Where We Left Off
Part 15 put this series in the dock. Fourteen objections went in, twelve landed, one half landed, one was survivable, and none failed outright — a result the chapter itself flagged as suspicious, because a prosecution where every charge sticks is a prosecution somebody curated.
The hits were these. The series asked what the rules are for and never asked, at the same depth, what they are costing right now and who is paying. Honest uncertainty is not neutral: because the burden of proof sits silently on whoever proposes a change, every careful “we do not know” scores for the arrangement already running. Sturdy evidence and fragile evidence were written in the same voice, so a reader could not tell a census from a hundred and eighty undergraduates. The scoreboards’ columns decided their own results. And a rule that gives every side its strongest case hands more to whichever side has been studied and funded.
Then Part 15 turned the knife on the confession itself. The assumption underneath all self-criticism, it said, is that being able to state an objection shows you are not subject to it. It does not. Naming a bias is the cheapest available response and produces the same feeling as fixing one. So the part ended with four checkable commitments instead of a conclusion, and said that if this part arrived and none of them existed, you would know exactly what that chapter had been.
This part reports on them. Two were kept and two were not, and Chapter Six says which — including the one I have decided not to do at all, and what that leaves standing against me.
Everything else here is the job the last part of a series has to do. Not a summary. A reckoning: what is established rather than argued, what each side actually won, why an answer outlives the question that produced it, what a reader can do with any of this on Monday morning, and the thing this series has refused for a thousand pages — what I think, and how much weight that deserves.
How to read this document
The Six Boxes
Every part of this series uses the same six boxes. They are the notation. Here is one live example of each, so you know what you are looking at before you need it.
Synthesis: putting separate findings together to see what they add up to. It is not a summary. A summary shortens each part; a synthesis asks what is true across all of them, and often that turns out to be less than the parts appeared to promise.
Why it matters here: this part is a synthesis, and Chapter One sets a hard rule for what is allowed into it. Nothing enters because it was argued well. Something enters only if it was established.
A Word Box explains a hard word the moment it first appears. Never later, never in a footnote, never in the glossary only. If a term is used, it is explained on the page where it is used.
Sixteen parts, about 1,070 pages, roughly 324,000 words. Read straight through at ordinary non-fiction pace, that is around twenty-six hours.
The companion document published with this one — The Short Version — is about 4,000 words. Twenty minutes. If the long series is a wall, the short one is a door, and Part 15 established that the wall was keeping out exactly the people the series is about.
An In Real Terms box takes a number nobody can picture and turns it into something with a body: a walk, a wage, a week, a room. A number you cannot picture is decoration.
What a claim actually rests on, and what that kind of evidence cannot show. This part leans on the box harder than any other, because its whole job is separating what has been checked from what has been argued.
Example: the strongest single finding in this series is that reserving village council seats for women in India, rotated by village serial number, changed how villagers judged female leaders and closed part of the gap in what parents wanted for their daughters. The evidence is a genuine randomisation — the rotation was mechanical, so the villages that got a woman leader did not differ from those that did not.
What it cannot show: whether the same effect appears where a woman holds office without being seen holding it. That distinction is the whole of Part 14’s design chapter, and it is an inference from the result rather than a separate measurement.
This is the box that builds the most trust and the one most writers skip. It names the actual thing — the register, the court record, the survey, the randomisation — and then says what that thing is bad at.
The closing document of a long series is written by somebody with a strong preference about what the series turned out to have been worth.
Every earlier part could afford a disappointing chapter, because there was always another one coming. The last part has to make sixteen of them add up. That is an enormous pull towards discovering that they do.
Nothing in this part is new evidence. Every finding it collects has already been published, dated, and made available, along with the grade of the evidence underneath it. A synthesis that only recombines already-public claims can be audited line by line by anybody willing to go back.
What would settle it: somebody else writing a rival synthesis from the same fifteen parts. That is genuinely possible — the material is all public — and I would rather read that than anything else about this subject.
An Argument box is used where serious people disagree. Each side gets its strongest case, not a version that is easy to knock down. Then a verdict says where things stand and what would settle it. Often the honest verdict is that nothing available would settle it.
The signature of this series. Not a caveat and not a limitation of the study. It digs out a premise that both sides of an argument accept without noticing, because that shared premise is usually where the real trouble is.
There are five in this part. The last of them is turned on the series as a whole, and it is the one I have been avoiding since about Part 4: that “the rules about women” is one thing at all. Sixteen parts assumed a single object with a single explanation. Chapter Nine asks what follows if it is five unrelated things wearing the same face.
Five is the range the method sets. Fewer and the series loses its signature. More and they stop landing.
Every chapter ends with one of these. Three to five short paragraphs, key words in bold, the last one being the whole chapter in a sentence.
The test is simple. A reader who reads only these boxes should still come away with the complete argument. In this part, doing that would take about eight minutes.
