, pesticides and environmental contaminants. The US National Residue Program, run by FSIS with other agencies, samples carcasses both randomly and in response to suspicion, for example when an animal shows injection-site lesions or comes from a producer with a history of violations. The European Union requires member states to operate residue monitoring plans, now under the official controls regulation and its associated delegated acts. Residue violations are relatively uncommon, and the main public health concern about antimicrobial drugs in food animals has shifted from residues in meat to the selection of resistant bacteria on farms, which residue testing does not address. The line speed problem In American poultry plants, the traditional inspection system required a government inspector to examine every bird, and the number of inspectors on a line limited how fast the line could run. This arrangement was increasingly seen as inefficient by the industry and inadequate by the agency. Inspectors were spending their time looking for visible defects, many of them quality issues rather than safety hazards, while the microbial hazards that caused human illness went largely unaddressed by their examination. Starting in the late 1990s, FSIS ran a pilot programme in a small number of plants, the HACCP-based Inspection Models Project, in which plant employees took over the task of sorting carcasses and removing those with defects, and government inspectors were redeployed to oversee the process and carry out verification tasks. In 2014, after years of controversy, FSIS published a final rule establishing the New Poultry Inspection System. Plants that opted into it would take responsibility for sorting carcasses, with one government online inspector per line examining carcasses at the end of the process and additional inspectors carrying out offline verification, including sampling. The maximum line speed for young chickens was set at 140 birds per minute; plants under the old system were limited to lower speeds. From 2018, FSIS allowed some plants to operate at up to 175 birds per minute under waivers. A similar system for pigs, the New Swine Slaughter Inspection System, was finalised in 2019 and removed the maximum line speed for participating plants. In 2021 a federal court in Minnesota vacated the line speed provision, in a case brought by unions who argued that FSIS had not properly considered worker safety, and FSIS subsequently allowed some plants to run at higher speeds under a time-limited trial while it collected data. In February 2026, FSIS published proposals to raise the permitted speeds for poultry plants under the new system and to remove the maximum line speed for swine plants under the new swine system, leaving inspectors authority to slow a line when food safety or inspection is compromised. At the time of writing, those proposals had not been finalised. The debate over line speeds has two strands that are often confused. One is about worker safety: faster lines increase repetitive strain injuries and other hazards for plant employees, who already work in one of the more dangerous occupations in the American economy. That concern is serious but falls outside veterinary public health as such, and FSIS has taken the position that worker safety is primarily the responsibility of the Occupational Safety and Health Administration. The other strand is about food safety: whether faster lines, with fewer government eyes on each carcass, increase the risk of contaminated or diseased product reaching consumers. The evidence on the second question is less clear than advocates on either side tend to claim. FSIS has argued that the redeployment of inspectors from sorting carcasses to verification tasks such as sampling and oversight of sanitation produces better food safety outcomes, and it has pointed to Salmonella data from plants in the pilot programme. Critics have argued that the data are limited and that plant employees sorting carcasses face commercial pressure to pass product. What is clear is that the question of how fast birds pass a government inspector is, by itself, not a good measure of how safe the product is, because the inspector at the line cannot see the hazard that matters most. The real question is whether the whole system, from farm through slaughter to verification testing, is reducing pathogen loads, and that question can only be answered with microbiological data. Europe and risk-based meat inspection The European Union has approached the same problem from a different direction. Beginning around 2010, the European Commission asked EFSA to assess whether traditional meat inspection methods were appropriate for the public health hazards now of greatest concern. EFSA published scientific opinions on meat inspection for pigs, poultry, cattle, small ruminants, farmed game and solipeds between 2011 and 2013. The conclusions were broadly consistent. The main public health hazards associated with meat were microbial, notably Salmonella, Campylobacter, pathogenic Yersinia enterocolitica, Shiga toxin-producing E. coli and Toxoplasma gondii, and none of them was detected by traditional visual inspection, palpation or incision. Palpation and incision, in fact, could spread contamination from one carcass to another or from one part of a carcass to another, because the inspector's hands and knife moved bacteria around. EFSA recommended an integrated system of meat safety assurance, in which information from farms, risk-based categorisation of herds and flocks, process hygiene measures in slaughterhouses and targeted testing would work together. The most concrete change so far concerns pigs. Commission Regulation (EU) No 219/2014 changed the rules so that post-mortem inspection of fattening pigs raised under controlled housing conditions would be visual only, with palpation and incision required only where the visual inspection, food chain information or other data indicated a possible risk. The reasoning was that palpation and incision of lymph nodes and organs in healthy pigs added little public health benefit and increased cross-contamination. Similar changes have been made for other species in specified circumstances. The broader integrated system EFSA envisioned, in which herds are classified by risk and inspection effort is allocated accordingly, has been adopted only partially, and member states vary considerably in how they implement it. European law also sets process hygiene criteria for slaughterhouses: microbiological limits that, if exceeded, require the business operator to improve hygiene and review its process controls. In 2018, the European Union introduced a process hygiene criterion for Campylobacter on the neck skin of broiler carcasses after chilling, with a limit of 1,000 colony-forming units per gram and a specified number of samples allowed to exceed it, the number of allowed exceedances being reduced in stages. This was a significant step, because Campylobacter is the most common bacterial cause of foodborne illness in Europe and the process hygiene criterion creates a direct incentive for slaughterhouses to reduce it. But it is a process criterion, not a food safety criterion: a batch that exceeds the limit is not withdrawn from sale, and the obligation falls on the slaughterhouse, not the farm that supplied the birds. Comparing the systems The American and European systems share a common structure: continuous official presence in slaughterhouses, ante-mortem and post-mortem inspection, HACCP-based obligations on businesses, and microbiological sampling. They differ in several ways that matter for the controlling idea of this book, as Table 1 summarises. Table 1. Main features of meat inspection approaches in the United States and the European Union. Feature US traditional inspection US modernised systems (NPIS, NSIS) EU official controls Who sorts carcasses for defects Government inspectors Plant employees, with government oversight Official veterinarian and auxiliaries, with plant staff Use of farm information Limited Limited Food chain information required for each batch Post-mortem method for healthy pigs Visual, palpation and incision Visual, palpation and incision (sorting by plant) Visual only under controlled housing, with targeted incision Microbial limits in slaughter plants Salmonella performance standards Salmonella performance standards Process hygiene criteria, including Campylobacter on broilers On-farm Salmonella obligations None federal for meat birds None federal for meat birds National control programmes with EU targets Source: FSIS regulations (9 CFR parts 309, 310, 381); Regulation (EU) 2017/625; Implementing Regulation (EU) 2019/627; Regulation (EC) No 2160/2003. The most important difference is the last row. The European Union places legal obligations on poultry farms to control Salmonella, through national control programmes and flock-level targets, while the American system relies on slaughterhouse-level performance standards and, for eggs, on the FDA's egg safety rule. This difference is not an accident; it reflects both different legal authorities and different political choices. FSIS has jurisdiction over slaughter and processing but no direct authority to regulate farms. It can create incentives for plants to demand cleaner birds from their suppliers, and much of the design of the withdrawn 2024 framework was aimed at doing exactly that, but it cannot directly require farms to test and act. The slaughterhouse as an observatory One function of the slaughterhouse is often overlooked: it is the single point through which almost every food animal passes, which makes it an unusually efficient place to observe the health of the national herd and flock. Diseases that are hard to find on farms can be found at slaughter. In Britain and Ireland, where bovine tuberculosis persists in cattle and badgers, a substantial share of newly infected herds is first identified when an inspector finds suspicious lesions in a slaughtered animal that had tested negative on the farm. Parasitic conditions such as liver fluke, which affects productivity and signals environmental risk, are recorded at slaughter and can be reported back to farmers. The European Union's rules require that the results of inspection be communicated to the farm of origin and to the veterinarian responsible for it, so that problems found at slaughter can be corrected where they arose. Slaughter sampling also underpins antimicrobial resistance surveillance. In the United States, the National Antimicrobial Resistance Monitoring System, a partnership of FDA, CDC and USDA established in 1996, has collected samples of intestinal contents from food animals at slaughter since the 2010s, alongside samples of retail meat and isolates from human patients. The European Union requires member states to test indicator bacteria and zoonotic bacteria from caecal samples at slaughter and from retail meat on a harmonised rotating schedule. Those data are the main source of information on how resistance in food animals is changing, and they depend on the sampling infrastructure that inspection provides. The slaughterhouse, in other words, can look back upstream, provided that the information it generates is actually used on the farms it came from. What inspection is for The honest summary of the state of meat inspection is that it does several valuable things and one thing that it was never designed to do and cannot do. It removes visibly diseased animals and carcasses. It enforces the removal of specified risk materials. It provides surveillance for notifiable animal diseases, for residues and, increasingly, for antimicrobial resistance, since sampling at slaughter is a key source of data for resistance monitoring programmes. It verifies that businesses follow their own food safety plans. And it provides a continuous official presence that has a deterrent effect, as the Hallmark/Westland scandal showed by counterexample: that abuse occurred when inspectors were not present. What inspection at the line cannot do is detect the microbial contamination that causes most foodborne illness. The response of the past thirty years, HACCP and microbiological testing, has improved the situation substantially, but it has also run into limits. A plant can control what it does. It cannot control what its animals bring through the door. That is the reason attention has increasingly turned to the farm and to surveillance systems capable of connecting human illness to its source, and it is why the next chapters focus on the organism that has most tested the limits of slaughterhouse-based control: Salmonella. Hashtags: #VeterinaryPublicHealth #FoodSafety #ZoonoticPathogenControl #OneHealth #FoodborneZoonoses #SalmonellaControl #CampylobacterControl #ShigaToxinProducingEColi #ListeriaMonocytogenes #MeatInspection #RiskBasedInspection #HACCP #FoodChainSafety #OnFarmBiosecurity #ZoonoticSurveillance #WholeGenomeSequencing #OutbreakInvestigation #SourceAttribution #AntimicrobialResistance #AntimicrobialStewardship #LivestockHealth #SlaughterhouseHygiene #PathogenReduction #UpstreamPrevention #FutureOfVeterinaryPublicHealth

Download the Book (PDF): Introduction For most of the twentieth century an eye examination answered one question well and a second question poorly. The first question was what lens would make the patient see clearly today. The second was whether anything was going wrong inside the eye. The first was answered with a phoropter and a letter chart, and the answer was precise to a quarter of a dioptre. The second was answered with an ophthalmoscope, a pair of trained eyes and a memory of what the optic nerve had looked like last time, if anyone had written it down. The two questions shared a room but not a philosophy. Refraction was measurement. Disease detection was impression. That division is dissolving, and the forces dissolving it are the subject of this booklet. Two developments, arriving from different directions over roughly the same twenty years, have changed what an eye examination can be. The first is the recognition that childhood myopia is not a harmless optical inconvenience to be corrected with ever-stronger glasses, but a disorder of eye growth that can be slowed, and that is worth slowing because every millimetre of excess length carries lifelong risk. The second is optical coherence tomography, a technology that lets a clinician measure the thickness of individual retinal layers in micrometres, in seconds, without touching the eye, and compare today's measurement with last year's. These look like separate stories. One belongs to paediatric optometry and contact lens practice; the other to glaucoma and retinal clinics. They are usually taught in different courses, discussed at different conferences and written about in different books. The argument of this booklet is that they are the same story told twice. The controlling idea The central claim is this: modern eye care is moving from correcting what the eye has become to measuring and changing what the eye is becoming, and the practices that serve patients best are those organised around quantified change over time rather than single-visit snapshots. In myopia, the snapshot is the prescription. A child who was minus one dioptre last year and minus one and three-quarters this year gets a new pair of glasses, and the examination is, by the old standard, a success: the child sees 6/6 again. By the new standard it is a failure of attention, because the prescription is a lagging and imprecise proxy for what actually matters, which is the axial length of the eye and how fast it is increasing. Axial length can be measured to a few hundredths of a millimetre with optical biometry. Plotted against age on a growth chart, it tells the clinician whether this child's eye is growing like an emmetropic eye, like a typical myopic eye, or like an eye heading toward the lengths at which myopic maculopathy, retinal detachment and glaucoma become common. And it tells the clinician, within a year or so, whether an intervention is working. In glaucoma and macular degeneration, the snapshot is the fundus appearance on the day. A cup-to-disc ratio estimated by eye varies between examiners and between visits by the same examiner; early ganglion cell loss is invisible to ophthalmoscopy; the earliest structural stages of atrophy in macular degeneration cannot be seen on a colour photograph at all. Optical coherence tomography replaces impression with measurement, and its real power is not in any single scan but in the series. A retinal nerve fibre layer that measures 85 micrometres means little on its own. The same layer measuring 95, then 91, then 85 over four years means a great deal. The two halves therefore share a method. Establish a baseline with a precise instrument. Measure again at intervals. Judge the rate of change against what is expected for that eye, that age and that population. Intervene when the rate is abnormal, and use the same measurement to judge whether the intervention is working. This is how cardiology treats blood pressure and how paediatrics treats height and weight. Eye care is late to it, and is now catching up quickly. Why now Three things make this the moment to take stock. The first is the scale of the myopia problem. A widely cited analysis published in 2016 by Brien Holden and colleagues projected that roughly half the world's population would be myopic by 2050, and nearly one in ten highly myopic. In the cities of East Asia, the future that projection describes has largely arrived: the great majority of young adults leaving secondary school are myopic, and a substantial minority are highly myopic. Other regions are following the same trajectory at different speeds. A generation of children is growing eyes that will, in middle and old age, produce a wave of sight-threatening complications that eye care systems are not built to absorb. The second is that the tools for slowing myopia have matured from promising to proven. Orthokeratology lenses worn overnight, specially designed daytime soft contact lenses, spectacle lenses with arrays of tiny peripheral lenslets, low-dose atropine eye drops and structured time outdoors all now have randomised trial evidence behind them. Regulators have begun to catch up. The United States authorised its first myopia-control soft contact lens in 2019 and its first myopia-control spectacle lens in 2025; the European Union approved its first low-dose atropine product in 2025. For a practitioner, the question is no longer whether myopia control is real but how to deliver it responsibly. The third is that retinal imaging has become both cheaper and smarter. Spectral-domain OCT is now standard equipment in many community optometry practices, not only in hospital clinics. OCT angiography lets clinicians see retinal and choroidal blood flow without dye injection. Artificial intelligence systems can segment retinal layers, flag fluid and grade disease with performance comparable to specialists in controlled studies. The first treatments that slow geographic atrophy, the advanced dry form of macular degeneration, were approved in the United States in 2023, which for the first time gave clinicians a reason to detect and measure atrophy early rather than simply document it. Home OCT devices that patients use daily have begun to receive regulatory authorisation. What this booklet covers, and what it leaves out The first half concerns myopia. It begins with the epidemic itself: how large it is, why it is concentrated where it is, and why the risks of high myopia justify intervention in children who currently see perfectly well in their glasses. It then turns to the biology of eye growth, because every myopia control strategy is an attempt to manipulate the visual signals that regulate how long the eye becomes, and the strategies make sense only once those signals are understood. A chapter follows on orthokeratology, the oldest of the optical approaches and in many ways the most demanding, with its own fitting craft and its own safety obligations. A further chapter covers the rest of the toolkit: soft lenses, spectacle designs, atropine and the more contested newcomers. The last myopia chapter is practical, concerned with how a practice actually runs a myopia management service: measurement, thresholds, communication with families, combination treatment and the unresolved question of when to stop. The second half concerns retinal imaging. It opens with how OCT works and what it can and cannot measure, including the artefacts that mislead the unwary. It then treats glaucoma, where OCT's contribution is the detection of structural loss before the visual field is damaged and the tracking of progression over years. The final chapter treats age-related macular degeneration, where OCT has transformed the monitoring of the wet form and is now reshaping how the dry form is classified, staged and treated. The two halves meet at several points. Highly myopic eyes are harder to image and interpret with OCT, because their stretched retinas and tilted optic discs fall outside the normative databases built from average eyes. Myopia is itself a risk factor for glaucoma. And the same habit of mind, measuring change rather than state, runs through both. Some things are deliberately left out. Refractive surgery, cataract surgery and the treatment of diabetic retinopathy each deserve their own book. Amblyopia and strabismus management appear only where they touch myopia. The detailed pharmacology of glaucoma medications and the surgical treatment of glaucoma are outside scope; the concern here is detection and monitoring. Where the booklet discusses specific products or regulatory decisions, it reflects the position as understood in late 2026, and readers should expect that position to keep moving. A note on evidence This is a field in which commercial interest and genuine scientific progress are tightly interwoven. Most of the pivotal trials of myopia control products were funded by their manufacturers. Many OCT normative databases are proprietary. Effect sizes quoted in marketing materials are often the most favourable subgroup of the most favourable trial. None of this makes the evidence worthless, but it does mean that the reader should attend to what was measured, over how long, in whom, and compared with what. Where the booklet quotes a figure, it names its source, and where the evidence is contested, it says so. Percentage reductions in myopia progression, in particular, are treacherous: a 50 per cent reduction in a population whose myopia progresses quickly is worth far more in dioptres than the same percentage in a population whose myopia progresses slowly. The absolute difference in axial length, in millimetres, is usually the more honest number. The reader this booklet has in mind is an intelligent non-specialist: a parent trying to make sense of options for a myopic child, a student entering optometry or ophthalmology, a general practitioner or paediatrician, a health policy analyst, or a practitioner trained a decade or more ago who wants to understand how the field has moved. Technical terms are explained where they first appear. No prior knowledge of optics beyond the school physics of lenses is assumed. Chapter 1: The Shape of the Epidemic A few decades ago, myopia in East Asia was common but far from universal; school surveys in Taiwan, for example, repeated at intervals since the 1980s, recorded a steady climb in every age group. By the 2010s, surveys of students finishing secondary school in Seoul, Shanghai, Taipei, Hong Kong and Singapore were finding myopia in something like eight or nine out of every ten, with a tenth to a fifth of them highly myopic. The change took roughly two generations. Genes do not change that fast. Something in the way children grow up changed, and their eyes changed with it. This chapter sets out what is known about the size and distribution of the problem, why it matters beyond the inconvenience of glasses, and why the right unit for thinking about it is not the prescription but the length of the eye. What myopia is A myopic eye focuses light from distant objects in front of the retina rather than on it. Near objects, whose light arrives diverging, can still be focused, which is why the condition is also called short-sightedness or near-sightedness. The simplest correction is a diverging (minus) lens, which spreads incoming light just enough to move the focus back onto the retina. The optical error can arise because the cornea or lens is too powerful, but in the overwhelming majority of childhood-onset myopia it arises because the eye is too long. The average adult emmetropic eye, one that needs no correction, is roughly 23 to 24 millimetres from the front of the cornea to the retina. Each additional millimetre of axial length adds, as a rule of thumb, somewhere between two and a half and three dioptres of myopia. A child whose eye has grown to 26 millimetres will typically be moderately to highly myopic; an adult at 28 or 30 millimetres is very highly myopic indeed. This is the most important physical fact in the whole subject, and it changes how everything else should be seen. Myopia is not a defect of focus that happens to need a lens. It is a structural change in the eye: the globe has elongated, mostly at its back, and in doing so has stretched the retina, the choroid beneath it and the sclera that forms the eye's outer wall. The glasses restore clear vision, but they do nothing to reverse the stretching, and it is the stretching that causes disease later in life. Definitions that the field now shares For decades, studies used different cut-offs for myopia and high myopia, which made comparing them difficult. In 2019 the International Myopia Institute, a collaboration of researchers and clinicians formed to produce consensus guidance, published standard definitions. They are worth knowing because they now appear in trials, guidelines and regulatory submissions, as Table 1 summarises. Table 1. Standard definitions of myopia used in research and clinical practice. Term Definition Practical meaning Myopia Spherical equivalent of −0.50 D or more myopic, with accommodation relaxed The threshold at which a child is counted as myopic Low myopia Between −0.50 D and −6.00 D Most myopic children and adults High myopia −6.00 D or more myopic Markedly raised risk of later complications Pre-myopia Refraction between +0.75 D and −0.50 D in a child whose age, baseline refraction and risk factors predict myopia A target for prevention rather than control Pathologic myopia Excessive axial elongation with characteristic structural changes in the sclera, choroid, retina or optic disc The disease stage that threatens sight Source: Flitcroft and colleagues, International Myopia Institute definitions report, Investigative Ophthalmology and Visual Science, 2019. Two details deserve comment. The first is the phrase "accommodation relaxed". Children have a powerful focusing system that can make a slightly long-sighted eye appear emmetropic, or an emmetropic eye appear myopic, during a routine test. Epidemiological studies that do not use cycloplegic drops to paralyse accommodation overestimate myopia in children, sometimes substantially. Clinicians who start a child on myopia control on the strength of a non-cycloplegic refraction risk treating eyes that are not myopic at all. The second is the concept of pre-myopia. Children are born mildly long-sighted, and their refraction drifts toward zero over the first years of life. A six-year-old who is already at plano, rather than at the one dioptre or so of hyperopia typical for that age, has used up the buffer that would normally protect against later myopia. That child is at high risk even though no lens is yet needed, and it is in children like this that prevention, as opposed to control, becomes possible. The size of the problem The most widely quoted projection comes from a meta-analysis led by Brien Holden, published in Ophthalmology in 2016. Pooling 145 studies covering about 2.1 million participants, the authors estimated that in 2000 about 1.4 billion people, 22.9 per cent of the world's population, were myopic, and about 163 million, 2.7 per cent, were highly myopic. Projecting forward on the basis of observed trends, they estimated that by 2050 about 4.8 billion people, 49.8 per cent of the world's population, would be myopic, and about 938 million, 9.8 per cent, highly myopic. Projections of this kind deserve scepticism. They extrapolate trends that may not continue and combine surveys that used different methods. But the direction is not in doubt, and in some places the projection is already conservative. The pattern has three features that matter for policy. The first is geography. The highest prevalences are in urban East and Southeast Asia: China, Taiwan, Hong Kong, Japan, South Korea and Singapore. A 2012 review in The Lancet by Ian Morgan, Kyoko Ohno-Matsui and Seang-Mei Saw described prevalences of 80 to 90 per cent among young adults completing secondary education in these settings, with 10 to 20 per cent highly myopic. Rates in Europe and North America are lower but rising; estimates for young European adults cluster around 40 to 50 per cent. Rates in rural sub-Saharan Africa and parts of South Asia remain comparatively low, though urbanising populations in those regions show the same upward drift. The second is urbanisation and schooling. Within any given country, urban children are more myopic than rural children, and children in intensive education systems are more myopic than those in less intensive ones. The sharpest natural experiments come from populations that are genetically similar but educated differently. Children of South Asian descent in Singapore are far more myopic than children in rural India. Ultra-orthodox Jewish boys in Israel, who spend long hours studying religious texts from early childhood, have been reported to have much higher myopia rates than their sisters