Three warnings specific to this part
What This Document Is
It is a closing, not a conclusion. A conclusion tells you what to think. What follows is a list of what has been checked, a list of what has not, and — separately and clearly labelled — one man’s opinion at the end, with an explicit statement of how little weight it should carry.
Chapters Three and Four are verdicts, not advocacy. Part 6 argued for the old rules and Part 7 audited feminism’s claims; those were briefs. These two chapters are scoring, and I have tried to make each list as unwelcome to its own side as the evidence requires. If Chapter Three feels too kind and Chapter Four feels too harsh, or the reverse, notice which one you had that reaction to.
Chapter Nine states my own view. Fifteen parts have refused to, on the grounds that the author is a man who is not subject to these rules and carries none of the risk of following his advice. That objection has not gone away. I state the view anyway, because refusing to state it while publishing a thousand pages that plainly imply one is a way of having it both ways. Read that chapter knowing it is the least evidenced thing in the document, and that Part 15 measured this author being wrong six times out of six on his own first expectations.
1The Shape of the Whole Thing
Fifteen parts produced a great deal of argument and a much smaller amount of knowledge. This chapter sets the rule for telling them apart, and the four grades of evidence used for the rest of the document.
1.1 — What is allowed in
A final part has one temptation and it is enormous. Sixteen documents have been written. A reader has spent hours. The obvious thing to do is gather the best sentences from each and arrange them so they point at a conclusion.
That would be a highlights reel, and a highlights reel is the opposite of what this should be. The best sentences in a long series are usually the best-argued ones, and being well argued is not evidence of anything except that the writer had time.
So here is the rule for this part, and it applies to every claim from here to the end.
Nothing enters because it was argued. Something enters only if it was established — and the evidence under it gets named where it is used.
That rule is expensive. Applied honestly it throws out a great deal of this series, including several passages I was pleased with. Chapter Two is the surviving list and it is shorter than sixteen parts would lead you to expect. That shortness is the most useful information in this document.
1.2 — The four grades
Part 15 found that this series had committed a specific sin. It had written sturdy evidence and fragile evidence in the same voice, in the same typeface, in the same shape of sentence — so a reader had no way of knowing whether a number came from a national register or from a single small study nobody has rechecked. The objection landed and I could not answer it.
The repair starts here. From this point on, every empirical claim carries a grade in plain words.
| Grade | What it means | Why it sits where it does |
|---|---|---|
| Solid | A count, a register, a court record, or a genuine randomised experiment. | Nobody collected it to win this argument. Births, deaths, school rolls and judgments exist because an office had to function, and the office had no stake in what a series would later say. |
| Good | A research finding that an independent team has run again and found in the same direction. | Replication is the only real check on a study. It is rare, so a claim that has one is unusual before it is anything else. |
| Single | One study, competently done, never rechecked. | Most published findings sit here. When 100 psychology studies were rerun by a large collaboration, 36 came back significant, at about half the original size. A single study is a lead, not a fact. |
| Soft | Self-report about behaviour people are punished for, or a projection rather than a measurement. | People answer these questions strategically, and the misreporting runs in the same direction as the thing being studied. A projection is a model’s output, not an observation. |
Two things follow from that table and both matter.
First, a Soft grade does not mean false. It means the number cannot carry weight on its own. Some of the most important claims in this subject are Soft and will stay Soft, because the honest measurement cannot be taken.
Second, a Solid grade does not mean important. A birth register is beyond dispute and tells you nothing about why anybody had the child. The grades measure how much a claim can bear, not how much it is worth.
Administrative record: information a government or institution collected in order to do its job — a birth register, a school roll, a court file, a land record, a police log.
Why it matters here: this is the most reliable evidence in the entire series, and the reason is not that officials are careful. It is that the record was made for a purpose unrelated to this argument. A clerk registering a birth in 1994 was not thinking about the sex ratio. That indifference is what makes the record hard to bend.
Notice what the grading does to the balance of this series. It promotes the dullest chapters and demotes the most interesting ones. Part 9’s demography and Part 11’s legal record climb. Part 4’s effect sizes and much of Part 10’s behavioural material fall. That is not a judgment about which chapters were better written. It is what happens when you stop letting prose quality stand in for evidential weight.
Both sides of this argument — and this series, until about two pages ago — share a premise that nobody states: that “established” is a stable category. That there is a settled core of known things, and the fighting happens at the edges of it.
There is no such core sitting independently of who did the work. What counts as established is a function of what got studied, and what got studied is a function of who was paying and what they found interesting. Part 11 had to be written almost from scratch because the comparison group — men — had barely been examined. Part 13 found exactly one genuine randomisation in the whole subject.
So Chapter Two is not a list of what is true about the rules governing women’s lives. It is a list of what has been checked. Those would only be the same list in a world where somebody had tried to check everything, and no such attempt has ever been funded.
The general form: a map of the known is also a map of the interests that paid for the survey. The blank areas are not where the ground is featureless. They are where nobody sent anyone.