or their secular peers. The third is age of onset. Myopia that begins at age six or seven has many more years of progression ahead of it than myopia beginning at twelve, and children who become myopic early are far more likely to end up highly myopic. In high-prevalence regions the age of onset has been falling. This is arguably the single most worrying trend in the data, because final refraction depends heavily on how early the eye starts to over-elongate. Genes, environment, and the interaction between them Myopia runs in families. A child with two myopic parents is several times more likely to become myopic than a child with none. Genome-wide association studies have identified hundreds of genetic variants associated with refractive error, many involved in light signalling within the retina, the structure of the extracellular matrix of the sclera, or the growth factors that pass messages from retina to sclera. Rare single-gene forms of very high myopia exist and are usually evident from infancy. Yet the genetic signal cannot explain the epidemic. Gene frequencies in a population do not change within two generations. What changed was environment, and the most plausible reading of the evidence is that genes shape susceptibility while environment determines how much of that susceptibility is expressed. A child who inherits a genetically sensitive eye and then spends childhood indoors reading is much more likely to become myopic than either factor alone would predict. Two environmental exposures dominate the evidence. The first is near work, meaning reading, writing and, increasingly, screen use at short distances. The association is real but weaker and less consistent than common belief supposes, and the mechanism remains debated. The second is time outdoors, which is consistently and protectively associated with less myopia onset. The protective effect appears to depend on the intensity of outdoor light rather than on distant viewing or physical activity, and it is strongest for preventing the onset of myopia rather than for slowing it once established. Chapter 2 examines both in detail, because they are central to understanding why the optical treatments work. A natural experiment during the COVID-19 pandemic added evidence. When schools closed and children in many countries spent months confined at home, several screening programmes recorded jumps in myopia among young children. A study of school screenings in Feicheng, China, published in JAMA Ophthalmology in 2021, found that the prevalence of myopia among six-year-olds screened in 2020 was several times that recorded in the same age group in previous years. The effect was strongest in the youngest children, whose eyes are thought to be most sensitive to environmental influence. Pandemic-era data are messy, and some of the effect may have reversed when normal life resumed, but the direction was consistent with the idea that indoor confinement promotes myopic growth. Why a few dioptres matter The objection most often raised by parents, and sometimes by clinicians, is simple: myopia is corrected with glasses, glasses are cheap, and plenty of people live happily with strong prescriptions. Why intervene? The answer is that high myopia is not only an optical condition. The elongated eye has a thinned sclera, a stretched and thinned choroid and a retina spread over a larger surface. These structural changes produce a series of complications that become common in middle and later life. • Myopic maculopathy, a group of degenerative changes at the centre of the retina including patchy and diffuse atrophy of the choroid and retina, lacquer cracks (breaks in Bruch's membrane) and, most dangerously, myopic choroidal neovascularisation, in which abnormal vessels grow under the retina and leak or bleed. Myopic maculopathy is now one of the leading causes of irreversible visual impairment in East Asia and a leading cause in several European populations. • Retinal detachment, in which the thinned peripheral retina develops tears through which fluid passes, lifting the retina away from the wall of the eye. Myopes of any degree carry substantially higher lifetime risk than emmetropes, and the risk rises steeply with the degree of myopia. • Glaucoma, where the association with myopia is consistent though more modest in size. Myopic optic discs are structurally different, larger, tilted and sometimes surrounded by zones of atrophy, which also makes glaucoma harder to detect, a point taken up in Chapter 7. • Cataract, particularly the posterior subcapsular and nuclear forms, which occur earlier in myopic eyes. In 2012, Daniel Ian Flitcroft of Dublin compiled published risk estimates and argued that, like blood pressure and cholesterol, myopia has no safe threshold. Even low myopia of one to three dioptres was associated with a measurably higher risk of myopic maculopathy and retinal detachment compared with emmetropia, and the risk climbed steeply with each additional band of myopia. The comparison with cardiovascular risk factors was deliberate: nobody would argue that because a moderately raised blood pressure causes no symptoms it should be left alone. A population analysis from Rotterdam by Willem Tideman and colleagues in Caroline Klaver's group, published in JAMA Ophthalmology in 2016, looked at the problem from the other end. Among older Europeans, the cumulative risk of visual impairment by age 75 rose steeply with axial length and degree of myopia. For those with the longest eyes and the highest myopia, visual impairment in later life was common rather than exceptional, and myopic maculopathy was the main cause. Each dioptre counts If risk rises continuously with myopia, then slowing myopia should reduce risk even if it does not prevent myopia altogether. This is the logic formalised by Mark Bullimore and Noel Brennan in a 2019 paper titled "Myopia control: why each diopter matters". Using published risk data, they estimated that slowing myopia by one dioptre should reduce the lifetime likelihood of myopic maculopathy by roughly 40 per cent, independent of the starting level. The calculation involves assumptions, but its message has shaped clinical thinking: a treatment does not need to stop myopia in its tracks to be worthwhile. Turning a child who would have ended at minus six into one who ends at minus four is a substantial gain in lifelong eye health, even though both still wear glasses. The argument also shifts attention from the refraction to the axial length. Myopia control trials increasingly report both, and some regulators and researchers regard axial length as the more meaningful endpoint. There are good reasons. Axial length is measured objectively with optical biometers precise to around a hundredth of a millimetre, whereas even cycloplegic refraction has measurement noise of a quarter dioptre or so. Axial length is what directly determines the structural risk. And some treatments, orthokeratology most obviously, change the shape of the cornea and so alter refraction without altering the length of the eye, which makes refraction alone useless for judging their effect. Axial length and the growth of the normal eye Eyes grow in childhood even when they are not becoming myopic. A newborn eye is around 16 to 17 millimetres long. It grows rapidly in the first two years, then more slowly. In an emmetropic child, the lengthening that continues through the school years is matched by a flattening of the cornea and a reduction in the power of the crystalline lens, so that the eye stays in focus as it grows. This coordinated process is called emmetropisation, and it is described in more detail in the next chapter. Because normal eyes grow, a single axial length measurement tells the clinician only so much. What matters is where the eye sits relative to others of the same age and sex, and how fast it is growing. Growth charts for axial length, analogous to the height and weight charts familiar from paediatrics, have been published from European cohorts, including one from Rotterdam led by the Klaver group, and from large Chinese populations. They show percentile curves: the median axial length at each age, and the curves for the heaviest-growing and slowest-growing children. An eye that sits on or above a high percentile at age eight is likely to finish long, and one that is crossing percentiles upward is growing abnormally fast. As a general orientation, emmetropic children in the later primary-school years typically lengthen by around a tenth of a millimetre a year or a little more, whereas untreated myopic children commonly lengthen by several tenths of a millimetre a year, with the fastest growth in the youngest. The published charts give the precise figures for particular populations; the principle is that the rate matters as much as the size. This way of seeing the eye, as a growing organ with a trajectory that can be charted and perhaps redirected, is the basis for everything that follows in the first half of this booklet. A public health problem, not only a clinical one Individual clinical treatment cannot, on its own, address an epidemic affecting half of a population. Even in wealthy settings, specialty contact lenses and regular biometry are expensive, and in many parts of the world they are unavailable. The most scalable intervention, increasing children's time outdoors, is a matter of school scheduling and urban design rather than clinical practice. Several governments have acted on this. Taiwan launched a school programme in 2010 encouraging children to spend two hours a day outdoors, often summarised as "Tian-Tian 120". Researchers there have reported that the rising prevalence of reduced visual acuity among schoolchildren reversed after the programme began, although attributing the change to one policy is inevitably uncertain. China's Ministry of Education, together with other agencies, issued a national plan in 2018 to reduce child myopia, setting prevalence targets and requiring schools to limit homework and increase outdoor activity. Singapore has run a national myopia prevention programme since the early 2000s. In most Western countries, by contrast, myopia is still treated as an optical matter. The World Council of Optometry adopted a resolution in 2021 calling for myopia management, including mitigation, measurement and management, to become the standard of care, and professional bodies in several countries have issued guidance. Clinical practice remains uneven. A child in one practice may be offered axial length measurement, a discussion of options and a treatment plan; a child with the same eye two streets away may simply be given stronger glasses each year. The rest of the myopia half of this booklet is concerned with closing that gap: first by understanding why eyes over-elongate, then by examining what can be done about it and how to do it well. Chapter 2: Why Eyes Grow Too Long Every treatment for myopia in use today rests on a single discovery made in laboratories in the 1970s and 1980s: the eye watches itself grow. The retina does not simply receive images; it evaluates their quality and sends signals that speed or slow the growth of the eye behind it. An eye deprived of clear images grows too long. An eye given images that are focused behind the retina grows longer to meet them. An eye given images focused in front of the retina slows down. This chapter traces how that discovery was made, what it implies, and how it connects to the two environmental factors, near work and time outdoors, that dominate the epidemiology. The optical treatments described in the next two chapters are, in essence, attempts to hijack the eye's own growth control system. Emmetropisation A newborn eye is short and, on average, moderately long-sighted. By the time a child starts school, most eyes have become nearly emmetropic. Across a population, the spread of refractive errors narrows sharply during the first years of life: newborns show a broad, roughly bell-shaped distribution of refractions, whereas young children cluster tightly around mild hyperopia. The eye has somehow grown to fit its own optics. This could in principle be a genetically programmed coincidence, with the cornea, lens and axial length each growing to a predetermined size. Two lines of evidence show that it is not. The first comes from children whose early vision is disrupted. Infants born with congenital cataracts, droopy eyelids covering the pupil or corneal scars often develop high myopia in the affected eye. The eye, denied a clear image, does not stop growing; it grows too much. The second line comes from animal experiments that made the phenomenon controllable. What the animals showed In 1977, Torsten Wiesel and Elio Raviola, working at Harvard on visual development in monkeys, reported that suturing the eyelids of infant monkeys closed caused the covered eyes to become myopic through axial elongation. This was an unexpected by-product of research on the visual cortex, and it launched the field of experimental myopia. Similar results followed in tree shrews, chicks, marmosets and guinea pigs. The phenomenon was named form-deprivation myopia: take away pattern vision, and the eye elongates. The next step came from Josh Wallman and colleagues, working with chicks, and from Frank Schaeffel and colleagues, who fitted young animals with spectacle lenses rather than occluders. A minus lens, which pushes the focus of the image behind the retina, made the eye grow longer until the retina caught up with the image; the eye became myopic by just the amount needed to compensate. A plus lens, which pulls the image in front of the retina, slowed growth, and the eye became correspondingly hyperopic. The eye was not merely responding to blur; it could tell the sign of the defocus, whether the image lay in front of or behind the retina, and grow accordingly. Several further findings shaped modern thinking. The response is local. Wallman showed that if a diffuser or lens covers only half of a chick's visual field, only the corresponding half of the eye elongates. The growth signal is generated within the retina and passed to the adjacent sclera, without needing the brain. Cutting the optic nerve in chicks does not abolish lens compensation. The choroid responds quickly. The choroid, the vascular layer between retina and sclera, thickens within hours in response to plus-lens defocus, pushing the retina forward toward the image, and thins in response to minus-lens defocus. In chicks these changes are large; in humans they are measured in tens of micrometres but are consistent. Choroidal thickness changes are now thought to be an early step in the signalling chain from retina to sclera, and choroidal thinning is a feature of myopic eyes. The periphery matters. A series of experiments by Earl Smith and colleagues at the University of Houston, published in the mid-2000s, used infant monkeys to show that optical defocus imposed on the peripheral retina alone could drive the growth of the whole eye, even when central vision was clear. In one striking experiment, the central fovea was ablated with a laser, and the monkeys still emmetropised normally. The fovea, the part of the retina we use for reading and fine detail, was not necessary for growth control. The periphery, with its much larger area, could govern it. Brief interruptions are powerful. Short periods of clear vision, or of plus-lens defocus, during a day otherwise spent wearing a minus lens can substantially reduce the myopia the minus lens would produce. The signal to slow growth appears to carry more weight, hour for hour, than the signal to accelerate it. This asymmetry is encouraging for treatment, since it suggests that a myopic defocus signal need not be present all day to have an effect. The sclera is the final effector. In mammals, including primates, the sclera of an eye developing experimental myopia undergoes active remodelling: the synthesis of collagen and other structural proteins falls, the fibrils become thinner, and the tissue becomes more extensible, so that normal intraocular pressure stretches it. This is not passive thinning but a regulated change in the biology of the tissue, which is why researchers look for signalling molecules, among them retinoic acid, growth factors and hypoxia-related signals from a thinned choroid, that might be targeted by future drugs. It also explains why the elongation, once it has occurred, is permanent. From animals to children: the peripheral defocus idea These findings produced an influential hypothesis about human myopia. A conventional single-vision spectacle lens or contact lens is designed to focus light precisely on the fovea. Because the eye is not a perfect sphere and myopic eyes are more elongated axially than equatorially, images falling on the peripheral retina of a corrected myopic eye tend to be focused behind it. Such relative peripheral hyperopic defocus would, if the animal work applies, act as a signal to grow longer, exactly the wrong message for an eye that is already too long. If that is correct, a lens that corrects the centre but places the peripheral image in front of the retina should slow growth. This is the design principle behind the dual-focus soft contact lens, the peripheral lenslet spectacle lenses, and, as a by-product of the way it reshapes the cornea, orthokeratology. The hypothesis has been challenged. Longitudinal studies in children have not consistently found that the amount of peripheral hyperopia predicts who will become myopic or how fast they will progress. The retinal shape of myopic eyes varies. And some effective treatments work in ways that are hard to explain by peripheral defocus alone. A more flexible version of the idea, which fits the evidence better, is that the retina integrates defocus signals across a large area, and that adding a competing myopic defocus signal, whether in the periphery or superimposed across the field, can counteract the growth drive. Treatments designed on this principle work, whatever the precise mechanism. Near work and the accommodation question Reading and close work have been blamed for myopia since at least the nineteenth century. The epidemiological association is real: children who read more, who attend more intensive schools and who achieve higher educational levels tend to be more myopic. Mendelian randomisation studies, which use genetic variants that predict years of education as a natural experiment, suggest that more education causes more myopia, not simply that myopic people are drawn to study. Why near work would cause myopia is less clear. The leading explanation for many years involved accommodative lag. When a child looks at a near target, the lens should increase in power by an amount that matches the target distance. Many myopic children accommodate slightly less than required, so that the near image falls a little behind the retina, precisely the hyperopic defocus that drives growth in animals. If sustained near work produced hours of such defocus each day, it could plausibly push the eye to elongate. This hypothesis was tested in the Correction of Myopia Evaluation Trial (COMET) in the United States, reported in 2003, which gave children progressive addition spectacle lenses to reduce the demand on accommodation during reading. The lenses slowed progression by only a small amount, around 0.2 dioptres over three years, which was statistically significant but clinically modest. Later trials of bifocals, notably one by Desmond Cheng and colleagues in Canada using executive bifocals, with and without a base-in prism, in Chinese-Canadian children whose myopia had been progressing quickly, showed substantially larger effects. The accommodation story is therefore part of the picture but does not explain it all. The modern view of near work adds several other ideas. Indoor near work usually takes place in dim light, which may itself be myogenic. It displaces time outdoors. And the visual content of text on a page, with its high-contrast, low-spatial-frequency patterns, may produce a particular pattern of retinal stimulation. A theory advanced in recent years proposes that the contrast signal between neighbouring cone photoreceptors helps regulate growth, and that the low-contrast, indoor visual environment is myogenic. This theory underlies a spectacle lens design that deliberately reduces contrast in the periphery, discussed in Chapter 4. It remains a hypothesis under test. Digital screens are a special concern of parents. The evidence that screen use as such, independent of total near work, causes myopia is still limited, though meta-analyses have found associations between smart device use and myopia. The simplest reading is that screens are a large and growing share of near work and a large and growing competitor for outdoor time, so they almost certainly contribute, but that there is nothing uniquely damaging about pixels compared with paper. Light and the outdoors The most robust environmental finding in myopia research is the protective effect of time spent outdoors. Cross-sectional studies in Australia, notably the Sydney Myopia Study led by Kathryn Rose and Ian Morgan in the mid-2000s, found that children who spent more time outdoors were less myopic, regardless of how much near work they did. Children of Chinese ethnicity in Sydney had far lower myopia prevalence than children of similar ancestry in Singapore, and the most striking difference between them was time outdoors. Correlation invited trials, and several were run. In Guangzhou, China, Mingguang He and colleagues randomised 12 schools. In intervention schools, first-grade children received an additional 40-minute outdoor activity class at the end of each school day, and parents were encouraged to take their children outdoors at weekends. After three years, reported in JAMA in 2015, the cumulative incidence of myopia was 30.4 per cent in the intervention group compared with 39.5 per cent in the control group. The difference, about a quarter fewer new cases, was achieved with an intervention that required no equipment and no clinical visits. In Taiwan, Pei-Chang Wu and colleagues ran a programme called Recess Outside Classroom, in which children in one school were encouraged to spend recess outdoors, while children in a nearby school continued as usual. After one year, reported in Ophthalmology in 2013, new myopia had developed in 8.41 per cent of the intervention children compared with 17.65 per cent of the controls. A later cluster-randomised trial by the same group found protective effects from around 11 hours a week outdoors, and suggested that even moderate light levels, such as those found in the shade of trees, provided benefit. In Shanghai, a trial led by Xiangui He and colleagues, published in 2022, randomised schools to an additional 40 or 80 minutes a day outdoors. Both groups had less incident myopia than controls, though the difference between 40 and 80 minutes was smaller than expected, possibly because children in the 80-minute arm did not consistently achieve the full exposure. The trials consistently show the effect is greater on preventing onset than on slowing established myopia. Once a child is myopic, extra time outdoors produces at most a modest slowing of progression. Why does outdoor time protect? Several explanations were considered: distant viewing, which relaxes accommodation; physical activity; vitamin D; and the brightness of outdoor light. The evidence now points strongly to light intensity. Indoor light, even in a well-lit classroom, typically measures a few hundred lux. Outdoor light on a cloudy day is usually several thousand lux and on a sunny day can exceed 50,000 or even 100,000 lux. In chicks, rearing under very bright light slows the development of experimental myopia. The best-studied mediator is dopamine, released by retinal amacrine cells in response to light. Drugs that block dopamine receptors in the retina can abolish the protective effect of bright light in animal models, and drugs that stimulate them can slow experimental myopia. Other candidate mechanisms include the spectral composition of outdoor light, with violet light receiving particular attention in some studies, and the pattern of defocus in outdoor scenes, where most objects are distant and focus uniformly. Several classroom studies have explored bringing the outdoors inside, with glass-walled classrooms or very bright artificial lighting. Results so far are mixed, and the simplest recommendation remains the most practical: children, particularly younger children before myopia begins, benefit from around two hours a day in daylight. Atropine and the chemical pathway The oldest drug treatment for myopia, atropine, fits awkwardly into the optical story. Atropine is a muscarinic antagonist that paralyses the ciliary muscle, stopping accommodation and dilating the pupil. For decades its effect on myopia was attributed to blocking accommodation during near work. But chicks, whose accommodation relies on different muscle types and receptors, respond to atropine anyway, and eyes without effective accommodation still respond. The current view is that atropine acts on receptors in the retina, choroid or sclera, not via accommodation, although the precise site of action remains uncertain. Its effect on pupil size, which increases retinal illumination, may also contribute. Rebound, reversal and the limits of control One further lesson from the laboratory shapes how clinicians talk to parents. Growth, once it has happened, cannot be undone. The stretched sclera does not shrink back. A treatment that slows elongation buys a smaller final eye, but it does not reverse what has already occurred. Several treatments also show some degree of rebound when stopped, with growth briefly accelerating above the untreated rate, a phenomenon discussed in Chapter 5. And the eye's growth control system appears to be most responsive in younger children, which is one reason why treatment begun early achieves more than treatment begun late. A final and still underappreciated finding concerns under-correction. For many years, some practitioners deliberately under-prescribed myopic children, on the theory that the reduced demand on accommodation would slow progression. Randomised trials, including one by Kahchuan Chung and colleagues in Malaysia published in 2002, found the reverse: full correction produced slower progression than under-correction, or no difference. Under-correction is now considered inappropriate as a myopia control strategy. The practice is a useful cautionary example of how plausible theories can mislead when not tested. What the biology implies for practice Taken together, the science suggests several practical principles that the rest of the myopia chapters build on. First, the retina is listening to optical signals that clinicians can control. The lenses a child wears are not neutral. A conventional single-vision lens corrects the blur but does nothing to discourage further growth, and may in some eyes encourage it. Second, a myopic defocus signal added to the visual field, whether in the periphery or across it, is the common element of the most effective optical interventions. Third, light matters, both for prevention and possibly as a component of treatment. Time outdoors is inexpensive, safe and supported by trials. Fourth, younger eyes are more plastic. Earlier intervention achieves more, and prevention in pre-myopic children is the most efficient of all. Fifth, the effect of any treatment should be judged by axial length, because some treatments change refraction without changing length, and because length is what carries the lifelong risk. With those principles established, it is possible to look at each treatment in turn, beginning with the one that most dramatically reshapes the eye: orthokeratology. Chapter 3: Orthokeratology for Children A child puts two rigid contact lenses in before bed, sleeps, removes them in the morning, and sees clearly all day without glasses or contact lenses. To parents encountering it for the first time, orthokeratology sounds like a trick. It is not; it is a well-characterised reshaping of the corneal surface that wears off gradually over a day or two after the lenses are stopped. What makes it relevant to this booklet is a discovery that was not part of its original purpose: children who wear orthokeratology lenses show slower axial elongation than children who wear ordinary glasses. Orthokeratology, commonly abbreviated to ortho-k, is the oldest of the optical myopia control methods and in many ways the most demanding. It requires more fitting skill, more aftercare and more attention to safety than any other option. In return, it offers spectacle-free daytime vision and a myopia control effect that has been replicated across many trials over two decades. This chapter explains how it works, what the evidence shows, how it is fitted, and what can go wrong. A short history The idea that a contact lens could alter the shape of the cornea emerged in the 1960s, when practitioners noticed that patients wearing rigid lenses made of polymethyl methacrylate sometimes had reduced myopia on removal. Deliberate attempts to exploit this, using progressively flatter lenses worn during the day, produced modest and unpredictable results, and early orthokeratology had a mixed reputation. Two developments in the late 1980s and 1990s transformed it. The first was the reverse geometry lens design, in which a relatively flat central zone is surrounded by a steeper reverse curve, followed by an alignment or landing zone that rests on the mid-peripheral cornea. This geometry produces a controlled, centred flattening of the central cornea much more rapidly than the older designs. The second was the development of rigid lens materials with high oxygen permeability, which made overnight wear physiologically acceptable. Wearing the lens during sleep, when the closed eyelid presses gently on the lens, turned out to be the ideal regime: the reshaping occurs overnight and the lens is not needed during the day. In the United States, the Food and Drug Administration approved Paragon Corneal Refractive Therapy lenses for overnight orthokeratology in 2002, for the temporary reduction of myopia of up to about six dioptres with limited astigmatism. Other designs were approved subsequently. Importantly, these approvals were for correcting refractive error, not for slowing myopia. Use of orthokeratology for myopia control in the United States remains an off-label application of approved devices, though one that professional guidelines widely accept. In several Asian markets, particularly China, orthokeratology has become a very large industry, and regulators there have introduced specific controls on who may fit the lenses. How the lens reshapes the cornea The front surface of the cornea is covered by an epithelium about 50 micrometres thick. This layer is soft, regenerates continuously, and can redistribute under pressure. When a reverse geometry lens is worn overnight, the central zone of the lens, which is flatter than the cornea, exerts a gentle positive pressure on the central epithelium, while the tear film under the steeper reverse curve exerts a negative, suction-like pressure on the mid-peripheral epithelium. The result, confirmed by OCT imaging of the cornea, is a thinning of the central epithelium and a thickening of the epithelium in a ring around it. The underlying stroma changes little. Because the power of the cornea depends on the curvature of its front surface, flattening the centre reduces the eye's focusing power, which reduces myopia. A treatment zone of roughly five to six millimetres across is created, within which the cornea is flatter and vision is clear. The ring of thickened epithelium around the treatment zone steepens the mid-peripheral cornea. Light passing through this steeper ring is focused more strongly than light passing through the centre, so it comes to a focus in front of the retina. The retinal periphery therefore receives myopic defocus, while the centre receives a focused image. This is exactly the optical pattern that the peripheral defocus theory, described in the previous chapter, predicts should slow eye growth. The same corneal change also increases the eye's higher-order aberrations, particularly positive spherical aberration, and some researchers believe this contributes to the effect independently. The effect is reversible. If the lenses are not worn, the epithelium gradually returns to its original contour over days to weeks. This is reassuring for safety, since stopping treatment returns the cornea to its prior state. It also means that the child must keep wearing the lenses every night, or nearly every night, to maintain clear daytime vision. The evidence for myopia control The first strong indication that orthokeratology slows myopia came from a pilot study in Hong Kong by Pauline Cho, Sin-Wan Cheung and Marion Edwards, published in 2005. Known as the Longitudinal Orthokeratology Research in Children study, it compared children wearing orthokeratology lenses with a historical control group wearing spectacles, and found that axial elongation over two years was substantially slower in the orthokeratology group. Because refraction cannot be used to judge the effect, since the lenses change refraction directly, orthokeratology studies rely on axial length. This turned out to be a benefit, forcing the field to adopt the more meaningful endpoint. The most influential trial was the Retardation of Myopia in Orthokeratology study, known as ROMIO, a randomised controlled trial by Cho and Cheung published in Investigative Ophthalmology and Visual Science in 2012. Children aged six to ten were randomised to orthokeratology or single-vision spectacles. After two years, axial elongation was about 43 per cent slower in the orthokeratology group. Younger children showed larger absolute effects, consistent with their faster underlying growth. Other trials extended the evidence. A study from the same group fitted toric orthokeratology lenses to children with moderate to high astigmatism and found a similar slowing of elongation. Another fitted children with high myopia with lenses designed to correct only part of their myopia, with the rest corrected by spectacles during the day, and still found substantial slowing. A Spanish trial by Jacinto Santodomingo-Rubido and colleagues found slower elongation in European children. Japanese studies by Takahiro Hiraoka and colleagues followed children for five years and found that the slowing was sustained, though it appeared greatest in the early years of treatment. Meta-analyses pooling these trials generally estimate that orthokeratology reduces axial elongation by around 40 to 50 per cent over two years, which in absolute terms is often of the order of a quarter of a millimetre. Converted into refraction using the usual rule of thumb, that is roughly two-thirds of a dioptre of myopia avoided over two years, and more if the effect persists through the growth years. Who benefits most Not every child responds equally. Several factors have been associated with greater effect. • Younger age at the start of treatment. Younger children progress faster without treatment, so there is more growth to slow. Many studies have found larger absolute slowing in children who start at age six to nine. • Larger pupil size. A larger pupil lets more light through the steeper mid-peripheral cornea, increasing the myopic defocus signal. Some studies have found that children with larger pupils respond better. • Smaller treatment zone. Several recent studies have explored lens designs with a smaller back optic zone, which produces a smaller central treatment zone and brings the ring of myopic defocus closer to the centre of the visual field. These designs have shown greater slowing than conventional designs in some trials, at the cost of more visual disturbance at night. • Higher baseline myopia. Children with higher initial myopia tend to show larger effects, possibly because the corneal reshaping and the resulting defocus ring are more pronounced. These associations help explain the variation between studies and suggest how fitting can be tailored, but none is strong enough to predict with certainty which child will respond. Fitting in practice Orthokeratology fitting is a craft. The fitter must choose a lens whose central zone produces the right amount of flattening, whose reverse curve creates a well-formed ring, and whose landing zone sits evenly on the mid-peripheral cornea so that the lens centres properly. A lens that is too flat will decentre, usually upwards, producing a treatment zone that sits off the pupil and causes blur and ghosting. A lens that is too steep produces a small, poorly defined treatment zone with a central area that has not flattened, known as a central island. The process usually runs as follows. Assessment. The fitter measures refraction, including under cycloplegia for children, and uses corneal topography to map the shape of the cornea across its surface. The topography identifies the central curvature, the eccentricity or rate of flattening toward the periphery, and any astigmatism. Candidates with myopia up to around four or five dioptres and low corneal astigmatism are most straightforward; higher myopia and substantial corneal toricity can be managed with specialised designs, but the task is harder. The fitter also examines the eyelids, tear film and cornea for any condition that would make overnight lens wear unsafe. Lens selection. Lenses can be chosen empirically, calculated by software from the topography and refraction, or by trial fitting with diagnostic lenses. Many modern systems combine the two, calculating a first lens and then refining it. Evaluating the fit. With the lens on the eye, the fitter instils fluorescein dye and examines the lens under blue light. A well-fitting reverse geometry lens produces a characteristic pattern often described as a bull's eye: a dark central area of close contact, a bright ring of pooled tears under the reverse curve, a dark ring of contact at the landing zone and a thin bright band at the edge. Deviations from this pattern indicate the adjustments needed. Overnight trial and topography. The decisive test is what happens after a night of wear. The child returns in the morning, still wearing the lenses or having just removed them, and the fitter measures vision, refraction and topography. A difference map, subtracting the pre-treatment topography from the post-treatment topography, shows the treatment zone. A well-centred bull's eye pattern of central flattening surrounded by a ring of steepening indicates success. A treatment zone displaced up or to one side, or a central island, signals the need for a different lens. Stabilisation. Full effect typically takes one to two weeks of nightly wear. During that period, the child may need temporary spectacles or soft lenses for residual myopia. Once stable, most children see well throughout the day, though some notice a slight return of myopia by evening, particularly with higher prescriptions. The visual side effects are mostly minor and tend to diminish with time: haloes and glare around lights at night, reduced contrast in dim light, and occasionally ghosting if the treatment zone is decentred. Children generally tolerate them well. Adults with large pupils are more troubled by them. Safety: the question that matters most Any contact lens that is worn overnight carries a risk of microbial keratitis, a corneal infection that can scar the cornea and, in the worst cases, cause permanent visual loss. This is the central safety concern with orthokeratology, and it deserves a direct discussion with every family considering it. In the early 2000s, a cluster of serious corneal infections associated with orthokeratology was reported, mainly from China and Taiwan. Many involved Pseudomonas aeruginosa and a notable number involved Acanthamoeba, a free-living amoeba found in tap water and soil that causes particularly severe and hard-to-treat infections. Reviews of these cases, including one by Kathleen Watt and Helen Swarbrick, found that many were associated with poor practices: unregulated fitting, poor hygiene, rinsing lenses or cases in tap water, and inadequate follow-up. The episode prompted regulatory tightening in China and elsewhere. The best estimate of the risk in properly supervised practice comes from a study by Mark Bullimore and colleagues, published in Ophthalmology in 2013. It surveyed a random sample of American practitioners fitting orthokeratology and identified cases of microbial keratitis over a defined period of wear. The estimated incidence in children was about 7.7 cases per 10,000 patient-years of wear. The confidence interval was wide, because the number of cases was small. This rate is of a similar order to the risk associated with overnight wear of soft contact lenses in adults, and higher than for daily disposable soft lens wear. It translates, for a child wearing orthokeratology for eight years, into a cumulative risk of the order of a few in a thousand of experiencing a corneal infection, most of which would resolve with treatment but a minority of which could cause lasting damage. How families weigh that risk against the benefits is a genuinely individual judgement. The risk needs to be set alongside the lifetime risk of complications of higher myopia that the treatment aims to reduce, which is also measured in percentages, and against the alternatives, some of which carry no infection risk at all. The clinician's role is to present the numbers honestly and ensure that, if orthokeratology is chosen, it is done safely. Safe practice rests on a few non-negotiable rules. • No tap water. Lenses and lens cases must never be rinsed with tap water, and the child should not swim or shower while wearing the lenses. This is the most important single rule for preventing Acanthamoeba infection. • Proper disinfection. Hydrogen peroxide systems are widely recommended for rigid lenses. The lens case should be cleaned, air-dried and replaced regularly. • Hand washing before handling lenses. • Parental supervision. Younger children cannot be expected to manage lens care alone. • Regular lens replacement. Rigid orthokeratology lenses typically need replacing about once a year, because deposits and surface wear accumulate. • Clear instructions for when to stop. A red, painful or light-sensitive eye, or reduced vision, means lenses out and an urgent visit to the practitioner. • Scheduled aftercare. Review the morning after the first night, then at about one week, one month and three months, and every six months thereafter, at minimum. Stopping, rebound and long-term use A question every family asks is how long the child must continue. Myopia typically progresses most rapidly in the younger school years and slows through the teenage years, though for many it continues into the early twenties. The usual advice is to continue at least until the mid-teens and to base decisions on measured axial growth rather than age alone. Rebound is a concern. A follow-up to the ROMIO trial by Cho and Cheung found that children who stopped orthokeratology experienced a period of faster axial elongation compared with those who continued, and that restarting slowed growth again. The rebound did not erase the benefit already gained, but it did reduce it. The general principle is to stop only once growth has slowed to near-normal rates, and to monitor axial length after stopping. Many children continue with orthokeratology into adulthood for convenience, since it allows them to go without glasses or daytime lenses. Adults who wish to stop can do so, and their corneal shape returns to its pre-treatment state. Those who later wish to have refractive surgery need a period off lenses to allow the cornea to stabilise before measurement. Where orthokeratology fits Orthokeratology suits motivated families who can manage nightly lens care, children who would benefit from freedom from glasses during the day, for example for sports, and children whose myopia and corneal shape fall within the range where fitting is straightforward. It is particularly attractive for younger children with fast progression, where its effect appears strongest. It is less suitable where hygiene cannot be assured, where access to regular aftercare is limited, or where the family is uncomfortable with the infection risk. Orthokeratology can also be combined with low-dose atropine eye drops, and several trials, from Japan, Hong Kong and China, have found that the combination produces more slowing than orthokeratology alone, particularly in children with lower myopia. The rationale is that the two treatments act through different pathways: one optical, one pharmacological. The combination is discussed further in Chapter 5. Finally, orthokeratology provides a salutary lesson for all myopia control. Its effect could not have been assessed by refraction, because the lens changes the refraction directly. It forced practitioners to measure axial length. That habit, once acquired, turned out to be the right way to monitor every myopia control treatment, and it is a habit that the rest of this booklet will return to repeatedly. Hashtags: #VisionScienceAndOptometry #MyopiaControl #RetinalImaging #ChildhoodMyopia #AxialLength #AxialElongation #OpticalBiometry #MyopiaManagement #MyopiaPrevention #Orthokeratology #PeripheralDefocus #MyopicDefocus #LowDoseAtropine #TimeOutdoors #Emmetropisation #HighMyopia #PathologicMyopia #MyopicMaculopathy #OpticalCoherenceTomography #OCTImaging #RetinalLayerAnalysis #GlaucomaDetection #MacularDegeneration #LongitudinalEyeCare #FutureOfVisionScience

Download the Book (PDF): Introduction Near the middle of Orson Welles's Citizen Kane (1941) there is a scene in a snowbound Colorado boarding house. A mother signs a document that hands her young son over to a banker, who will raise him and manage the fortune that has fallen into the family's lap. Most viewers remember the scene as sad. Fewer remember why it is sad in the particular way it is, and almost nobody on first viewing notices that the boy is visible throughout, through a window at the back of the room, playing in the snow and shouting, while the adults decide his life in the foreground. The camera retreats as the mother walks from the window to the table; it holds the father, the banker and the mother in one composition; and the child remains framed in the distant window like a picture on the wall. There is no cut to a close-up of the boy's face at the moment of decision. There does not need to be. The arrangement of bodies in depth has already said what a close-up would have said, and said something a close-up could not: that the decision is being made over him, at a distance from him, while he is absorbed in play he is about to lose. That observation is the kind this book teaches you to make and to defend. It is not an impression, and it is not a matter of taste. It is a claim about how a specific configuration of visible elements, chosen from among many possible configurations, shapes what a viewer is led to understand and feel. It can be checked against the film. It can be argued with. It can be written up in a paragraph that another scholar can test by watching the same forty seconds. This is what a formal close reading of a film is, and it is the cinematic equivalent of what literary scholars have done with lines of verse for a century. Why film needs its own close reading Humanities scholars who come to film from literature, history, philosophy or art history usually arrive well equipped to talk about what a film is about. They can discuss its themes, its historical moment, its politics, the director's biography, the source novel. What they often lack is a disciplined vocabulary for talking about how a film means what it means at the level of the shot and the cut. The result is a familiar kind of writing in which a film is treated as if it were a novel with pictures: plot is summarized, dialogue is quoted, and the images are mentioned only when they happen to illustrate a point already made on other grounds. The problem with that approach is not that it is wrong about themes. It is that it discards most of the evidence. A film is made of hundreds or thousands of decisions about where to put the camera, how high, with which lens, how the actors stand in relation to one another and to the edges of the frame, where the light comes from and how hard it is, how long each shot lasts and what it is joined to. Those decisions are where a filmmaker's thinking is most visible and least paraphrasable. A scholar who ignores them is like a critic of poetry who reads only the prose summary. The opposite failure is just as common and harder to correct. Students who have been taught a little film vocabulary often apply it as a code book: a low angle means power, a high angle means weakness, a red dress means passion, a diagonal composition means instability. The trouble with a code book is that it is nearly always wrong in the particular case. Welles shoots Kane from low angles when he is triumphant and when he is ruined; the same angle does opposite work depending on what surrounds it. A close reading that treats film elements as a fixed dictionary produces claims that are both easy to make and easy to refute, and it trains the reader to stop looking at precisely the moment when looking should begin. The controlling idea This book rests on one idea, and every chapter applies it to a different layer of the film image. The idea is this: every visible feature of a shot, and every join between shots, is a choice made against alternatives, and its meaning lies in the difference between what was chosen and what could have been chosen in its place. A low angle means nothing by itself. It means something against the eye-level angles the film has trained you to expect, against the high angle used on the same character two scenes earlier, against the flat lighting of the preceding sequence. Meaning in cinema, as in language, is relational. The practical consequence of this idea is a method. When you look at a shot, you ask what else could have been done at this point, given this film's own habits and the options available to filmmakers at the time, and what difference the actual choice makes. Semioticians call this the commutation test: mentally substitute one element for another and see whether the meaning changes. If replacing a wide shot with a close-up would change what the viewer infers, then the wide shot is doing semiotic work, and you can describe that work precisely. If the substitution would change nothing, the element is not carrying meaning at that moment and you should not claim that it is. The test protects you from both failures described above. It forces attention to the image, and it disciplines interpretation by making every claim answer the question: compared to what? How the book is organized The first chapter sets out the conceptual tools: sign, code, paradigm and syntagm, denotation and connotation, and the commutation test itself, drawn from the semiotic tradition of Ferdinand de Saussure, Charles Sanders Peirce, Roland Barthes and Christian Metz, but stripped to what a working critic needs. The chapters that follow move through the layers of the film image in roughly the order a filmmaker builds them. Chapter 2 treats blocking, the placement and movement of bodies in the space of the set, and shows how staging in depth, lateral arrangement and the edges of the frame encode relations of power, intimacy and exclusion. Chapter 3 treats lighting, from the three-point system of the Hollywood studios through film noir and naturalistic lighting, and argues that light is always an argument about what deserves to be seen, including an argument about whose faces the technology was built to render. Chapter 4 treats the camera's position: its height, its angle, its distance and the lens it uses, and the way these govern the viewer's relation to what is shown. The next three chapters move from the single shot to the relation between shots. Chapter 5 examines the continuity system, the set of conventions by which classical cinema makes cuts invisible, and asks what that invisibility conceals. Chapter 6 turns to Soviet montage, the tradition that made the cut visible and treated the collision of shots as a way of producing ideas. Chapter 7 treats duration and pacing: how long shots last, how the rhythm of editing has changed over a century, and what the long take does that editing cannot. Chapter 8 widens the lens to visual rhetoric, the ways that all these elements combine to persuade, and the ways film form has served propaganda, ideology and resistance. Chapter 9 brings everything together into a working protocol for close reading, with a full worked example. The conclusion asks what this kind of reading is for, where its limits lie, and what remains unsettled. What you need to bring You need no technical knowledge of filmmaking, though a little helps and the book supplies it where needed. You do need access to the films discussed and the patience to watch scenes more than once, with the ability to pause. Close reading of film was nearly impossible for most scholars before home video; the pause button made it a practice rather than a feat of memory. Use it. Most of the scenes discussed here are short and widely available, and the arguments will be more convincing, and more contestable, if you check them against the images themselves. The films discussed are drawn mostly from the canon that film scholarship has built up over the past century: Hollywood studio films, Soviet silent cinema, European art cinema, and a range of films from Japan, Hong Kong, Iran and elsewhere. This is a practical choice, not a claim that these films matter more than others. They are the films on which the analytic vocabulary was developed, and they are the films a reader is most likely to be able to find. The method itself applies to any moving image, from a television advertisement to a vertical video on a phone, and the conclusion returns to that point. One more preliminary. Film analysis is sometimes accused of reading too much into images, of finding intentions the filmmakers never had. The method taught here sidesteps that worry without dismissing it. A close reading does not need to prove that a director intended a particular effect. It needs to show that the effect is there, produced by specific and describable choices, available to any attentive viewer. Whether Welles consciously meant the window to frame the boy like a picture is a question for biography. That the window does frame him that way, and that this arrangement differs meaningfully from the alternatives, is a question for analysis, and it is the question this book equips you to answer. Chapter 1. The Frame as a Sign Semiotics is the study of signs: of how anything, a word, a gesture, a traffic light, a photograph, comes to stand for something other than itself. The field has two founding figures who worked independently and never met. Ferdinand de Saussure, a Swiss linguist whose lectures were published by his students as the Course in General Linguistics in 1916, treated language as a system of differences. Charles Sanders Peirce, an American logician writing in the late nineteenth and early twentieth centuries, built a general theory of signs that included pictures, symptoms and diagrams as well as words. Film scholars have drawn on both, often uneasily, because the moving image fits neither model cleanly. This chapter takes from each what a working critic needs and leaves the rest. Three kinds of sign, and why film uses all of them Peirce's most useful distinction for film is between three ways a sign can relate to what it stands for. An icon signifies by resemblance: a portrait resembles its sitter, a map resembles its territory in certain respects. An index signifies by physical connection or causal contiguity: smoke indicates fire, a footprint indicates a foot, a weathervane indicates the direction of the wind. A symbol signifies by convention alone: the word "tree" has no resemblance to or physical connection with trees, and signifies only because speakers of English agree that it does. The photographic image is unusual because it is both icon and index at once. A photograph of a face resembles the face, which makes it iconic, and it was also produced by light reflected from that face striking a sensitive surface, which makes it indexical. The French critic André Bazin built much of his film theory on this second property. In his essay "The Ontology of the Photographic Image," first published in 1945 and collected in What Is Cinema?, he argued that the photograph carries a trace of the real object in a way no painting can, and that cinema's vocation is to respect that trace. Digital cinema has complicated the indexical claim, since an image can now be synthesized without any object in front of a lens, but the viewer's habit of treating a moving photographic image as a record of something that was there remains powerful, and filmmakers continue to exploit it. Film is also full of symbols in Peirce's sense. A wedding ring, a national flag, a crucifix on a wall, a clock showing midnight: these objects mean what they mean by cultural convention, and a film can deploy them just as a novel deploys words. And film develops its own conventional signs over time. The fade to black that signals the passage of time, the dissolve that signals a memory, the slow push-in toward a face that signals a character's dawning realization: none of these resemble or are caused by what they signify. They are conventions, learned by audiences through repetition, and they can be broken or revised. The practical lesson is that you cannot read a film image with a single model of the sign. The same frame may present an iconic likeness of an actor, an indexical record of a particular afternoon's light on a particular location, and a symbolic prop, all at once, and a close reading must be able to move among these registers. When a critic says a shot of a crumbling wall "represents" the decay of an aristocratic family, the claim is about symbolic or metaphorical meaning built on top of an iconic image. When a critic says that the wobble of a handheld camera "conveys" urgency, the claim is about an indexical trace of the operator's body read as a sign of a state of affairs. Being clear about which register you are working in makes your claims sharper and easier to test. Paradigm and syntagm Saussure's great insight was that a sign's meaning does not come from its relation to a thing in the world but from its relations to other signs in the same system. The word "cat" means what it means partly because it is not "bat," "cap," "cot" or "dog." Saussure distinguished two axes along which these relations run. The syntagmatic axis is the axis of combination: the sequence in which signs are actually arranged, as words follow one another in a sentence. The paradigmatic axis is the axis of selection: the set of alternatives that could have occupied any given slot but did not. In the sentence "the cat sat on the mat," the syntagm is the actual chain of words; the paradigm for the slot occupied by "sat" includes "lay," "slept," "stood," "waited," and so on. This distinction is the single most useful piece of theory a film critic can own, because it maps directly onto how films are made. At every moment, a filmmaker selects from a paradigm of options and combines the selection with what came before and what comes after. For any given shot, the paradigms include camera distance (extreme long shot, long shot, medium shot, close-up, extreme close-up), angle (high, eye-level, low, overhead, canted), lens (wide, normal, long), lighting key (high or low), color palette, the position of each actor in the frame, the direction each is facing, whether the camera moves and how, and how long the shot lasts. The syntagm is the sequence of shots as it actually unfolds in the finished film. A close reading works on both axes. On the paradigmatic axis, it asks: what was chosen here, and from among what alternatives? On the syntagmatic axis, it asks: how does this choice relate to the ones before and after it? The two questions are inseparable. A close-up means something different after five minutes of long shots than it does in a sequence made entirely of close-ups. Carl Theodor Dreyer's The Passion of Joan of Arc (1928) is built almost entirely from close-ups of faces, many against blank white walls. In that film a close-up is the norm, and what stands out, what becomes marked, are the rare shots that pull back to show bodies in space. In a Western shot mostly in long shot, like John Ford's The Searchers (1956), a close-up of John Wayne's face is an event. The commutation test The method that follows from Saussure's insight was formalized by linguists as the commutation test, and it transfers directly to film. To determine whether an element is meaningful, substitute an alternative from the same paradigm and ask whether the meaning of the whole changes. In phonology, the test establishes which sound differences matter in a language: swap the "p" in "pat" for "b" and you get a different word, so the difference between p and b is meaningful in English. In film, the test establishes which visual choices are doing work at a given moment. Consider a simple case. A woman receives a telephone call telling her that her husband has died. The filmmaker must decide, among many other things, how close the camera is to her. Try the commutation mentally. In a close-up, the viewer scrutinizes her face for its reaction, and the film invites empathetic identification with her inner state. In a long shot, she is small in the frame, perhaps surrounded by the furniture of a domestic interior, and the film invites the viewer to see her situation rather than her feeling: her isolation, the ordinary room that will now be different. If the camera is outside the house looking in through a window, the viewer is placed as an observer, perhaps an intruder, and the film withholds her reaction entirely. Each of these is a legitimate choice. They produce different meanings. The commutation test makes those meanings describable by contrast. Notice what the test does not do. It does not assign a fixed meaning to "close-up" or "long shot." It asks what difference the choice makes in this context. That is why it protects against the code-book fallacy. A critic who says "the long shot creates distance" has said something almost empty, because a long shot does not always create distance; in a comedy by Jacques Tati, the long shot is the space in which jokes happen, and it produces engagement rather than detachment. A critic who says "by holding the camera in long shot when we most expect a close-up, the film refuses us access to her grief and makes us register the room instead" has made a claim that depends on a commutation, and it can be checked. The test also teaches you where to look. Elements that are fully determined by genre convention, technical necessity or narrative requirement are weakly meaningful, because there was no real alternative. If every shot in a film is lit in high key, the high-key lighting of a particular shot tells you little. Elements that deviate from the film's own norms, or from the norms of its period and genre, are strongly meaningful. The Russian Formalist critics of the 1910s and 1920s, whose ideas film scholars such as Kristin Thompson later adapted, called this the principle of foregrounding: an artwork draws attention to its devices by departing from the background of expectation. Much of the skill of close reading consists in knowing what the background is, so that you can see what stands out from it. Denotation, connotation and the problem of the unit Roland Barthes, the French critic whose essays of the 1950s and 1960s did more than anyone's to carry semiotics beyond linguistics, drew a distinction that film critics use constantly. Denotation is the literal, first-order meaning of a sign: a photograph of a man in a uniform denotes a man in a uniform. Connotation is the second-order meaning that attaches to the denoted content by cultural association: the uniform may connote authority, militarism, national pride or menace, depending on context. In "Rhetoric of the Image," a 1964 essay analyzing a magazine advertisement for Panzani pasta, Barthes showed how a photograph of tomatoes, peppers, a string bag and packaged pasta connoted "Italianicity" through its colors and objects, and how the caption anchored that connotation and closed down other possible readings. His term for this closing-down function was anchorage. Film images work the same way, with the added complication that everything moves and changes. Dialogue, music and sound effects anchor images much as captions do, and the sequence of shots anchors each shot by setting up expectations. A shot of an empty chair denotes an empty chair. After a funeral scene, it connotes absence and loss. After a scene in which a character has stormed out of a meeting, it connotes rupture or defiance. The image is the same; the syntagm changes its connotation. Christian Metz, the most rigorous of the early film semioticians, spent much of his career on a problem that follows from this. In Film Language, a collection of essays published in French in 1968 and translated into English in 1974, Metz asked whether cinema is a language in Saussure's sense. His answer was that it is not, or not quite. Language has minimal units (phonemes, which carry no meaning on their own, and morphemes, which do) that combine according to rules. Film has no equivalent. A shot is not like a word; it is more like a whole sentence or a statement, since even the simplest shot of a revolver says something like "here is a revolver" rather than merely naming one. And there is no finite vocabulary of shots from which filmmakers select, as speakers select from a lexicon. Every shot is, in a sense, a new invention. Metz concluded that cinema is a language without a language system: it has codes and conventions but no grammar in the strict linguistic sense. Metz's conclusion matters for the practicing critic in two ways. First, it warns against pushing linguistic analogies too far. There is no dictionary of shots, and a critic who writes as if there were will be wrong. Second, it points to where the codes actually are. Metz's most famous contribution, the grande syntagmatique, was a classification of the ways shots combine into larger units in classical narrative films: the scene, the sequence, the alternating syntagm that cuts between two lines of action, and so on. What Metz found was that the conventions of cinema live less in individual images than in patterns of combination. That insight anticipates the argument of the middle chapters of this book. What the frame does before anything is put in it One element of the image is so basic that it is easy to overlook: the frame itself. Unlike a painting, which has a fixed edge that closes off a composition, a film frame cuts out a portion of a continuous space that seems to extend beyond it. Bazin put the contrast sharply: the frame of a painting is centripetal, drawing the eye inward to the composition, while the edge of a film screen is centrifugal, a mask that shows only part of a world that continues on every side. The distinction is not absolute, and some filmmakers compose with painterly closure, but it identifies something real. What lies just outside the frame, what film theorists call offscreen space, is always potentially present. Noël Burch, in Theory of Film Practice (published in French in 1969 and in English in 1973), distinguished six zones of offscreen space: the four spaces beyond each edge of the frame, the space behind the camera, and the space behind the set or beyond the horizon. A filmmaker can activate any of these. A character looks off to the left, and the viewer infers someone or something there. A hand enters from the bottom of the frame. A voice speaks from behind the camera. Horror films depend heavily on offscreen space, because what cannot be seen is more frightening than what can. Yasujirō Ozu frequently cuts to a shot of an empty room just after the characters have left it, and the offscreen space they have entered, which the viewer never sees, becomes a presence in the shot. The shape of the frame is itself a choice. The aspect ratio, the ratio of width to height, has varied through film history: the nearly square frame of early sound cinema, standardized in the early 1930s at 1.37:1 and known as the Academy ratio; the widescreen processes introduced in the 1950s, such as CinemaScope, at roughly 2.35:1; the 1.85:1 ratio common in American theatrical releases since; and the 16:9 of modern television. Each ratio favors certain compositions. The Academy ratio suits the vertical human figure and the close-up of a face. Widescreen suits landscapes and horizontal groupings of figures, and it tends to hold more than one person in a close shot. When Wes Anderson's The Grand Budapest Hotel (2014) switches aspect ratios to mark different time periods, the frame shape becomes an explicit sign. More often its work is quieter, a constraint within which every other choice is made, and a close reading should note it. The chapters that follow take the contents of the frame one layer at a time. But the tools introduced here remain constant throughout: sort the image into its registers of sign, identify the paradigm from which each element was selected, place it on the syntagm of the film's unfolding, and test its meaning by imagining the alternatives. The discipline of asking "compared to what?" is the whole method in miniature. Chapter 2. Blocking and the Geography of Power Mise-en-scène is a term borrowed from the French theater, where it meant "putting on stage": the arrangement of everything that appears before the audience. Film critics adopted it to name everything that appears before the camera: setting, costume, makeup, props, lighting and the placement and movement of actors. Some critics include the camera's framing under the term; others reserve it for what is staged in front of the lens. This chapter concentrates on the part of mise-en-scène that most directly shapes relations between people on screen, and that is most often ignored by viewers because it seems natural: blocking, the positioning and movement of actors in the space of the scene. Blocking is where drama becomes geometry. Who stands closer to the camera and who stands farther away, who occupies the center of the frame and who is pushed to its margins, who sits and who stands, who moves and who stays still, who faces the camera and who turns away: these are choices, each drawn from a paradigm of alternatives, and together they produce a map of the relations among characters at a given moment. That map is continually redrawn as the actors move. A well-staged scene is a sequence of such maps, and reading it closely means tracking how the geography of bodies changes and what those changes express. Depth, lateral space and the hierarchy of the frame The film frame offers two main axes for arranging figures: the lateral axis, running from left to right across the screen, and the depth axis, running from the foreground near the camera to the background far from it. Each produces its own kinds of relation. Lateral arrangement is the older and simpler system, inherited from the proscenium stage and from early cinema's habit of placing the camera at a distance and letting actors perform across the width of a set. When figures are arranged side by side at roughly the same distance from the camera, they tend to read as equals, or at least as participants in the same space. Separation along the lateral axis reads as social or emotional distance, and movement across it reads as approach or withdrawal. A classic device is to place two characters at opposite edges of the frame with a gap of empty space between them, a composition that speaks of estrangement without a word of dialogue. The final shot of Michelangelo Antonioni's La Notte (1961), in which a married couple lie tangled together on a golf course at dawn, draws its force partly from the preceding scenes, in which the two have been kept apart across the frame, often separated by architectural divisions. Depth staging is more complex and more powerful. When one figure stands in the foreground and another in the background, the difference in their size on screen creates an immediate hierarchy of visual prominence. The foreground figure dominates, simply by occupying more of the frame, yet the background figure may be the focus of attention if lit more brightly, if placed in the center, if moving while the foreground figure is still, or if the foreground figure is turned away from the camera. Depth staging allows a filmmaker to set up competition for the viewer's attention within a single shot and to shift that attention without cutting. The film historian David Bordwell has done more than anyone to show how inventive early filmmakers were in this regard. In On the History of Film Style (1997) and Figures Traced in Light (2005), he analyzed European films of the 1910s, especially by the French director Louis Feuillade and the Danish and Swedish filmmakers of the period, in which characters move toward and away from a fixed camera in carefully choreographed patterns. A figure steps forward to reveal a figure behind; another crosses in the foreground and briefly blocks the view; a third enters through a doorway in the background. These films, made before continuity editing became the dominant way of directing attention, used blocking to do the work that editing later took over. Bordwell's point is that staging is not a primitive substitute for editing but a different and sophisticated system with its own resources. Deep focus and the democratic image The Welles scene described in the introduction belongs to a later and more famous moment in the history of depth staging. When Citizen Kane was released in 1941, its cinematographer Gregg Toland had developed, with fast film stock, powerful lighting and wide-angle lenses stopped down to small apertures, a technique that kept planes from the near foreground to the far background in sharp focus at once. This is called deep focus. Deep focus was not new in principle, but Toland and Welles used it with a boldness that seemed revolutionary, and they combined it with staging that pushed figures into extreme foreground and background positions. Return to the boarding house scene. Mrs. Kane stands in the foreground, her face large and rigid. Thatcher, the banker, sits near her at the table. Mr. Kane, the father, hovers in the middle ground, protesting weakly. Through the window at the back, small and sharp, the boy Charles plays in the snow. Every element is in focus. The viewer is free to look at any of them. But the composition is not neutral. The mother's dominance of the foreground matches her dominance of the decision; the father's position in the middle, standing and ineffectual, matches his marginality to it; the child's position at the far end of the depth axis, framed by the window, matches his exclusion from the decision about his own life. When the mother signs, the boy's voice can be heard crying out in his game. The juxtaposition of adult transaction and childish play in one image, rather than in two alternating shots, forces the viewer to hold both at once. Bazin was the most eloquent defender of this style. In "The Evolution of the Language of Cinema," an essay assembled from articles written between 1950 and 1955 and published in the first volume of What Is Cinema?, he argued that deep focus restores to the viewer a freedom that editing takes away. When a scene is broken into close-ups and reverse shots, the editor decides at each moment what the viewer should look at. When the scene is held in a single deep-focus shot, the viewer must choose, as in real life, and the ambiguity of reality is preserved. Bazin saw this as both more realistic and more democratic, and he found the same virtue in the films of William Wyler, for whom Toland also shot, and in Jean Renoir's The Rules of the Game (1939), where parallel actions unfold in different planes of the frame. Bazin's argument is powerful, but a close reader should notice its limits. The Kane scene does not simply leave the viewer free. It directs attention through composition, lighting and sound as firmly as editing would, and the boy's place in the window is as calculated as any close-up. What deep-focus staging does is not to abolish direction but to change its mode: from successive direction, in which each shot tells you where to look, to simultaneous direction, in which the whole composition arranges a hierarchy and invites the eye to move within it. That difference is real and worth describing, but it is not the difference between freedom and control. Centers, edges and the space outside The center of the frame is its point of greatest stability and, by default, its point of greatest importance. Figures placed in the center tend to read as the subject of the shot; figures pushed to the edges read as peripheral, threatened, or on their way out. Classical Hollywood cinema generally centered its principal figures, adjusted framing to keep them centered as they moved, and used the edges for secondary characters and for entrances and exits. Departures from this norm are therefore meaningful. Some filmmakers make those departures systematic. The Hong Kong director Wong Kar-wai and his cinematographer Christopher Doyle, in In the Mood for Love (2000), repeatedly frame the two principal characters, neighbors whose spouses are having an affair with each other, through doorways, window frames, half-open curtains and narrow corridors, often placing them off-center or cutting off parts of their bodies with the edge of the frame. The result is an image of intimacy that is always partial, obstructed and watched. The characters are never quite given the full frame, just as they never quite permit themselves the relationship they want. A critic who says "the framing expresses repression" has said something vague. A critic who notes that in scene after scene the couple are placed off-center and partly masked by architectural elements, so that the viewer, like the gossiping neighbors in their building, sees them only in fragments, has made a claim that the commutation test can support: imagine the same scenes shot in centered medium two-shots, and the sense of surveillance and constraint disappears. At the other extreme, symmetrical centered compositions can produce effects that have nothing to do with naturalism. Stanley Kubrick favored one-point perspective, with the vanishing point of converging lines placed at the center of the frame and the principal figure placed on it. In The Shining (1980) the device, used in the hotel's corridors, gives the space an imposing, rigid and ultimately menacing order, as if the building itself were organizing the characters. Wes Anderson uses centered symmetry to produce the opposite effect, a doll's-house artificiality that marks his films as fables. The same formal choice, central placement in symmetrical frames, serves dread in one case and whimsy in the other. What decides the difference is the surrounding syntagm: lighting, color, performance, music, the pace of cutting and the genre expectations that each film sets up. Blocking also uses the relation between the space inside the frame and the space outside. A character who is pushed toward the edge of the frame by another's advance is under pressure; a character who walks out of frame leaves the world of the shot, and whether the camera follows, holds on the empty space or cuts to the next scene is itself a choice. In many films of Ozu, the camera stays in a room after its occupants have left, holding on the emptied space for several seconds. This produces a sense of time continuing independent of human purpose, and it quietly shifts the film's attention from individual drama to the settings, rooms, corridors and streets, that outlast it. Movement, height and the dynamics of staging Blocking is not a sequence of still compositions. Actors move, and movement redistributes power within the frame. A character who rises from a chair gains height and prominence; one who sits down cedes them. A character who crosses in front of another interrupts the second's visibility and often seizes the scene. A character who moves toward the camera grows in the frame and increases in intensity; one who moves away diminishes. Relative height is one of the oldest and most reliable resources of staging. In most cultures, being higher reads as having more power, and filmmakers exploit this constantly. In the courtroom, the judge sits elevated; in the throne room, the king. But the effect of height depends on context. A standing character looming over a seated one may be dominant, or may be anxious, pleading and exposed; a seated character may be powerless, or may be the calm center around whom others orbit. In the opening scene of Francis Ford Coppola's The Godfather (1972), Don Corleone sits behind his desk in his darkened study while supplicants stand before him or sit in lower chairs. His seated stillness, not his height, marks his power: others move, explain, plead; he barely shifts, and the camera comes to him. Here stillness is authority and movement is need. A useful exercise is to watch a dialogue scene with the sound off and chart the movements of each character: who initiates movement, who responds, who closes distance and who opens it, who ends the scene in the dominant position. In a well-directed scene, this silent map will often tell you as much about the balance of power as the dialogue does, and sometimes it will contradict the dialogue. Characters who say they are in charge may be positioned as subordinate; characters who apologize may be staged as the victors. When blocking and dialogue diverge, the blocking is usually closer to what the film wants you to understand. Directors differ enormously in how much they rely on blocking. Some treat it as the primary means of expression; others do most of their shaping through editing and use staging mainly to give editors usable coverage. The difference can be felt without technical knowledge. In a film by Kenji Mizoguchi, the Japanese director whose Sansho the Bailiff (1954) and Ugetsu (1953) are classics of long-take staging, the camera often holds back while figures move through complex spaces, and the dramatic climax of a scene may occur at a distance, with the key figure small in the frame. In the films of the Taiwanese director Hou Hsiao-hsien, such as The Puppetmaster (1993) and Flowers of Shanghai (1998), scenes are frequently played in long shots from a fixed position, with actors moving in and out of patches of light and in and out of view behind doorframes and furniture. The viewer learns to read the scene from the whole disposition of bodies, not from faces. This difference in approach matters for close reading because it tells you where to look for meaning in a given film. In a heavily edited film, blocking within each shot is often simple and the significant relations are built by the cuts. In a staging-based film, the single shot is the essential unit, and a close reading that skipped over the arrangement of figures within it would miss most of what the film does. Part of reading well is identifying which kind of film you are watching, which is to say, which paradigms the filmmaker is actively choosing from and which are held constant. A director who uses the same fixed camera distance throughout has taken distance off the table as a variable, and the viewer's attention shifts to what the director does vary. Blocking, in such films, carries the weight that elsewhere would be spread across many kinds of choice. Chapter 3. Light as Argument Nothing is visible on film without light, and nothing about the light in a film is accidental. Even when a cinematographer works entirely with available daylight, choices have been made about the time of day, the direction of the camera relative to the sun, the exposure, and what is allowed to fall into shadow. In the studio era, when almost every scene was shot on a stage, every source of light in the image was placed by hand. Lighting is therefore one of the most complete sets of paradigmatic choices a filmmaker makes, and it is among the least noticed by viewers, who tend to register it only as mood. This chapter argues that light in cinema is always an argument about what matters. It tells the viewer where to look, what is known and what is hidden, which faces are to be admired and which to be feared, and what kind of world the film takes place in. It does this through a small number of variables that can be described precisely, and the critic who learns to name them gains access to a layer of meaning that most commentary treats as atmosphere. The variables of light Four properties of light do most of the work, and each forms a paradigm from which choices are made. The first is quality, the difference between hard and soft light. Hard light comes from a small or distant source, such as a bare bulb, a spotlight or the sun on a clear day, and produces sharp-edged shadows and strong texture. Soft light comes from a large or diffused source, such as an overcast sky, a light bounced off a white card or a lamp behind a silk screen, and produces gradual transitions from light to shadow and a smoother rendering of surfaces. Hard light tends to dramatize and reveal texture, including wrinkles, pores and scars; soft light tends to flatter and to calm. The second is direction. Light from the front flattens features and reduces modeling. Light from the side models the face into a lit half and a shadowed half, producing the effect painters call chiaroscuro. Light from behind, called backlight, rims the figure with a halo and separates it from the background. Light from below, called underlighting, reverses the ordinary pattern of shadows cast by sunlight and overhead lamps, and because it is unfamiliar it tends to look unnatural and sinister, which is why it became a staple of horror. Light from directly above can make eye sockets into dark pits. The third is source, the question of whether the light seems to come from somewhere within the world of the film, such as a window, a lamp or a fire, or whether it is simply there. Cinematographers speak of motivated light, which is placed to appear as if it comes from a visible or implied source in the scene, and unmotivated light, which has no such justification. Classical cinema motivated its key light loosely; later traditions made motivation a point of principle. The fourth is color, including both the color temperature of the light, from the warm orange of candlelight and tungsten bulbs to the cool blue of daylight and moonlight, and any deliberate coloring by gels or grading. Color temperature carries strong connotations in most cultures: warmth, intimacy and domesticity at one end, coldness, clinical detachment or night at the other. Filmmakers use these connotations and sometimes invert them. On top of these four sits the overall key of the lighting, the ratio of bright to dark across the image. High-key lighting keeps the ratio low, with fill light brightening the shadows so that the whole frame is evenly visible. Low-key lighting keeps the ratio high, leaving large areas of deep shadow. The terms are used loosely, but the distinction is fundamental and easy to see. The three-point system and its ideology The Hollywood studios, by the 1930s, had standardized an approach to lighting figures that is still taught in film schools as the basic grammar of the craft. It is usually called three-point lighting. A key light, the brightest source, is placed to one side of the camera and somewhat above the subject, modeling the face and establishing the main direction of illumination. A fill light, softer and weaker, is placed on the other side of the camera to lighten the shadows cast by the key, controlling how dark those shadows are. A backlight is placed above and behind the subject, outlining the hair and shoulders and separating the figure from the background. Additional lights illuminate the set. Three-point lighting is often described as a neutral technique, but it embodies a set of values. It is designed to make the principal figures of a scene legible, modeled and separated from their surroundings: to make stars look like stars. The backlight's halo, especially on blond hair, gave the leading women of the studio era a glowing, almost sanctified quality. The systematic separation of figure from ground reinforced the classical cinema's focus on individual characters as the drivers of the story, with settings as backdrop. And the careful balance of key and fill ensured that faces were never obscured by shadow unless the film wanted them to be. The system made human faces the center of the visual world, and it ranked those faces by how much light they received. The film scholar Richard Dyer pressed this point further. In White (1997), he argued that the lighting technologies and conventions of Western cinema and photography were developed around the rendering of white skin, and that the ideal of the glowing, luminous face was racially coded. Film stocks were calibrated, exposure standards set and lighting practices refined with pale faces as the norm, so that darker skin was routinely under-lit, under-exposed or rendered as a flat mass lacking detail. The historian of photography Lorna Roth documented a related history in the calibration cards, known in the industry as "Shirley cards" after the name of one of the early models, that color film laboratories used for decades to balance their printing, which featured white women. The point is not that individual cinematographers set out to exclude anyone, but that a technical system presented as neutral carried assumptions about whose face was the standard case. This history matters for close reading because it shows how deeply the paradigm of lighting choices is shaped by technology and institution before any individual filmmaker makes a decision. It also matters because contemporary filmmakers have made explicit counter-arguments in light. Barry Jenkins's Moonlight (2016), photographed by James Laxton, bathes its Black characters in saturated blues, purples and ambers, with skin rendered richly and with evident care for its range of tones. The film's lighting is an argument about who deserves luminous attention, and it reads more sharply against the history Dyer describes. A critic who notes only that the film is "beautifully lit" misses that its beauty is pointed. Low key, noir and the moral economy of shadow The Hollywood crime films of the 1940s and 1950s that French critics grouped under the name film noir are the most celebrated case of low-key lighting as a style. Their characteristic look, with faces half in shadow, streets wet and glistening under streetlamps, venetian blinds casting bars across rooms and figures silhouetted in doorways, was partly a result of economy, since shadows hide the cheapness of sets, partly an inheritance from German expressionist cinema of the 1920s, brought to Hollywood by émigré directors and cinematographers, and partly a deliberate effort to render a world of moral uncertainty. The cinematographer John Alton, who shot T-Men (1947) and The Big Combo (1955), published a manual on the craft, Painting with Light (1949), which remains one of the most direct accounts of how these images were made. Noir lighting encodes a moral economy. Where the classical three-point system keeps faces legible, noir lets shadow invade them, so that the viewer cannot be sure of a character's expression or intentions. Where classical lighting separates figure from ground, noir often lets figures dissolve into the darkness of the set, as if the city were absorbing them. The hard, raking light from low or unusual angles, the pools of illumination surrounded by black, the shadows of objects such as blinds or banisters projected across faces and walls: these produce an image of a world in which knowledge is partial and people are trapped. The bars of shadow cast by blinds are a common example because they are so easy to read as prison bars, and a code-book critic will make that reading every time. The commutation test asks for more: what if the scene were lit evenly? What exactly changes? Usually the answer is not a single symbol but a shift in the viewer's relation to the characters, from confident knowledge to uneasy guessing. Low-key lighting is not the property of noir alone. Gordon Willis, the cinematographer of The Godfather (1972) and its first sequel, lit the Corleone family's interiors from overhead with minimal fill, so that Marlon Brando's eyes often sit in dark pools and are hard to read. The choice was controversial at the time, since it broke the studio rule that a star's eyes should always be visible, and the cinematographer Conrad Hall is widely reported to have nicknamed Willis "the Prince of Darkness." In the film's opening scene, the contrast between the shadowed study, where the Don conducts his business, and the bright sunlit garden outside, where his daughter's wedding is under way, is a structural principle of the whole film: the family's public face is lit, its business is conducted in shadow, and the story tracks Michael Corleone's movement from one to the other. Naturalism, motivation and the ethics of source A contrary tradition treats light primarily as a record of the conditions under which a scene would have been seen. Here the question is not how to make the image expressive but how to make it faithful to the sources of light that exist in the world of the film. This tradition gained force in European cinema of the 1960s, when lighter cameras and faster film stocks made location shooting with available light practical, and cinematographers such as Raoul Coutard, who shot many of Jean-Luc Godard's early films, bounced light off ceilings to create a diffuse, even illumination that looked unlit. The most extreme case is Stanley Kubrick's Barry Lyndon (1975), set in the eighteenth century and photographed by John Alcott. For its interior night scenes, the production used exceptionally fast lenses originally made by Zeiss for NASA, which allowed the scenes to be filmed by candlelight alone. The result is an image that looks like eighteenth-century painting, soft and warm, with faces emerging from surrounding darkness, and a depth of field so shallow that figures slightly behind or in front of the plane of focus blur. The effect is historical in a precise sense: it shows a world as it would have looked when candles were the only light after dark. It is also expressive, since the pools of golden light surrounded by darkness give the aristocratic interiors a fragile, enclosed quality, as if the whole society were a brief illumination about to be snuffed out. Terrence Malick's Days of Heaven (1978), photographed largely by Néstor Almendros, is another landmark of source-conscious lighting. Many of its exterior scenes were shot in the brief interval around sunset when the sky provides soft, directionless, warm light, which filmmakers call magic hour. The practice was costly, since only a short window of shooting was possible each day, and the result is a landscape bathed in a glow that seems at once real and elegiac. The light makes a claim about the characters' world: that it is beautiful, transient, and about to end. As Table 1 sets out, these traditions differ systematically in their answers to the four basic lighting variables. The table is a simplification; many films mix approaches, and each tradition contains internal variety. Table 1. Four lighting traditions compared on their basic variables. Tradition Quality Key Source logic Typical claim Classical three-point Mixed, soft fill High Loosely motivated Stars are legible, central Expressionist and noir Hard Low Often unmotivated The world is unstable Available-light naturalism Soft, bounced Variable Strictly motivated This is how it looked Painterly source lighting Soft, directional Low Strictly motivated This world is fragile Because light is so pervasive, it can be hard to know where to start with it in a close reading. Three questions help. First, where is the brightest point in the frame, and is it where the story says it should be? The eye goes first to the brightest area, and classical films place their protagonist's face there. When the brightest point is elsewhere, on an object, a background, a secondary character, the film is directing attention away from its ostensible center, and that displacement is worth describing. Second, what is hidden? Every lighting design leaves some things in shadow. What those things are, and whether their hiding is motivated by the world of the film or imposed on it, tells you what the film wants withheld. In the opening scene of The Godfather, what is hidden is the Don's eyes, and with them the easy access to his inner life that close-ups of stars normally provide. The film makes him a figure to be interpreted rather than known. Third, how does the lighting change? Light in a well-made film is not static. It changes between scenes, often in patterns that track the story's arc, and it can change within a scene as characters move through it or as sources are switched on and off. A character who begins a film in full, flattering light and ends it in shadow has been given a visual trajectory, and so has one who moves the other way. Tracking these changes across a whole film is laborious, but it often reveals a structure the dialogue never states. The larger lesson of this chapter is that there is no neutral lighting. Even the flattest, most even illumination is a choice, made against alternatives, with consequences for what can be seen and how it feels. The illusion of neutrality is itself one of the most powerful effects a lighting system can produce, and it is the effect most in need of being noticed. Hashtags: #VisualSemioticsOfCinema #MiseEnScene #Montage #VisualRhetoric #FilmSemiotics #CloseReading #CinematicMeaning #CommutationTest #ParadigmAndSyntagm #DenotationAndConnotation #FilmSigns #Blocking #DeepFocus #FrameComposition #OffscreenSpace #CameraAngle #LensChoice #CinematicLighting #ThreePointLighting #FilmNoir #ContinuityEditing #SovietMontage #ShotDuration #VisualIdeology #FutureOfFilmAnalysis

Download the Book (PDF): Introduction In the summer of 1920 a Canadian-born American explorer named Robert Flaherty set up his cameras near Inukjuak on the eastern shore of Hudson Bay. He had been financed by the French fur company Revillon Frères, and he wanted to film the life of an Inuit hunter as it had been lived before traders like his sponsors arrived. The man he cast as the hunter, Allakariallak, was given a new name for the screen: Nanook. The women introduced as his wives were not his wives. The rifle that Inuit hunters had by then used for years was kept out of frame in favour of the harpoon. An igloo was built at outsized scale and cut in half so that the camera could see inside it in usable light. When Nanook of the North opened in 1922 it was hailed as a new kind of film, one that showed reality rather than invented stories, and it has remained on film-school syllabi ever since as the founding work of documentary. Nothing in that account is secret. Flaherty's methods have been discussed for a century, and his defenders point out, fairly, that Allakariallak and his companions collaborated with the filming, suggested scenes, repaired the camera, and developed the footage. What matters for this book is not whether Flaherty was dishonest. It is that the terms of the collaboration were not the Inuit's to set. They could perform, advise, and labour, but the decisions about what counted as authentically Inuit, what would be cut, what the finished film would argue, and where it would be shown and to whom, belonged to a man from elsewhere and the companies and audiences behind him. Eighty years later an Inuk filmmaker from Igloolik, Zacharias Kunuk, released Atanarjuat: The Fast Runner (2001), an epic built from a legend told by Igloolik elders, performed in Inuktitut by a largely Inuit cast, and produced by an Inuit-controlled company. It won the Caméra d'Or at Cannes. The distance between those two films is the subject of this book. A term and its meaning The phrase "visual sovereignty" was put into circulation by the Tuscarora artist and scholar Jolene Rickard in the mid-1990s, in writing about Native photography, and was developed at length by the Seneca scholar Michelle Raheja in her study of Native Americans in film. Rickard's point was that sovereignty, for Indigenous nations, is not only a legal status negotiated with states. It is also a practice carried out in culture: a people's continuing capacity to define itself, to hold its own knowledge, and to decide how it appears. Raheja extended the idea to cinema, arguing that Native filmmakers and actors have long used the screen both to answer the stereotypes produced about them and to address their own communities directly, making images that serve Indigenous purposes even when they pass through non-Indigenous institutions. Sovereignty is a strong word, and it is chosen deliberately. It is not a synonym for "representation" or "diversity." A film can include Indigenous faces, stories, and even crew, and still be organised entirely by other people's priorities. Visual sovereignty asks a harder set of questions. Who decides what is filmed and what is left unfilmed? Whose protocols govern access to ceremony, land, and the dead? Who owns the footage, and who can say no to its reuse? Whose sense of time shapes the edit? Who is the imagined first audience? When these questions are answered by Indigenous people acting as members of their communities, rather than as individual talents recruited into someone else's project, something changes not only in what films show but in how they are made and how they look. The argument The controlling claim of this book is that visual sovereignty is a matter of form and process, not only of content. Replacing a stereotyped image with a positive one does not by itself undo the colonial gaze, because that gaze was never just a collection of bad pictures. It was a structure: a way of placing the camera, of organising time, of extracting stories, and of assigning the right to look to one party and the duty to be looked at to another. Indigenous filmmakers who have most fully disrupted that structure have done so by rebuilding it at each point where it operated. They have moved the camera, sometimes literally, to where their communities stand. They have organised films around cyclical, layered, or genealogical time rather than the single forward line of the settler narrative. They have developed consultation protocols that make consent collective and continuing rather than a signature on a release form. And they have adapted oral storytelling traditions, which carry their own rules about who may speak and how, to the demands of documentary and dramatic screen media. The chapters follow that structure. Chapter 1 examines the colonial archive itself, from early ethnographic footage to the Hollywood western, and describes the gaze these films built. Chapter 2 traces the first experiments in which Indigenous people took hold of cameras, from the Navajo film project of the 1960s to community video in Central Australia and the Amazon. Chapter 3 sets out the main theoretical frameworks that have grown from that work, including the Māori filmmaker Barry Barclay's idea of a Fourth Cinema. Chapters 4, 5, and 6 turn to the craft of the image: cinematography and the ethics of framing, the handling of time and chronology, and the adaptation of oral storytelling. Chapter 7 examines the consultation and consent protocols that now govern much Indigenous screen production. Chapter 8 looks at documentary as a form of witness in moments of confrontation with the state. Chapter 9 considers the institutions, broadcasters, and streaming economies within which Indigenous screen work is now made, and asks what sovereignty can mean inside them. Scope and limits Indigenous cinema is not one tradition. The Inuit of the eastern Arctic, the Yolŋu of Arnhem Land, the Māori of Aotearoa, the Aymara of the Bolivian altiplano, the Sámi of northern Scandinavia, and the hundreds of Native nations within the United States and Canada have different languages, political histories, and relations to the states that surround them. A book of this length cannot survey them all, and it does not try. It concentrates on the places where Indigenous screen practice has been most sustained and most discussed: Canada and the United States, Australia, Aotearoa New Zealand, and, more briefly, Latin America and Sápmi. Examples are chosen because they show a mechanism clearly, not because they are representative of a whole people. Two further cautions govern what follows. The first is that the author of a book about visual sovereignty is not thereby its holder. Where Indigenous filmmakers and scholars have explained their own work, this book relies on their explanations and attributes them. Where communities have chosen to keep material restricted, the book does not attempt to describe it. The second caution is that sovereignty is not purity. Almost every film discussed here was made with money, equipment, training, or distribution from outside the community that made it. Kunuk's Isuma worked with Norman Cohn, a New York-born video artist. Rolf de Heer, a Dutch-born Australian, co-directed Ten Canoes with the Yolŋu man Peter Djigirr. The question is never whether outsiders were present, but who held the decisions that mattered, and whether those decisions were accountable to the people on screen. What emerges from these cases is not a checklist of Indigenous style. It is a set of practices, argued over and revised, through which communities that were long the objects of cinema have made themselves its authors. Those practices have things to teach anyone who holds a camera, including those who will never make a film about Indigenous people at all, because they expose assumptions about looking, time, consent, and story that most screen media still takes for granted. Chapter 1: The Colonial Gaze and Its Archive Cinema and colonial science grew up together. In 1895, the year the Lumière brothers first projected films to a paying audience in Paris, the French physician Félix-Louis Regnault used a chronophotographic camera at an ethnographic exposition in the same city to record a Wolof woman making pottery. Regnault believed that moving images could give anthropology the precision of the laboratory, capturing the gestures of peoples he regarded as closer to the origins of humanity. Three years later, the Cambridge zoologist-turned-anthropologist Alfred Cort Haddon carried a camera to the Torres Strait between Australia and New Guinea and filmed dancers on Mer. In 1901 the biologist Baldwin Spencer, travelling through Central Australia with the telegraph official Francis Gillen, filmed Arrernte ceremonial performances near Alice Springs. These are among the oldest surviving fragments of ethnographic film, and they were made within a few years of the medium's invention. The film historian Fatimah Tobing Rony, in The Third Eye (1996), argued that this timing was not coincidental. Early cinema offered a new way of possessing the world, and the peoples whom European science had classified as primitive were among its first subjects. Rony called the resulting attitude "fascinating cannibalism," a hunger to consume images of the racialised other that mixed curiosity with appetite and fear. Her title borrowed from W. E. B. Du Bois's account of double consciousness: the experience of seeing oneself through the eyes of those who hold one in contempt. For Indigenous viewers encountering these films, often decades later in archives or classrooms, the third eye is the uncanny sensation of watching one's grandparents turned into specimens. This chapter examines the structure of that gaze. It does so not to rehearse a catalogue of offensive images, which would be easy and not very useful, but to identify the specific operations through which colonial cinema positioned Indigenous people. Each of those operations later became a point of intervention for Indigenous filmmakers, and the rest of this book follows them one by one. The salvage paradigm The most durable idea in early ethnographic cinema was that Indigenous peoples were disappearing and had to be recorded before they vanished. The American photographer Edward S. Curtis made the idea explicit. One of the first images in his vast project The North American Indian, published between 1907 and 1930, shows a line of Navajo riders moving away from the camera into shadow. Curtis titled it The Vanishing Race. The riders are not dying; they are riding. The photograph turns an ordinary movement into a prophecy. Curtis extended the project to film. In 1914 he released In the Land of the Head Hunters, a feature-length drama shot on the coast of British Columbia with Kwakwaka'wakw performers. He worked closely with George Hunt, a man of Tlingit and English descent raised among the Kwakwaka'wakw, who had earlier been the principal collaborator of the anthropologist Franz Boas. The film invented a melodrama of love, sorcery, and warfare, set in a precontact past. Performers wore wigs and costumes to remove signs of contemporary life, and props were made or borrowed to supply what Curtis considered authentic. The actors were paid, and many clearly relished the performances. But the film's premise was that the Kwakwaka'wakw worth showing were those of an imagined past, and that the living community around the camera, which included wage workers, Christians, and people fighting the Canadian government's ban on the potlatch, was a corruption of that past. The anthropologist James Clifford later named this attitude the "salvage paradigm": the conviction that authentic culture lies in the past, that the present is decline, and that the outside observer's task is to rescue what remains. The paradigm has a tidy logic. If Indigenous cultures are vanishing, then the observer who records them becomes their custodian. The films, photographs, and recordings pass into museums and universities, where they are preserved, and the living descendants of the people recorded become, at best, informants about a heritage now held by others. Nanook of the North belongs to the same paradigm, though it is warmer in tone. Flaherty's intertitles insist on the happiness and resourcefulness of his Inuit subjects, and the film's affection is real. But its method was salvage. Flaherty staged scenes of walrus and seal hunting with harpoons rather than rifles, suppressed the presence of the trading post that structured Inuit economic life in the region, and presented a family composed for the camera as a natural unit. An intertitle at the end of later prints reports Nanook's death from starvation on a hunting trip. Scholars have long doubted the claim, which appears to have been a dramatic embellishment. The point is that the film ends by confirming what the salvage paradigm needed: the hunter vanishes, and the film survives him. The denial of shared time The anthropologist Johannes Fabian, in Time and the Other (1983), gave a name to a habit he found throughout his discipline. Anthropologists, he argued, routinely described the people they studied as living in a different time from the observer's own, as though a trip to the field were a trip backward. Fabian called this the "denial of coevalness": the refusal to grant that the observer and the observed share the same historical present. Cinema made this denial vivid. Ethnographic films were edited to remove clocks, cars, clothing from mail-order catalogues, tinned food, and anything else that would place their subjects in the twentieth century. The effect was to present Indigenous people as survivors from an earlier stage of human development, visible now only because the camera had arrived in time. When contemporary signs did appear, they were framed as contamination, a sign that authenticity was slipping away. Curtis's wigs and Flaherty's harpoons are instances of the same editing principle. Popular cinema took up the same temporal logic in a different register. The Hollywood western, which dominated American film production for half a century, almost always placed Indigenous people in the nineteenth century, as obstacles to settlement or as noble relics of a doomed order. There was little room in this genre for Indigenous people who drove trucks, served in the armed forces, held jobs in cities, or ran tribal governments. When the audience left the cinema, the Indians stayed behind in the past. This matters because time is one of the places where Indigenous filmmakers have most deeply reworked the image. If the colonial gaze fixed Indigenous people in a vanished past, then a sovereign cinema must be able to show Indigenous people as contemporaries, and also to represent the relation between past and present on Indigenous terms, which are rarely the terms of a simple line from before to after. Chapter 5 takes up how filmmakers have done this. Redfacing and the performed Indian Hollywood rarely cast Native actors as leads in Native roles. Michelle Raheja, in Reservation Reelism (2010), uses the term "redfacing" for the practice of non-Native performers playing Native characters, often with makeup, wigs, and invented speech. The practice ran from the silent period through the studio era and beyond. Some of the most recognisable screen Indians of the twentieth century were played by actors of European descent. Iron Eyes Cody, who appeared in scores of westerns and in a famous 1971 anti-littering advertisement in which a tear rolls down his cheek, presented himself for decades as Native American; he was born to Italian immigrant parents in Louisiana. Redfacing was not only a matter of casting. It extended to language, where actors spoke invented gibberish or broken English, and to material culture, where Plains headdresses and tipis were used as generic signs of Indianness regardless of the nation depicted. John Ford filmed many of his westerns in Monument Valley on the Navajo Nation and employed Navajo people as extras and riders. In several films Navajo performers played Comanche or Apache characters and spoke Navajo on screen, sometimes, according to later accounts by Navajo viewers, making jokes at the expense of the production that no one on set could understand. Raheja treats such moments seriously. They are evidence that even inside the most confining productions, Native performers found ways to address their own communities over the heads of the film's intended audience. That observation is the seed of the concept of visual sovereignty. Raheja does not claim that these performers controlled the films in which they appeared. They clearly did not. She argues that they exercised a limited, tactical form of authorship within them, and that later Native filmmakers built on this tradition of double address. It is a reminder that the history of Indigenous people and cinema is not simply a story of exclusion followed by entry. People were present on screen and behind the camera from the beginning, working, negotiating, and sometimes subverting. What they lacked was control over the terms. The structure of the gaze It is now possible to describe the colonial gaze less as a set of images than as a set of operations. Five stand out. The first is the one-way look. The camera belongs to the observer, and the observed have no reciprocal right to look back, to see the footage, or to control its circulation. In most early ethnographic filming, subjects never saw the finished work. Recordings left the community and entered archives where the community had no standing. The second is typification. Individuals are presented as examples of a type: "the Eskimo," "the Aborigine," "the Plains Indian." Names, when used, are often invented or simplified, as Allakariallak became Nanook. The individual's particular history, kin relations, and opinions matter less than the degree to which he or she illustrates the category. The third is the denial of shared time, described above. The subject is placed in a past that the observer has travelled to reach and will leave behind. The fourth is extraction. Stories, songs, ceremonies, and images are taken from the community and become the property of the filmmaker, the studio, or the institution. This is true whether the material was freely offered, purchased, or obtained without real consent. The community's own rules about who may see or hear such material, which in many Indigenous societies are elaborate and binding, are not recognised. The fifth is the authority of the external voice. Whether through intertitles, voice-over narration, or the implied perspective of the edit, the film explains its subjects to an audience that is assumed not to include them. The subjects may speak, but their speech is framed, translated, and interpreted by someone else. Each of these operations can be reversed, and each has been. Indigenous filmmakers have made the look reciprocal by addressing their own communities first and by returning footage to them. They have refused typification by making films about particular people with particular names, genealogies, and grievances. They have asserted shared time, and more than that, they have introduced ways of structuring time that do not assume a single line of progress. They have replaced extraction with protocols of consultation, consent, and ownership. And they have displaced the external voice by building films around Indigenous speech, languages, and storytelling forms. The chapters that follow examine each of these reversals. Living with the archive The colonial archive has not gone away. The footage shot by Spencer, Curtis, Flaherty, and hundreds of lesser-known observers survives in national film archives, museum collections, and university libraries. Much of it is freely viewable online. For Indigenous communities it is a difficult inheritance. It contains the faces and voices of ancestors, sometimes the only moving images of them that exist, and it records songs, dances, and ceremonial knowledge that may since have been lost or reshaped. It also carries the gaze under which it was made, and in some cases it shows material that the community's own law restricts: ceremonies meant only for initiated men or women, or images of people who have died, which in many Aboriginal Australian communities should not be shown or named for a period after death. Communities have responded to the archive in several ways. One is repatriation and restricted access. Australian film and sound archives now routinely consult Aboriginal and Torres Strait Islander communities about access to culturally sensitive material, and many collections carry warnings or have been placed under restricted conditions. Another response is reinterpretation. In 2008 a team led by the anthropologist Aaron Glass and the literary scholar Brad Evans, working with the U'mista Cultural Centre in Alert Bay and Kwakwaka'wakw performers, restored Curtis's 1914 film under its original title and toured it with live performances by descendants of the original cast. The restoration did not erase Curtis's authorship. It placed the film in a new frame, one in which the Kwakwaka'wakw were visibly present as contemporaries interpreting their own ancestors' work. A third response is reuse. Indigenous filmmakers have incorporated archival footage into new works that turn its meanings around. The Australian artist Tracey Moffatt's short film Night Cries: A Rural Tragedy (1989) reworks, as its point of departure, the imagery of Charles Chauvel's 1955 feature Jedda, the first Australian feature shot in colour, in which a young Aboriginal woman raised by a white station family is drawn back toward Aboriginal life with fatal results. Moffatt's film imagines an adult Aboriginal woman caring for her dying white adoptive mother in a claustrophobic studio landscape. It says nothing directly about Jedda's politics; it simply occupies the space the earlier film left out, the long aftermath of the removal that Chauvel's story treated as tragedy. These responses share a principle. They do not treat the colonial archive as neutral evidence to be mined, nor as poison to be destroyed. They treat it as a record of a relationship, one in which Indigenous people were present and active but did not hold the decisive power, and they seek to change the relationship. The same principle runs through the rest of this book. Visual sovereignty does not begin from nothing. It begins from a century of being filmed, and it takes its shape from the specific ways that filming was done. What the gaze teaches It is tempting to treat the early ethnographic film and the Hollywood western as relics, easily seen through by modern audiences. The operations they depended on are still in wide use. Wildlife-style documentaries about "uncontacted" Amazonian peoples, travel programmes that present Indigenous ceremonies as spectacle, advertising that uses Indigenous imagery as a sign of spiritual authenticity, and news coverage that appears only in moments of crisis all reproduce elements of the structure. The one-way look persists wherever footage leaves a community without any continuing relation to it. Typification persists wherever a single speaker is made to stand for a whole people. The denial of shared time persists whenever Indigenous life is illustrated by feathers and drums rather than by the ordinary present. The persistence of these operations explains why Indigenous filmmakers have been wary of solutions that simply swap bad images for good ones. A positive stereotype is still a stereotype. A film that celebrates Indigenous spirituality with reverent slow motion and swelling music can typify and extract just as thoroughly as one that mocks it. The shift that visual sovereignty requires is not from negative to positive, but from object to author, and that shift reaches into every decision a filmmaker makes. The first attempts to make it happen, in the 1960s and 1970s, showed both how much could change when Indigenous people held the camera and how much of the old structure could survive that change. Chapter 2: Taking the Camera In the summer of 1966 two researchers from the University of Pennsylvania, the communications scholar Sol Worth and the anthropologist John Adair, arrived at Pine Springs, a small Navajo community in Arizona, with sixteen-millimetre cameras and a question. Was film a universal language, readable in the same way by all people, or would people from a different culture, given cameras and minimal instruction, organise images according to their own sense of how the world should be seen? With the help of the graduate student Richard Chalfen, they taught a handful of Navajo participants, among them Johnny Nelson, Al Clah, Susie Benally, Mike Anderson, and Maxine and Mary Jane Tsosie, how to operate the cameras and splice film, and deliberately refrained from teaching the conventions of Hollywood continuity editing. The results were published in 1972 as Through Navajo Eyes. The films themselves are short and unassuming: a silversmith at work, a weaver preparing wool, a portrait of a spring, a meditation by Al Clah called Intrepid Shadows that follows the play of light and wind around a doll. Worth and Adair noted features they took to be culturally significant. The Navajo films contained long sequences of walking, which the researchers had not expected, and they showed processes, such as the gathering of materials for weaving, that a Hollywood editor would have compressed or cut. Some participants were reluctant to show close-ups of faces. The researchers suggested that the films reflected Navajo patterns of narrative and perception, in which journeys and the steps of a process carried meaning in themselves. The experiment and its limits Through Navajo Eyes is often cited as the first Indigenous-authored film project, and it deserves attention both for what it achieved and for how it was framed. It demonstrated, against the assumption that cinema was a transparent window, that people with different cultural training would make different films. That finding was important. It helped undermine the idea that the conventions of Western cinema were natural rather than learned. But the project also reproduced much of the structure it seemed to challenge. The Navajo filmmakers were participants in an experiment designed by outsiders to answer outsiders' questions. The findings were interpreted and published by the researchers. The films circulated mainly in academic contexts. Later critics, including the anthropologist Sam Pack, who revisited the project and its participants, have pointed out that the researchers tended to attribute the films' features to a static Navajo cultural essence rather than to the specific experiences, preferences, and circumstances of the people who made them. Some of those features may have owed as much to inexperience, the constraints of the equipment, or the participants' understanding of what the researchers wanted as to Navajo worldview. The project gave Navajo people cameras. It did not give them control of the project's meaning. The tension is worth dwelling on because it recurs. When outsiders design a programme to put cameras in Indigenous hands, the resulting films are often read as windows into a culture, as though their makers were transmitting their people's worldview rather than making choices as individuals. That reading can be flattering, but it is a subtle form of the typification described in the previous chapter. The Navajo filmmaker becomes an example of Navajo seeing, much as the subject of an ethnographic film was an example of Navajo life. Visual sovereignty requires something more: that Indigenous filmmakers set the terms of the project, decide what it is for, and answer primarily to their own communities rather than to the researchers or funders who made it possible. Challenge for Change and the Indian Film Crew In 1967 the National Film Board of Canada launched a programme called Challenge for Change, which aimed to use film and video as tools of social change by giving marginalised communities the means to document their own concerns and present them to government. The best-known early experiment took place on Fogo Island, Newfoundland, where films of residents' discussions were shown back to the community and to officials in a cycle intended to open dialogue. The programme also created an Indian Film Crew, a group of young First Nations people trained in filmmaking at the NFB. The most influential film associated with the crew is You Are on Indian Land (1969), which documents a December 1968 blockade of the international bridge crossing Cornwall Island, in the Mohawk community of Akwesasne on the border between Ontario and New York. The Mohawk protesters were asserting their rights under the Jay Treaty of 1794 to cross the border without paying duty on personal goods. The film shows the confrontation with police, the arrests, and the discussions among community members. For many years it was credited to the NFB director Mort Ransen. The Board has since recognised the Mohawk organiser Mike Kanentakeron Mitchell, who was a member of the Indian Film Crew, as its director, with Ransen as producer. The change of credit is itself a small act of visual sovereignty applied retrospectively: the film is now attributed to the community member who initiated it rather than to the institution's staff filmmaker. What distinguished You Are on Indian Land from earlier documentaries about Indigenous protest was its position. The camera stands among the protesters, not behind the police lines or at the observer's distance of a television crew. The film treats the protesters' arguments as the centre of the story and allows long passages of their speech. The NFB then used the film in the way Challenge for Change intended, screening it for both the community and government officials. It was not a work of independent Indigenous media in the full sense, because it was produced inside a federal institution, but it showed what a change of position could do. The Indian Film Crew did not last long as a formal unit, but several of its members went on to careers in media and politics, and it helped prepare the ground for Alanis Obomsawin, the Abenaki singer and storyteller who joined the NFB in 1967 and became one of the most important documentary filmmakers in Canada. Her work is the subject of a later chapter. Inventing television in the desert The most influential theoretical account of early Indigenous media came from Central Australia. In 1982 the American researcher Eric Michaels arrived at Yuendumu, a Warlpiri settlement northwest of Alice Springs, to study the likely effects of satellite television, which was about to reach remote Australia. The fear, widely shared, was that mainstream television would overwhelm Aboriginal languages and cultures in communities that had so far been beyond its reach. Michaels found that the Warlpiri were not waiting to be overwhelmed. Working with Warlpiri people, most prominently Francis Jupurrurla Kelly, he helped establish a local video production and, from 1985, an unlicensed community television service broadcasting Warlpiri-made programmes. The organisation became the Warlpiri Media Association, later PAW Media. Michaels described the process in a report whose title captured his argument: The Aboriginal Invention of Television in Central Australia, 1982–1986. The Warlpiri, he argued, did not simply adopt television; they reinvented it according to their own social rules. Video production followed Warlpiri structures of authority over knowledge. The right to tell a particular story, perform a particular song, or show a particular place was held by particular people by virtue of their kinship and their relation to the country concerned. A video recording of a ceremony required the involvement of those who held the relevant rights, and its circulation had to respect the distinctions between material for everyone and material restricted by gender or initiation. Michaels also observed features of the videos' form. Warlpiri videographers, he reported, often included long, slow pans across apparently empty landscape, which non-Aboriginal viewers found tedious. For Warlpiri viewers these shots were full. They traced the paths of ancestral beings across country, the tracks along which the Dreaming stories run, and they showed places whose names, owners, and associated stories viewers could supply from their own knowledge. The camera was not looking at scenery. It was following a story that the land itself held. What appeared to outsiders as the absence of action was, for the intended audience, a dense text. This finding had implications that reach through the rest of this book. It suggested that Indigenous screen forms might differ from Western ones not because of inexperience but because they answered to different audiences with different knowledge. It suggested that the relation between camera and land could itself carry cultural meaning. And it showed that questions of who may film and who may watch were not afterthoughts to production but its organising principle. Video in the villages At roughly the same time, Indigenous groups in the Brazilian Amazon were taking up video for purposes that were both cultural and bluntly political. The anthropologist Terence Turner, who had worked for many years with the Kayapó of central Brazil, described in a 1992 article how Kayapó communities had adopted video cameras in the 1980s. They used them to record their own ceremonies, to document their encounters with Brazilian officials so that promises could not later be denied, and to perform their presence for national and international media. At a large gathering at Altamira in 1989, organised to oppose a planned hydroelectric scheme on the Xingu River, Kayapó participants carried video cameras, and photographs of Kayapó men in traditional dress filming the proceedings circulated widely. Turner argued that the Kayapó were not being absorbed into a foreign medium but were appropriating it to strengthen their own social and political projects, including the conscious performance of cultural identity for outside audiences. The Kayapó cameras were matched by an organised programme. In 1986 the indigenist Vincent Carelli founded Vídeo nas Aldeias, "Video in the Villages," which began by filming Indigenous communities and showing them the results. An early and often-recounted episode involved the Nambikwara. Having watched footage of their girls' initiation ceremony, members of the community were dissatisfied with how they appeared in it, dressed in the Western clothes of everyday life. They decided to perform the ceremony again for the camera, this time in older forms of dress and adornment, so that the video would show it as they wished it to be seen. Carelli's film of the process, A Festa da Moça (1987), includes their reaction. Outside observers could read this as an invention of tradition for the camera. The more interesting reading is that the Nambikwara were using the camera to make a collective decision about how to represent and renew their own practices. Over the following decades Vídeo nas Aldeias shifted from filming communities to training Indigenous filmmakers, who went on to make their own films in their own languages. The Faustian contract By the late 1980s enough Indigenous media existed for scholars to begin asking what it meant. The American anthropologist Faye Ginsburg, who worked extensively with Aboriginal media makers in Australia, framed the central dilemma in a 1991 article titled "Indigenous Media: Faustian Contract or Global Village?" The danger was plain. Television, video, and film were technologies built by the same societies that had colonised Indigenous peoples, and they carried assumptions about narrative, audience, and value that might gradually reshape any culture that used them. To take up the camera might be to sign a Faustian contract, gaining power at the price of the very distinctiveness one hoped to protect. Ginsburg's answer was that the contract was not inevitable. The evidence from Yuendumu, the Kayapó, and elsewhere suggested that communities were using media to reinforce their own social relations, languages, and knowledge systems, and to intervene in the politics of the states that surrounded them. The outcome depended on who controlled production and distribution, and on whether media practices were embedded in community life or imposed from outside. Media could be a means of cultural continuity rather than its end. This debate prepared the ground for the rise of Indigenous feature filmmaking. In the eastern Arctic, Zacharias Kunuk, a young Inuk from Igloolik, reportedly sold carvings in the early 1980s to buy a video camera. He worked with the Inuit Broadcasting Corporation, which was established in 1981 to produce Inuktitut-language television, and in 1990 founded Igloolik Isuma Productions with Paul Apak Angilirq, Pauloosie Qulitalik, and Norman Cohn. Their early works, including the television series Nunavut (1994–95), dramatised life in the Igloolik region in the 1940s, with community members performing daily activities and the dialogue largely improvised in Inuktitut. The idea behind these works was that the knowledge of elders could be carried forward by re-enacting it, and that the camera could serve the community's own memory rather than an outsider's archive. From these projects came Atanarjuat. What changed and what did not These early experiments share a pattern. In each, the camera passed into Indigenous hands, and in each the films that resulted looked and worked differently from what outsiders had made. Landscape carried stories. Processes were shown whole. Speech was in Indigenous languages. Footage was shown to the community first. Political encounters were recorded as evidence rather than as spectacle. Rules about who could film and who could watch shaped what was recorded at all. Yet in most of these cases the institutional frame remained external. Worth and Adair's project was a university experiment. The Indian Film Crew worked inside a federal agency. Michaels was a researcher paid to study television's effects. Vídeo nas Aldeias was founded by non-Indigenous activists. The history of Indigenous media is not a story of independence won in a single stroke. It is a story of Indigenous people working inside, alongside, and eventually against the institutions that first gave them access, gradually shifting control of decisions from those institutions to themselves. The next chapter examines the ideas that Indigenous filmmakers and scholars developed to describe and guide that shift. Chapter 3: Fourth Cinema and the Language of Sovereignty In 1969 two Argentine filmmakers, Fernando Solanas and Octavio Getino, published a manifesto called "Towards a Third Cinema." They divided the world's film production into three kinds. First Cinema was Hollywood and its imitators: industrial, commercial, built around stars and spectacle, and designed to turn audiences into passive consumers. Second Cinema was the European art film and its heirs, the cinema of the individual auteur, which offered personal expression but remained inside the same system of production and distribution. Third Cinema would be something else: a militant cinema of liberation, made collectively in the countries of the colonised and neocolonised world, designed to provoke discussion and action rather than contemplation. Solanas and Getino's own film La hora de los hornos (The Hour of the Furnaces, 1968) was meant to be stopped during screenings so that audiences could debate what they had seen. Third Cinema was influential across Latin America, Africa, and Asia, and it shaped the vocabulary of politically committed filmmaking for a generation. But it was a cinema of the nation. Its heroes were the people of Argentina, Cuba, Algeria, or Bolivia struggling against imperialism and domestic elites. Indigenous peoples could appear within it, but usually as part of a larger national or class struggle, and the national projects Third Cinema supported were not always friendly to Indigenous claims. A postcolonial state could be as determined as a colonial one to settle Indigenous lands, suppress Indigenous languages, and absorb Indigenous people into a single national identity. Barclay's Fourth Cinema Barry Barclay, a filmmaker of Ngāti Apa descent, was the first Māori to direct a feature film, Ngati (1987), set in a small East Coast community in 1948. He had earlier worked with the historian Michael King on the television documentary series Tangata Whenua (1974), which brought Māori community life and speech to New Zealand television in a form that allowed people to talk at length in their own settings. Over the following decades he made documentaries, including The Neglected Miracle (1985), about the appropriation of plant genetic resources from Indigenous and peasant farmers, and The Feathers of Peace (2000), about the Moriori of the Chatham Islands. He also wrote a book on filmmaking, Our Own Image (1990), and later a study of Māori intellectual property, Mana Tuturu (2005). In a 2002 lecture at the University of Auckland, later published in the film journal Illusions, Barclay proposed that Indigenous film deserved its own category, which he called Fourth Cinema. His reasoning extended Solanas and Getino's scheme. First, Second, and Third Cinema, for all their differences, were cinemas of the modern nation-state. They assumed that the nation was the frame within which film was made and viewed, even when they opposed a particular government. Fourth Cinema, by contrast, was the cinema of peoples who predated the nation-state and who remained, in some sense, outside it: the Indigenous peoples of the settler colonies and beyond, whose ties to land, language, and kin were older than the states that had enclosed them. Barclay was careful not to define Fourth Cinema by a list of stylistic features. He spoke instead of "interior" values, the codes and relationships that governed Indigenous communities and that a film made from inside such a community would carry, often invisibly to outsiders. A Fourth Cinema film would not simply depict Māori or Aboriginal or Inuit life. It would be made according to the protocols of the people depicted, would answer to them, and would carry their ways of speaking, relating, and understanding into its structure. He acknowledged that the category was a proposal rather than a finished theory, and that he could describe it only partially. Barclay's most quoted image comes from Our Own Image. Thinking about the arrival of European ships in the Pacific, he contrasted two possible camera positions. The histories that New Zealanders had inherited were filmed, so to speak, from the deck of the ship, looking toward the people gathered on the shore. A Māori cinema would put the camera on the shore, among the people watching the ships come in. The image is simple, but its implications are large. It locates the difference between colonial and Indigenous cinema not in subject matter but in position: where the camera stands, whose perspective it shares, who is inside the frame and who is outside it. It is an argument about form. Barclay's own practice matched the image. On Tangata Whenua and Ngati he insisted on approaching communities through their own processes of discussion and consent, on hiring Māori crew, and on letting people speak at their own pace. He argued that the camera should enter a community as a guest, subject to the obligations of a guest, and that the community's leaders should have real power over what was filmed and how it was used. For Barclay, Fourth Cinema was as much an ethics of production as a style. Visual sovereignty The phrase "visual sovereignty" came from a different tradition, that of Native American art and photography. Jolene Rickard, a Tuscarora artist and scholar, used it in a 1995 essay in the photography journal Aperture titled "Sovereignty: A Line in the Sand." Rickard was responding to a history in which photographs of Native people had been made almost entirely by outsiders, and to a present in which Native photographers were beginning to produce a body of work of their own. She argued that sovereignty, a concept central to the political and legal struggles of Native nations in the United States, also had to be understood culturally. The ability of a people to define themselves in images, to carry forward their own knowledge, and to resist incorporation into the visual regimes of the settler state was itself a form of sovereignty, and one that underpinned the legal and political forms. Michelle Raheja, a Seneca scholar, took the concept into film studies in a 2007 article on Atanarjuat and in her book Reservation Reelism (2010). Her definition is characteristically double. Visual sovereignty, for Raheja, involves Native filmmakers and performers engaging with the stereotypes and conventions of mainstream media, sometimes working within them, sometimes turning them inside out, while at the same time addressing Native audiences with images and meanings rooted in their own communities. It is not a pure or separatist position. It accepts that Native people live within, and must respond to, a media world dominated by others, and it looks for the ways in which they have exercised authorship within that world. Raheja illustrates the point with a reading of Nanook of the North itself. In one famous scene Nanook is shown a gramophone at the trading post and, according to the intertitle, puzzles over it and bites the record. The scene was intended to demonstrate Inuit innocence of modern technology. Raheja argues that Allakariallak's performance can be read as knowing, even ironic: an Inuk hunter who had lived near a trading post for years, playing the part of the naive primitive for a camera he knew well, with a smile that might be addressed to Inuit viewers rather than to Flaherty's audience. Whatever Allakariallak intended, the reading models a way of watching the colonial archive that looks for the traces of Indigenous agency within it. The same book introduces the idea of the "virtual reservation," a space created by Native media in which Native knowledge, humour, and imagination can be expressed and preserved, a space not bounded by the physical reservation or by the expectations of non-Native audiences. The phrase captures something that Barclay's "interior" also names: a zone of meaning addressed first to insiders. Embedded aesthetics and survivance Faye Ginsburg, whose work on Aboriginal media in Australia was introduced in the previous chapter, contributed a further concept in a 1994 article, "Embedded Aesthetics." Ginsburg noticed that when Aboriginal media makers judged their own work, they did not always use the criteria that film critics used. A video might be praised not for its visual beauty or narrative economy but because it had been made properly: because the right people had been consulted, because the relevant owners of a story or place had been involved, because the production had strengthened relationships rather than strained them. The aesthetic value of the work was embedded in the social relations of its making. Ginsburg's concept is important because it shifts attention away from the finished text toward the process. It suggests that two films with identical images could have different value, even different meaning, depending on how they were made. It also gives a theoretical basis to what Barclay described as the ethics of the guest and what later became formalised in consultation protocols. For many Indigenous filmmakers, a film made without proper consent is not merely unethical. It is a bad film, because it has failed in its most basic task of carrying the community's voice. The Anishinaabe writer and theorist Gerald Vizenor contributed a term that has been widely adopted in Indigenous film criticism: "survivance." Vizenor coined it in Manifest Manners (1994) and elaborated it in later work. It combines survival and resistance, but its meaning is more specific than either. Survivance names an active sense of presence in Native stories, a refusal of the tragic narratives of victimry and disappearance that dominated representations of Native people. A story of survivance does not deny suffering. It refuses to let suffering be the whole story, insisting instead on continuance, wit, irony, and self-determination. The concept has proved especially useful for reading Native comedy and genre filmmaking, which has often been dismissed as less serious than documentary or historical drama but which may do more to undermine the vanishing-Indian narrative than any number of solemn laments. Comparing the frameworks These concepts overlap, and critics often use them together, but they emphasise different things. As Table 1 sets out, some locate the Indigenous difference in the political relation to the nation-state, some in the process of production, some in the address to audiences, and some in the narrative stance of the work. Table 1. Frameworks for Indigenous screen practice. Framework Main source Core claim Chief emphasis Third Cinema Solanas and Getino Film as a tool of national liberation Political struggle Fourth Cinema Barry Barclay Indigenous film stands outside the nation-state frame Interior values, community Visual sovereignty Jolene Rickard; Michelle Raheja Self-definition through images is a form of sovereignty Double address, authorship Embedded aesthetics Faye Ginsburg A work's value lies in how it was made Social process Survivance Gerald Vizenor Active presence over narratives of victimry Narrative stance The frameworks are not rival theories so much as angles on a single problem. Each responds to one or more of the operations of the colonial gaze described in Chapter 1. Fourth Cinema and embedded aesthetics answer extraction and the one-way look by relocating authority in the community. Visual sovereignty answers the external voice by describing a double address in which Indigenous audiences are primary. Survivance answers the denial of shared time and the salvage paradigm by insisting on continuance rather than disappearance. Objections and refinements These ideas have not gone unchallenged. One objection is that they risk essentialism, the assumption that Indigenous films must share a common character because their makers share an identity. Critics have pointed out that Indigenous filmmakers vary enormously in style, politics, and ambition, and that a theory of Indigenous cinema can become a new cage, requiring Indigenous artists to make films that display their difference. The Māori director Taika Waititi, whose films range from the coming-of-age comedy Boy (2010) to the Hollywood superhero franchise, has been celebrated and criticised in roughly equal measure for refusing to be confined to what an Indigenous filmmaker is supposed to make. Barclay anticipated the objection. His emphasis on interior values and on the processes of production, rather than on a checklist of forms, was intended to avoid defining Fourth Cinema by surface features. A Fourth Cinema film could be a comedy, a thriller, or a documentary. What mattered was its relation to the community it came from. Ginsburg's embedded aesthetics makes the same move. Neither framework claims that all Indigenous films look alike. A second objection runs the other way: that concepts like visual sovereignty are too accommodating, treating any Indigenous presence in mainstream media as a form of resistance. Raheja's readings of Native actors in Hollywood westerns, for example, have been questioned by critics who ask how much agency a performer really exercised within a production designed to reproduce stereotypes. Raheja's reply is that sovereignty is always exercised under constraint, and that to deny any agency to Native performers would be to repeat the colonial gaze's refusal to see them as actors in their own history. The disagreement is not easily resolved, and it matters for how we judge contemporary work made within commercial studios and streaming platforms, discussed in Chapter 9. A third objection concerns the category of the Indigenous itself. The peoples grouped under the term have different histories, legal situations, and relations to the states that surround them. The Sámi of northern Norway, Sweden, Finland, and Russia, the Aymara of Bolivia, where Indigenous people form a large share of the population, and the Aboriginal peoples of Australia, who are a small minority, face different political conditions. A theory of Fourth Cinema developed in Aotearoa, where the Treaty of Waitangi gives Māori claims a distinctive legal basis, may not travel without adjustment. The international networks through which Indigenous filmmakers now meet, including festivals such as imagineNATIVE in Toronto and the Sundance Institute's Indigenous program, have fostered a shared vocabulary, but that vocabulary has to be translated back into each local situation. From theory to craft These frameworks agree on one point that governs the rest of this book: the Indigenous difference in cinema is not only a matter of what is shown but of how. Barclay's camera on the shore, the Warlpiri pan across country that tracks an ancestral path, Ginsburg's work judged by the propriety of its making, and Vizenor's refusal of victimry all locate sovereignty in decisions of form and process. The next three chapters take up those decisions in turn, beginning with the most immediate of them: where the camera stands, how it frames people and land, and what it refuses to show. Hashtags: #VisualSovereignty #IndigenousCinema #CounterVisualPractices #IndigenousFilmmaking #ColonialGaze #VisualSelfDetermination #FourthCinema #EmbeddedAesthetics #Survivance #IndigenousMedia #DecolonizingCinema #IndigenousStorytelling #CommunityAuthorship #CollectiveConsent #CulturalProtocols #IndigenousLanguages #OralStorytelling #CameraPosition #IndigenousCinematography #IndigenousArchives #ArchivalRepatriation #MediaSovereignty #CommunityAccountability #CounterVisuality #FutureOfIndigenousCinema

Download the Book (PDF): Introduction In Agile software development, traditional requirements documents have been replaced by user stories. Jeff Patton's methodology of story mapping helps teams visualize the entire user journey, preventing the common trap of building a disjointed product. While the book is highly practical and conversational, students often struggle to extract a formalized, citable framework that satisfies strict university evaluation metrics for product management courses. This companion formalizes the methodology. It is explicitly written to explain User Story Mapping by Jeff Patton and Peter Economy, organizing the conversational advice into a rigorous academic model. We break down the mechanics of framing the journey, slicing out Minimum Viable Products (MVPs), and prioritizing features based on user value. Packed with clear definitions and essay prompts, this guide equips you to critically analyze modern Agile product management. The book and its author User Story Mapping: Discover the Whole Story, Build the Right Product was published by O'Reilly Media in 2014. Jeff Patton wrote it with Peter Economy, a professional business writer, and it carries three forewords, by Martin Fowler, Alan Cooper and Marty Cagan. Those three names are a useful clue to where the book sits. Fowler is one of the signatories of the 2001 Manifesto for Agile Software Development and a long-standing voice on software design. Cooper is the interaction designer who popularized personas and goal-directed design. Cagan is the product management writer and former technology executive whose book Inspired shaped how many companies think about product discovery. Patton's book stands where their three fields meet: Agile delivery, user-centered design and product management. Patton came to the method through practice rather than theory. He worked for years as a developer, product manager and consultant, and by the mid-2000s he was writing and teaching about arranging user stories spatially so that a team could see the whole of a product at once. His 2008 essay "The New User Story Backlog Is a Map" gave the practice a name that stuck. By the time the book appeared, story mapping was already in wide use, and the book set out to explain not only the mechanics but the thinking behind them. The book is written in a deliberately informal voice. It is full of anecdotes from Patton's consulting work, hand-drawn sketches, sidebars contributed by practitioners, and short, memorable maxims. Its eighteen chapters move from the big picture of a map, through the history and proper use of user stories, to the way stories are broken down, discovered, built and learned from. That style is a strength for working practitioners and a difficulty for students. The ideas are there, but they are distributed across stories and asides rather than stated as definitions, principles and procedures. A student writing an essay or sitting an examination needs the second form. Why story mapping matters Most software teams working in an Agile way keep their work in a backlog: an ordered list of items waiting to be built. A flat list has a well-known weakness. It shows what comes next but hides how the pieces relate to one another and to the experience of the person using the product. A team can finish a hundred backlog items and still ship something that does not hang together, because nobody could see, from a list, that an essential step in the user's journey had been skipped or that three features solved the same problem. Patton's answer is to lay the stories out in two dimensions. Along the top, from left to right, run the big activities a user performs, in the order a user would perform them. Beneath each activity sit the smaller steps and the details, alternatives and variations that belong to it. The horizontal axis tells the story of use; the vertical axis carries priority and detail. Once the map exists, the team can draw horizontal lines across it to decide what belongs in a first release and what can wait. The map becomes a shared picture of the product that everyone in the room helped to build. The deeper claim of the book, and the one this guide treats as its spine, concerns understanding rather than documentation. Patton argues that the purpose of stories is not to write better requirements but to help people build a shared understanding of what they are making and why. Documents can be passed from hand to hand without anyone understanding them the same way. Conversation, especially conversation anchored by pictures and cards, is how a group arrives at the same understanding. The map is a tool for that conversation. The controlling idea of this guide Stated in one sentence, the idea that organizes this guide is this: story mapping is a disciplined way of using a shared picture of the user's whole journey to decide the least a team must build to achieve the outcomes that matter, and then to learn whether it did. Each part of that sentence corresponds to a strand of Patton's argument. A shared picture refers to his case for conversation and shared understanding over handed-off documents. The user's whole journey refers to the map's backbone and narrative flow. The least a team must build refers to his insistence on minimizing output, and to his treatment of minimum viable products and release slicing. The outcomes that matter refers to his distinction between what a team produces and what changes in the world as a result. To learn whether it did refers to the discovery practices, validated learning and post-release reflection that close the loop. The guide adds its own material in three places, and marks it clearly when it does. It sets Patton's ideas beside their sources and neighbours, such as Kent Beck's and Ron Jeffries's early work on stories, Bill Wake's INVEST criteria, Eric Ries's Lean Startup, Desirée Sy's account of dual-track development, and the Jobs to Be Done tradition. It works through an extended example map for a product invented purely for teaching purposes. And it offers critical assessment of where the method is strong, where it is vague, and where it is commonly misapplied. Where the text says "Patton argues" or "the book recommends", it is reporting the source; where it says "this guide suggests" or offers a critique, the view is the guide's own. How the guide is organized The argument begins where Patton's does, with the problem of shared understanding and the difference between output and outcome, because every later technique depends on those two ideas. It then takes the map apart, layer by layer, explaining the backbone of activities and user tasks, the walking skeleton, the body of details and alternatives, and the discipline of working a mile wide before going an inch deep. Because a map is only as good as the understanding of users behind it, the guide next turns to framing: personas, user research, and the Jobs to Be Done lens. A fully worked example map follows, built step by step for an invented community-garden service, so that the abstractions can be seen operating together. With the map in hand, the guide moves to decisions about scope. It examines how releases are sliced for outcomes, what a minimum viable product is in Ries's definition and in Patton's reframing, and why the two are not quite the same thing. It then explains Patton's chess-like opening game, midgame and endgame as a strategy for building iteratively and incrementally, and how validated learning fits alongside it. The later chapters step back from the map to the stories on it. They trace the origins of user stories in Extreme Programming, set out the Card, Conversation and Confirmation model and the INVEST criteria, and weigh Patton's caveats about the popular story template. They explain his rock-breaking metaphor for epics, themes and stories, and show how acceptance criteria and story workshops turn a large idea into buildable work. The final chapters place all of this inside a larger cycle of opportunities, discovery and delivery, including dual-track development, and close with reflection and learning after release. The Conclusion then assesses the method as a whole and considers what it asks of the people who practise it. Every chapter ends with key takeaways and review questions. Several of the questions are written as essay prompts of the kind used in university product management and software engineering courses. They are designed to be answered with reference to the source book, and the best answers will go beyond summary to evaluation. A short glossary and an annotated list of further reading close the guide. Neither replaces the source. Patton's book is lively, generous with examples and often funny, and students who read it alongside this guide will find the structure here easier to hold and the stories there easier to remember. Chapter 1: Shared Understanding, Output and Outcome Every method rests on a diagnosis. Before Patton teaches anyone to arrange sticky notes on a wall, he identifies what goes wrong when teams build software, and his diagnosis is not the one students usually expect. The common failure, in his account, is not that requirements are badly written, that developers are careless, or that estimates are poor. It is that the people involved do not understand the same thing, and do not know that they do not. The second failure follows from the first: teams measure themselves by how much they produce rather than by what changes because of it. This chapter sets out both ideas, because every technique in the rest of the book is a response to them. The problem with handing off documents The traditional model of software requirements is a relay race. Someone close to the business or the customer gathers needs, writes them into a document, and passes the document to designers and developers, who build what it says. The model assumes that a sufficiently complete and careful document transfers understanding from one head to another. Patton rejects that assumption. His paraphrasable point, and one of the most quoted lines associated with the book, is that shared documents are not the same as shared understanding. The reasoning is straightforward once stated. A document is written by someone who already holds a rich mental picture of the problem: the customers they have met, the complaints they have heard, the constraints they have negotiated. The words on the page are a compressed trace of that picture. A reader decompresses the words using their own experience, which is different. Two people can read the same sentence, both nod, and walk away imagining different products. Neither notices, because the document gives them no way to compare the pictures in their heads. The divergence surfaces weeks or months later, when something is built and the author says it is not what they meant. Patton illustrates this with a simple sequence of sketches, an idea he returns to repeatedly. Two people discuss an idea and each imagines something different. When they externalize their thinking by drawing or writing it where both can see, the difference becomes visible, and they can talk about it. Through that conversation they converge on a combined picture that is better than either started with. The lesson is that understanding is built by people working together over shared, visible artifacts, not transmitted by one person to another. The book adds a second image that students find memorable. Patton compares the documents a team produces during good conversations to holiday photographs. For the people who took them, the photographs trigger vivid memories of the trip: what happened just before, who said what, why the moment mattered. For someone who was not there, the same photographs are just pictures of strangers in front of buildings. Story cards, sketches and whiteboard notes work the same way. They are highly effective reminders for those who took part in the conversation and weak substitutes for it for anyone else. From this comparison comes a practical rule Patton summarizes as "talk and doc". Teams should talk, and while they talk they should capture what they decide, on cards, sketches or photographs of whiteboards, so that the words do not simply evaporate. The documentation is not a replacement for the conversation; it is a record that helps the participants remember it and helps them pass on the context later, ideally by telling the story again rather than by forwarding the file. It is worth being precise about what Patton does and does not claim here, because students sometimes caricature the position. He does not argue that documentation is useless or that teams should write nothing down. He argues that documents cannot carry understanding on their own and that a process built on handing documents from one group to another will reliably produce misunderstanding. The remedy is to put the people who need to understand into the same conversation, and to design the documents to support conversation rather than replace it. Stories as a tool for conversation This diagnosis explains Patton's view of user stories, which the later chapters of this guide examine in detail. The standard definition of a user story is a short description of a piece of functionality from the perspective of the person who will use it. Many teams treat stories as a smaller, more fashionable kind of requirement: shorter documents, written in a template, handed off in the same way as the long ones. Patton argues that this misses the point entirely. In his account, stories get their name from how they are used, not from how they are written. A story is a prompt for people to tell each other what is needed, why it is needed, who needs it and what would count as done. The card on which it is written is a placeholder for that conversation. The story map extends the same logic from a single story to a whole product. A list of stories, however well written, still lacks the context that lets people understand how the pieces fit together. A map arranges the stories into the narrative of a user's experience, so that when a group stands in front of it they are, in effect, telling the product's story together. Gaps, overlaps and disagreements show up as physical features of the wall: an empty column, two cards that say the same thing, a card someone moves back and forth. The map is shared understanding made visible. Patton also puts the point in terms of what a team's job really is. The job, he argues in one of his most direct passages, is not to get the requirements right. It is to change the world, at least the small part of it the product touches, for the people who use it. Requirements matter only as a means to that end. This reframing leads directly to the second idea of the chapter. Output, outcome and impact Patton distinguishes three things that are routinely confused. Output is what a team produces: features, screens, reports, releases, lines of code. Output is easy to count and easy to celebrate. Velocity reports, release notes and feature lists all measure it. Outcome is what happens after the output reaches people: how their behaviour, experience or capability changes. An outcome is observed in the world, not in the codebase. Customers complete a task they previously abandoned; support staff spend less time on a certain kind of call; users return to the product weekly rather than monthly. Impact is the longer-term consequence of outcomes for the organization that built the product and for its wider environment: more revenue, lower costs, greater market share, a stronger reputation, or a public service that reaches more of the people it is meant to serve. The relationship among the three is causal but not guaranteed. Output is the only one a team directly controls. Outcomes depend on whether people actually use and value the output. Impact depends on whether the outcomes, at scale and over time, matter to the organization. A team can deliver large amounts of output with no useful outcome, and outcomes that are welcome to users can still fail to produce the impact the organization needed. The three levels, and the questions each asks, are summarized in Table 1. Table 1. Output, outcome and impact compared. Level What it is Who controls it Typical measure Question it answers Output What the team builds The team Features shipped Did we build it? Outcome Change in what people do Users, influenced by the team Task completion, adoption Did it change behaviour? Impact Longer-term organizational effect The market and the organization Revenue, cost, retention Did it matter to the business? From the distinction follows the maxim most often attributed to the book: teams should minimize output while maximizing outcome and impact. The reasoning is economic. Patton stresses that there is always more to build than there is time or money for. Every feature carries a cost to build and a continuing cost to maintain, test, document and support, and every feature adds complexity that makes the product harder to use and harder to change. If two plans achieve the same outcome, the one that builds less is better. The aim is not to build less for its own sake but to find the smallest amount of output that produces the outcomes that matter. This is a genuinely different stance from the one many organizations take, in which success is declared when the planned scope has been delivered on time and within budget. A project can meet all three of those conditions and still fail, because nothing it delivered changed what anybody did. Conversely, a team that delivers a fraction of the originally imagined scope can succeed spectacularly if the fraction it chose was the fraction that mattered. Why the distinction is hard in practice The guide adds several observations about why teams find this distinction difficult to act on, even when they accept it in principle. First, outcomes arrive late. Output can be counted at the end of every sprint; changes in user behaviour may take weeks or months to appear, and they are tangled up with other influences such as marketing, seasonality and competitors. Management systems that need a number every fortnight will drift toward output. Second, outcomes require a hypothesis. To say that a feature should produce a particular outcome is to make a prediction that might be wrong. Many organizations are uncomfortable with plans that openly admit uncertainty, and prefer the apparent certainty of a scope list. Third, output is how individuals are often recognized. A developer who closes many tickets or a product manager who ships many features can point to visible evidence of work. It takes a deliberate change in incentives to reward the person who argued successfully that something should not be built. Patton's method does not remove these difficulties, but it gives teams a way to confront them. A story map with outcome-based release slices, discussed in later chapters, forces a group to state for each slice what it expects to change and for whom. That statement becomes the basis for later measurement. A case in point: HealthCare.gov A widely reported public example illustrates what happens when output is delivered in a single large batch without a shared understanding of the whole user journey or early evidence of outcomes. HealthCare.gov, the United States federal website through which people were to shop for and enrol in health insurance under the Affordable Care Act, opened on 1 October 2013. The launch went badly. Users met error messages, long waits and pages that failed to load, and the account-creation step at the very start of the journey became a bottleneck. Internal notes later released by the House Oversight Committee, and reported by outlets including NBC News and NPR, indicated that only six people managed to complete enrolment in a health plan through the site on its first day. The failure has been analyzed at length by journalists and oversight bodies, and it had many causes, including contracting arrangements, divided responsibility across several contractors and the government agency that managed them, late changes in policy decisions, and too little end-to-end testing before launch. This guide does not claim that story mapping would have prevented it. The case is useful for a narrower reason. It shows, at a very large scale, several patterns Patton warns against. The work was divided among many parties who each delivered their part, so a great deal of output existed, but the journey as a user experienced it, from creating an account to choosing a plan, had not been proven to work end to end. The measure that mattered to the public, whether people could actually enrol, was the outcome, and in the first days it was close to zero despite the scale of the output. The recovery is also instructive. The rescue effort that followed in late 2013 concentrated on making the core path through the site work reliably, fixing the most damaging problems first and monitoring actual user success closely, rather than adding features. Whatever one's view of the policy, it is a vivid illustration of the difference between building everything and building the path that works. What good conversation needs If conversation is the mechanism, its quality matters, and Patton's book is full of practical advice on how to make it productive. Three conditions recur. The first is the right mix of people. Patton repeatedly recommends small working groups, typically three to five people, that combine perspectives: someone who understands the business and customers, someone who understands the user experience, and someone who understands how the thing could be built and tested. A group of this size can still talk as a group; a meeting of twenty people turns into an audience listening to a few speakers. The second is something to point at. Conversations that stay purely verbal drift, because each speaker refers to pictures only they can see. Writing ideas on cards, sketching screens, and placing both on a table or wall gives the group a common reference. Patton notes that people naturally gesture at these artifacts, move them, and group them, and that this physical interaction is part of how agreement forms. The third is a clear question. A conversation about "the requirements" has no natural end. A conversation about who the users are, what they are trying to do, and what the team hopes will change for them has a direction. This is why, as later chapters show, Patton begins every mapping exercise by framing the problem before touching a single story. Implications for the rest of the method Two conclusions from this chapter carry through the rest of the guide. The first is that story mapping is a social practice before it is a technique. The map is valuable because the people who need to understand the product build it together, argue over it and refer back to it. A map drawn by one person and emailed to the team as a file loses most of its value, for exactly the reasons Patton gives about documents in general. Students evaluating the method should notice that its benefits depend heavily on who is in the room. The second is that every decision a map supports is ultimately a decision about outcomes. The backbone describes what users are trying to do. The slices describe which outcomes a release is meant to produce. The discovery and learning practices test whether those outcomes happened. When a team uses a story map simply to organize a large list of features into a tidier shape, without asking what each slice is supposed to change, it has kept the form of the method and lost its purpose. Key Takeaways · Patton's central diagnosis is that shared documents do not create shared understanding; understanding is built through conversation over visible, shared artifacts. · Documents from good conversations work like holiday photographs: powerful reminders for participants, weak substitutes for everyone else. Hence "talk and doc". · User stories are named for how they are used, as prompts for conversation, not for how they are written. · Output is what a team builds, outcome is how people's behaviour changes, and impact is the longer-term effect on the organization. · Because there is always more to build than time allows, teams should aim to minimize output while maximizing outcome and impact. · The HealthCare.gov launch shows how large volumes of output can coexist with almost no outcome when the end-to-end journey has not been proven. Review Questions 1. Explain in your own words why Patton holds that shared documents do not produce shared understanding. What role does conversation play in his alternative? 1. Define output, outcome and impact, and give an original example of a feature that could deliver substantial output with no meaningful outcome. 2. What does "talk and doc" mean, and how does the holiday photograph comparison justify it? 3. Essay prompt: "Measuring a product team by the features it ships is not merely incomplete but actively harmful." Evaluate this claim with reference to Patton's distinction between output and outcome and to at least one real case. 4. Essay prompt: Using the HealthCare.gov launch as a case, discuss the limits of what a collaborative technique such as story mapping can achieve when organizational and contractual structures divide responsibility for a user journey. Chapter 2: The Anatomy of a Story Map A story map is simple to look at and surprisingly easy to build badly. At first sight it is just cards arranged in rows and columns. Its usefulness comes from the rules governing where each card goes, and from the order in which a group builds it. This chapter formalizes that structure. It defines each layer of the map, explains the principles that give the arrangement its meaning, and sets out the sequence Patton recommends for building one. Starting with something you already know Patton introduces mapping through an exercise that requires no software at all, and it remains the best way to understand the logic. Take a pad of sticky notes and write down, one per note, the things you did this morning from the moment you woke until you left the house: switch off the alarm, shower, get dressed, make coffee, check messages, and so on. Most people produce a dozen or two notes in a few minutes. Now lay the notes out in a line from left to right in the order they happened. Already the line tells a story; anyone can read along it and say "and then... and then...". Next, look for groups of notes that belong together and give each group a name that sums it up: getting cleaned up, getting dressed, eating breakfast, heading out. Place those summary names above their groups. Finally, think about variations. What would you do differently on a day off? What if you were out of coffee? What if you overslept by forty minutes? Put those alternatives beneath the step they replace or modify. That last question is the key. When you imagine oversleeping, you do not abandon the story. You still get cleaned up, dressed, fed and out of the door; you just take the quickest version of each. You might skip the shower in favour of washing your face, and grab a banana instead of cooking. Drawing a line across the map and putting only the fastest variants above it produces a complete, if spartan, morning. That line is a release slice in miniature, and the exercise demonstrates the book's central claim that people already know how to think this way. Patton's point, in the chapter he calls "You Already Know How", is that mapping formalizes a kind of storytelling everyone does naturally. The layers of a map A mature story map has a consistent structure. Using Patton's vocabulary, with some terms the guide defines more tightly for study purposes, it has the following layers, which Table 2 summarizes. Activities. Along the very top of the map sit the large things a user does to reach a goal, such as manage email campaigns or, in the morning example, get ready. Activities are made up of several related tasks and often have no single sequence among themselves beyond a rough order of use. They are the chapter headings of the product's story. User tasks and the backbone. Directly beneath the activities, arranged left to right in narrative order, sit the user tasks: short verb phrases describing the steps a user takes, such as create a mailing list, write a message, send a message, review results. Together, activities and the top row of tasks form what Patton calls the backbone of the map. The backbone should be readable as a continuous narrative of use. The body. Below the backbone hang the details: subtasks, alternative ways of performing a task, variations for different kinds of users, exceptions, and the features that support each step. Cards lower in a column are generally less essential or more detailed than cards higher up. This region is the body of the map. Release slices. Horizontal lines or bands drawn across the body divide it into slices. Each slice takes cards from across the width of the map and represents a coherent release or increment. The top slice is the smallest version of the product that still tells a complete story. Walking skeleton. The term comes from Alistair Cockburn, who used it for a tiny implementation of a system that performs a small end-to-end function, connecting all the main architectural parts. In mapping practice it refers to the thinnest possible slice that touches every step of the backbone, so that a user, or at least a tester, can move through the whole journey even if each step is crude. Table 2. The layers of a story map. Layer What it contains Orientation Question it answers Activities Large goal-directed groups of work Top row What are users trying to accomplish? User tasks Verb-phrase steps in order of use Second row, left to right What do they do, and in what sequence? Body Details, alternatives, exceptions Below each task, top to bottom What else could happen at this step? Release slices Horizontal bands across the body Across the full width What belongs together in one release? Walking skeleton Thinnest end-to-end slice Top band What is the least that works end to end? Two axes, two meanings The two directions of the map carry different information, and confusing them is the most common beginner's mistake. The horizontal axis represents narrative order: the sequence in which a user experiences the product. Moving a card left or right changes where it sits in the story. The vertical axis represents priority and detail: higher cards are more essential or more general, lower cards less essential or more specific. Moving a card up or down changes when it is likely to be built, not where it sits in the story. A flat backlog collapses both dimensions into one. A card's position in an ordinary backlog says only that it comes before or after other cards; it cannot say whether a card comes early in the user's journey or early in the build. The map separates the two, which is exactly why it can reveal gaps. If a column has no card above the first release line, the first release has a hole in the middle of the story. Levels of goal To decide what counts as an activity and what counts as a task, Patton borrows another idea from Cockburn, whose book Writing Effective Use Cases described goals at different altitudes. A "sea level" goal is one a user would expect to complete in a single sitting and then feel they had achieved something, such as sending a newsletter. Higher, "summary" goals span several sittings, such as running a marketing campaign over a month. Lower, "sub-function" goals are steps inside a sitting, such as choosing a font. In mapping terms, user tasks in the backbone should be roughly at sea level; activities sit at the summary level; the body holds sub-function detail and alternatives. The metaphor gives teams a practical test. If a card in the backbone describes something nobody would stop and feel satisfied about having done, it is probably a detail that belongs lower down. Principles that give the map meaning Several principles run through Patton's treatment. Stated formally, they are as follows. Narrative flow. The backbone must tell a story that can be read aloud from left to right. Patton recommends literally doing so: one person walks the map, narrating the user's experience, while others listen for places where the story jumps or stalls. The narrative test catches missing steps that a list conceals. User perspective. Cards in the backbone describe what users do, not what the system does or how it is built. Search for a flight is a user task; index flight database is not. Technical work matters, but it hangs below the user steps it supports rather than replacing them in the backbone. Mile wide, inch deep. Patton advises mapping the whole breadth of the experience before going deep into any part of it. Teams naturally dive into the first interesting step and spend an hour enumerating its details, leaving the rest of the journey unexplored. Working breadth first ensures that the scope of the whole story is visible early, which is what makes sensible slicing possible later. Depth is added afterwards, and only where it is needed. Focus on the user's goals, not the features. Every card should connect, directly or through its column, to something a user is trying to achieve. Features that cannot be placed under any user task are either serving a user the team has not yet named or serving nobody. Explore before you cut. Patton encourages groups to generate alternatives, variations and exceptions freely during the exploration stage. Questions such as "what could go wrong here?", "what else might a user do?" and "what about users who...?" are what he calls "what abouts". Only once the body is full does the team begin cutting, because cutting is only meaningful when you can see what you are choosing among. Stories are for conversation. The map is never finished and never self-explanatory. Its value lies in the discussion around it, which is why Patton favours building it in person with a small, cross-functional group. The six steps of mapping The book offers a compact sequence of steps that the guide treats as the core procedure of the method. The order matters, and each step answers a different question. 1. Frame the problem. Before writing any cards, agree who the product is for, what problem it solves for them, and why the organization wants to solve it. Without this, the map has no criteria for relevance. 1. Map the big picture. Build the backbone breadth first, telling the story of the users' experience from start to finish. Keep it a mile wide and an inch deep. 2. Explore. Fill in the body with details, alternatives, exceptions and ideas for different kinds of users. Invite others to challenge and add. 3. Slice out a release strategy. Draw lines across the map to identify releases, each aimed at specific outcomes for specific users. 4. Slice out a learning strategy. Identify what the team does not yet know, and slice the map to find the smallest experiments and prototypes that would reduce that uncertainty. 5. Slice out a development strategy. Within a release, slice again to decide the order in which to build, starting with the pieces that make the product work end to end and that carry most technical risk. Students should note that steps four, five and six are all slicing, but with different purposes: the first slices for outcomes in the market, the second for knowledge, and the third for the practical sequence of construction. Later chapters take each in turn. Maps at different scales, and their neighbours Patton opens the book not with definitions but with a story about Gary Levitt, the founder of the email marketing service Mad Mimi, who came to him with an ambitious idea for a new product. Working through the idea together on a map let both of them see its whole shape at once, and seeing the whole made it far easier to recognize how large it really was and how much of it could wait. The anecdote establishes a theme that recurs through the book: the first practical benefit of a map is often not a better plan but a clearer view of how much is being proposed. That example involves a whole product, but maps work at several scales. A team can map an entire product to plan its first release, map a single new capability within an existing product, or map one complex workflow to understand it before redesigning it. The structure is the same at every scale; what changes is the altitude of the cards. On a whole-product map a backbone task might be manage subscribers; on a feature map the same idea might be the whole map, with backbone tasks such as import a list, clean duplicates and segment by interest. Students should expect to see maps nested in this way in practice, and should be clear about which scale a given map is working at. The guide also notes that story maps have close neighbours in the design world, and the differences are instructive. A customer journey map, widely used in service design and user experience research, describes a customer's experience across all touchpoints with an organization, often including emotions, pain points and channels, and usually describes the current state. A user story map describes what a product will let users do and organizes the work of building it; it is oriented toward future decisions about scope. The two are complementary. A journey map produced during research often supplies the understanding from which a story map's backbone is drawn, and a team that has one should use it rather than reinvent it. Physical and digital maps Patton's book, written before remote work became common in software teams, strongly favours physical maps built on walls or tables with paper cards. The advantages he describes are real: everyone can touch and move cards, several people can work at once, and the map remains visible between sessions as a constant reference. The guide adds that most teams today, many of them distributed, build maps with online whiteboard tools or with story-mapping features inside backlog management software. These tools preserve the two-dimensional layout and make maps easy to share and keep. They also bring risks that follow directly from Patton's diagnosis. A digital map can easily become a document that one person maintains and others glance at, rather than a shared artifact the whole group builds together. It can grow without limit, because there is no wall to run out of, and so lose the discipline that physical space imposes. Teams using digital maps tend to get the most from them when they build them together in live sessions, with several people editing at once and the narrative spoken aloud, rather than asynchronously. Common mistakes Four errors appear often enough in practice to be worth naming. The first is building the map from the system's point of view, with columns such as database, API and interface. This produces an outline of the architecture, not a story map, and it cannot reveal gaps in the user experience. The second is going deep too early, so that the first two columns are richly detailed and the rest barely exist. The resulting slices are lopsided. The third is treating the map as a one-off planning artifact that is filed once releases are agreed. Patton describes maps as living tools that teams return to as they learn, reshaping them as understanding improves. The fourth is mapping one undifferentiated user. Most products serve several kinds of people whose journeys diverge, and a map that ignores this hides important decisions. The next chapter addresses how to frame those users properly. Key Takeaways · A story map arranges activities and user tasks left to right in narrative order to form a backbone, with details, alternatives and exceptions hanging below as the body. · The horizontal axis carries the order of the user's experience; the vertical axis carries priority and level of detail. · The walking skeleton, a term from Alistair Cockburn, is the thinnest slice that works end to end across the whole backbone. · Patton's principles include narrative flow, user perspective, "mile wide, inch deep" breadth-first mapping, and exploring freely before cutting. · The six steps are: frame the problem, map the big picture, explore, then slice for release, for learning, and for development. Review Questions 1. Describe the morning-routine exercise and explain what each stage of it teaches about story mapping. 1. Distinguish the meaning of the horizontal and vertical axes of a story map. Why can a flat backlog not show both? 2. Using Cockburn's goal levels, explain how a team might decide whether a card belongs in the backbone or in the body. 3. What does "mile wide, inch deep" mean, and what failure does it prevent? 4. Essay prompt: Critically assess the claim that physical story maps are superior to digital ones. Draw on Patton's argument about shared understanding and on the realities of distributed teams. Chapter 3: Framing the Journey: Users, Personas and Jobs to Be Done A story map tells the story of someone using a product. If the team does not know who that someone is, or what they are trying to achieve, the map can still be built, but it will tell a plausible fiction rather than a true story. Patton's first step, framing the problem, exists to prevent this. This chapter formalizes framing as three linked tasks: stating the business intent, understanding the people involved, and understanding what those people are trying to get done. It also introduces two tools from outside the book that students are likely to meet alongside it: personas, which Patton uses, and Jobs to Be Done, which the guide adds as a complementary lens. Framing the idea In the part of the book that deals with discovery, Patton describes framing as the first of four essential steps: frame the idea, understand customers and users, envision the solution, and minimize and plan. Framing is about the idea's purpose from the organization's point of view. The guide formalizes it as a small set of questions a team should be able to answer, in writing, before it begins to map: · What is the idea? A short statement of the product or capability, in plain language. · Why build it? The business problem or opportunity: what the organization hopes to gain, and why now. · Who is it for? The kinds of customers and users it serves, named specifically. · What problem does it solve for them? The difficulty or desire on the users' side that the idea addresses. · How will we know it worked? The outcomes and impact that would count as success, stated in terms that could, at least in principle, be observed. These questions echo the output, outcome and impact distinction from Chapter 1. The "why" is about impact; the "problem it solves" is about outcome; the idea itself is only a proposed output. A team that cannot answer the last question has no basis for later deciding what to cut, because it has no criterion by which one slice is better than another. Patton stresses that framing is a group activity. Writing the answers down together exposes disagreement early. It is common for a product owner and an engineering lead to discover, at this stage, that they believe the product is for different people. Customers, users and other stakeholders Patton is careful to distinguish several roles that everyday language blurs. A customer is someone who chooses to acquire the product, often by paying for it. A user is someone who actually operates it. In consumer products the two are often the same person, but in business software they frequently differ: an operations director chooses a scheduling system, and shift supervisors and nurses use it every day. The two groups care about different things. The customer cares about cost, compliance and reporting; the user cares about speed, clarity and not being interrupted. A map built only from the customer's perspective will miss the daily experience that determines whether the product is adopted. Beyond customers and users are other stakeholders: people inside and outside the organization whose interests the product affects, such as support staff, sales teams, regulators and partners. Some of them never touch the product, yet their needs still create cards on the map, typically in the body beneath the user tasks that generate their concerns. For mapping purposes, the practical rule is to name user types explicitly and to decide which of them the map is primarily about. Patton recommends choosing a focus. A product can serve many kinds of users, but a first release that tries to serve all of them equally tends to serve none of them well. Deciding whose story the backbone tells is itself a prioritization decision. Personas A persona is a short, concrete description of an archetypal user, written as if describing a real person: a name, a role, a context, goals, frustrations and typical behaviour. The technique was popularized by Alan Cooper, whose 1999 book The Inmates Are Running the Asylum argued that software designed for an abstract "user" ends up designed for nobody, and that designing for a specific, well-understood archetype produces better products for the real people that archetype represents. Cooper's foreword to Patton's book reflects that shared ground. Patton uses personas pragmatically. Rather than waiting for a large research programme to produce polished documents, he describes teams sketching simple personas together, on flip-chart paper or cards, from what the group already knows about its users. A useful lightweight persona typically records: · a name and a sketch or photo to make the person memorable; · the person's role and context of use; · what they are trying to achieve with the product; · the pains or frustrations that make the problem worth solving; · what the team believes would delight or reassure them. Such personas are hypotheses. Written from the team's current beliefs, they capture assumptions that may be wrong, and the act of writing them makes those assumptions visible and testable. The designer Jeff Gothelf, in Lean UX (2013), calls this kind of persona a "proto-persona" and recommends validating it with research. Patton's advice is in the same spirit: build personas quickly, then go and meet real users to correct them. Personas help the map in two ways. First, they give the backbone a subject. Telling the story as "Maria logs in, checks her list, and..." keeps the group anchored in a specific experience rather than a generic system description. Second, they sharpen slicing decisions. When a team debates whether a feature belongs in the first release, "does Maria need this to succeed?" is an easier question to answer than "is this important?". Criticisms of personas Students should know that personas have critics. Some researchers argue that personas invented without research can entrench stereotypes, lend false authority to guesses, and become decorative artifacts that nobody consults. Others note that demographic details such as age and hobbies are often irrelevant to design decisions and can distract from the goals and contexts that actually matter. These criticisms apply mainly to personas used as finished documents rather than as working hypotheses. Patton's collaborative, lightweight approach is partly a response to them, but it does not remove the risk. A persona is only as good as the evidence behind it, and a team that never tests its personas against real users is building a map of its own imagination. Understanding users through research Patton's second discovery step is to understand customers and users. He repeatedly encourages teams to go and meet the people they are building for, observe them in their own context, and listen to them describe their work. This echoes the advice, associated with the entrepreneur and educator Steve Blank, to "get out of the building", and the book draws on the design thinking tradition associated with the design firm IDEO and Stanford's d.school, which places empathy with users at the start of the design process. The guide summarizes the most common research methods students will meet: · Interviews, in which team members ask users open questions about their goals, current practices and difficulties. The most useful interviews focus on what people actually did recently rather than on what they say they would like. · Observation and contextual inquiry, in which researchers watch people doing real work in their real environment and ask questions as they go. Hugh Beyer and Karen Holtzblatt's Contextual Design (1998) set out this approach in detail. · Analysis of existing data, such as support tickets, usage analytics and search logs, which show what people struggle with at scale. · Prototype testing, in which users try a rough version of an idea while the team observes; this belongs as much to validated learning as to research, and later chapters return to it. Research feeds the map directly. One technique the guide recommends, consistent with Patton's approach, is to map the world as it is today before mapping the product. A team builds a backbone of what users currently do to reach their goal, including workarounds, spreadsheets and phone calls, and marks the pain points. The future-state map is then built beside it, and the team can check that each significant pain has a response somewhere in the new story. Jobs to Be Done The Jobs to Be Done (JTBD) framework is not part of Patton's book, but it is widely taught alongside story mapping and addresses the same underlying question from a different angle. It is associated above all with Clayton Christensen, who developed it in The Innovator's Solution (2003, with Michael Raynor) and set it out most fully in Competing Against Luck (2016, with Taddy Hall, Karen Dillon and David Duncan). Related work includes Anthony Ulwick's Outcome-Driven Innovation and Bob Moesta's interviewing methods. The core idea is that people do not buy products for their features or because of their demographic profile; they "hire" products to make progress in a particular circumstance. The unit of analysis is the job, meaning the progress a person is trying to make in a given situation, rather than the person or the product. Christensen's best-known illustration concerns a fast-food chain that wanted to sell more milkshakes. Research found that many milkshakes were bought early in the morning by commuters who wanted something that would occupy a long, dull drive and keep them full until lunch. Understood that way, the milkshake's competitors were bananas, bagels and boredom, not other milkshakes, and improvements such as making the shake thicker and easier to buy quickly made sense in a way that generic taste tests could not reveal. JTBD has produced its own story format, the job story, developed at the software company Intercom and popularized by Alan Klement. A job story takes the form "When [situation], I want to [motivation], so I can [expected outcome]". It deliberately replaces the persona in the classic user story template with a situation, on the argument that the context of a need often predicts behaviour better than the identity of the person. Reconciling personas and jobs Students sometimes treat personas and JTBD as rival schools. The guide suggests they are better seen as answering different questions that a story map needs answered. Personas answer who the story is about and help a team empathize with a specific experience. Jobs answer what progress that person is trying to make and in what situation, which is exactly what a backbone must capture if it is to describe goals rather than features. In mapping practice, the two combine naturally. The framing step can state the primary job the product will be hired for. Activities in the backbone then correspond to the main stages of getting that job done, and user tasks to the steps within each stage. Personas identify which people perform the job and how their situations differ, which often explains the alternatives and variations in the body. Outcomes for release slices can be expressed as improvements in how well a job gets done, for whom. Writing outcome statements that can be tested The final framing question, how the team will know the idea worked, is the one most often answered vaguely. "Improve the customer experience" or "increase engagement" cannot guide a slicing decision, because almost any feature could be argued to contribute to them. The guide offers a more disciplined form, consistent with Patton's emphasis on outcomes, for students who need to write framing statements in assignments. A testable outcome statement names four things: who will behave differently, what they will do differently, by how much or how often, and how the team will observe it. For example: "Small-business owners who sign up will send their first campaign within their first week, and we will see this in our activation data." Each part constrains the design. The "who" tells the team whose story the backbone must tell. The "what" identifies the path through the map that must work. The "how much" gives a threshold for deciding whether a release succeeded. The "how observed" forces the team to plan for measurement before building rather than after. It helps to distinguish leading indicators from lagging indicators. Lagging indicators, such as annual revenue or renewal rates, capture impact but arrive too late to guide the next decision. Leading indicators, such as whether new users complete a key task in their first session, are closer to the outcome level and appear quickly. Good framing names at least one leading indicator per release, so that the team can learn within weeks rather than years. Outcome statements should also be honest about uncertainty. Patton's approach treats a plan as a set of bets. Writing "we believe that..." in front of an outcome statement is a small but useful habit: it reminds the team that the statement is a hypothesis to be tested, which prepares the ground for the validated learning discussed later in this guide. The case for a primary user Patton's advice to choose a focus deserves a fuller defence, because students often find it counter-intuitive. Organizations like to believe that a product which serves everyone will have the largest market. In practice, early products that try to serve every user type tend to be shallow for all of them: each group finds some of what it needs but not enough to change its behaviour, so the product produces output without outcome. Choosing a primary user does not mean ignoring others. It means that when the needs of different users conflict, or when time runs short, the team already knows whose success comes first. Secondary users appear on the map, typically lower in the body or in later slices. A focused first release that genuinely works for one group also creates a base from which to learn, and what the team learns from that group often transfers to others more cheaply than building for all of them at once would have done. From framing to the backbone Framing ends when a team can state its idea, its business reason, its primary users, the job those users are trying to get done, and the outcomes it hopes to see. At that point the team has what it needs to build a backbone that tells a true story rather than an imagined one. The questions stay on the wall beside the map, and the team returns to them every time it has to decide what to cut. The next chapter puts all of this to work. It builds a complete example map, step by step, for an invented product, so that framing, backbone, body and slices can be seen together. Key Takeaways · Framing answers what the idea is, why the organization wants it, who it is for, what problem it solves for them, and how success would be recognized. · Patton distinguishes customers who choose a product from users who operate it and from other stakeholders; maps should name user types explicitly and choose a focus. · Personas, popularized by Alan Cooper, are concrete archetypes; Patton favours lightweight personas built collaboratively and then tested against real users. · User research through interviews, observation, data and prototypes grounds the map in reality; mapping today's experience first exposes pain points. · Jobs to Be Done, associated with Clayton Christensen, focuses on the progress people seek in a situation and complements personas when building a backbone. Review Questions 1. List the framing questions set out in this chapter and explain how each relates to output, outcome or impact. 1. Why does Patton distinguish customers from users? Give an example of a product where the two differ and explain the consequences for the map. 2. What is a proto-persona, and what risks arise if a team never validates its personas? 3. Explain the milkshake example and what it shows about the unit of analysis in Jobs to Be Done. 4. Essay prompt: "Personas describe people; jobs describe progress. A product team needs both." Evaluate this statement with reference to story mapping, Cooper's persona method and Christensen's Jobs to Be Done theory. Hashtags: #VisualizingTheProduct #UserStoryMapping #AgileProductManagement #UserStories #SharedUnderstanding #ProductDiscovery #ProductDelivery #UserJourney #StoryMapBackbone #UserActivities #UserTasks #WalkingSkeleton #ReleaseSlicing #MinimumViableProduct #OutcomeDrivenDevelopment #OutputOutcomeImpact #ProductPrioritization #NarrativeFlow #CrossFunctionalCollaboration #TalkAndDoc #INVESTCriteria #CardConversationConfirmation #DualTrackDevelopment #ValidatedLearning #FutureOfUserStoryMapping