That box is the reason Chapter Eight exists. A synthesis that lists only what is known is telling you half a story; the other half is the shape of what nobody looked for, and that shape is informative on its own.
1.3 — The map
Before the findings, the terrain, in one pass. If you have read all fifteen parts you can skip this. If you arrived at the last one first — which some people do, and it is a reasonable thing to do — this is the minimum you need.
Parts 1 to 3 asked where the rules came from. What the question is really asking, how the rules arose historically, and how they are transmitted and enforced in India specifically.
Parts 4 and 5 asked about human beings and about the countries that changed. What sex differences are actually measured to be, with the caveats that an average difference between groups licenses almost nothing about any individual. Then the countries that dismantled the rules, where outcomes matched nobody’s predictions.
Parts 6 and 7 put each side’s case at full strength. Six argued for the old rules as an advocate would. Seven audited what feminism actually claims, as an auditor would. Both scored themselves at the end.
Parts 8 to 10 counted. A ledger of the equality project with a column asking who each change reached. The arithmetic of birth rates and the pronatalist argument. And the core traditionalist warning — that a woman who breaks these rules ruins her own life — unpacked into four claims and tested.
Parts 11 to 13 looked where nobody looks. Men, as the missing comparison group. Enforcement once a community loses its edges and the phone becomes both the exit and the weapon. And the handful of times history ran something like an experiment.
Parts 14 and 15 asked what works and then what was wrong with the asking. Four intervention shapes scored, and then the whole series put in the dock.
Fifteen parts, about a thousand pages, roughly three hundred thousand words. Out of all of that, Chapter Two lists eleven findings.
That is about one established finding for every ninety pages. Put another way: if you read this series at a steady pace for a full working day, you would collect roughly one thing that is properly known.
This is not a complaint about the series. It is what research on a contested social question actually yields, and the reason nobody says so is that a book which announced “eleven findings” on its cover would not be read.
1.4 — Who is speaking, one last time
The method this series uses says to declare your position wherever it could pull the writing, at the point where it applies rather than in a preface nobody reads. This is the last time, so let me be complete.
I am an Indian man writing about rules that fall mostly on Indian women. I have never had my clothing, movements or friendships negotiated by anyone. I wrote the fifteen parts this document is grading, which means I am marking my own work, and Chapter Six is a report on promises I made to myself.
I publish under my own name on a site I own. No funder, no editor, no commission — which removes one kind of pressure and adds the pressure of having nobody to tell me a chapter is bad.
And there is a specific pull on this part that did not apply to the others. A final document has to justify the whole. Every incentive in my position points towards discovering that sixteen parts were worth writing. Where you find this chapter reaching for a conclusion that makes the series look necessary, distrust it, and check the grade attached to whatever it is claiming.
The rule for this part: nothing enters because it was argued. Something enters only if it was established, and the evidence gets named where it is used.
Four grades from here on. Solid means a count, a register, a court record or a randomised experiment. Good means a study an independent team has run again. Single means one study, never rechecked. Soft means self-report about punished behaviour, or a projection rather than a measurement.
Soft does not mean false, and Solid does not mean important. The grades measure what a claim can bear, not what it is worth.
Underneath all of it: “established” is not a stable category. The list in Chapter Two is a list of what has been checked, and what got checked was decided by who was paying.
Fifteen parts and about a thousand pages produced eleven findings — roughly one for every ninety pages, which is what honest research on a contested social question actually yields.
2What Is Actually Established
Eleven findings survive the rule set in Chapter One. Each one has its evidence named and graded. Then the shorter, more uncomfortable list of things almost everybody believes that did not survive.
2.1 — The eleven
These are in no order of importance. They are grouped by what they are about: how the rules work, what they do to a woman, and what happens when somebody tries to change them.
One. The rule falls where it can be enforced, not where belief points. A household can reach a daughter who lives in it. It cannot reach a son who has moved to another city for work. Across the material in Part 11, the pattern of who gets policed tracks physical reach far better than it tracks anything anyone says they believe about the sexes. Grade: the counts underneath are Solid — censuses and household surveys — but the reading of them is an inference, and a rival reading exists. It is the most important claim in this series and it is not proved.
Two. Daily enforcement is largely administered by women on women. In the Indian household survey data, a mother-in-law’s view of contraception predicts a couple’s use of it better than the wife’s own view does. The lethal end of enforcement is a different matter and is usually delegated to the most legally expendable male in the family. Grade: Good. Large national surveys, and the pattern appears across more than one South Asian dataset.
Three. The stated function was solved thirty years ago and nothing was retired. Certainty about who fathered a child is the reason the tradition itself gives, most often and most explicitly, for rules on female sexual conduct. A cheap and conclusive test for that has existed since the 1990s. No society has dismantled any part of the apparatus in response. Grade: Solid on both halves — the availability of the test is a fact about the world, and the absence of any retirement is an absence anybody can check. This is the single most damaging finding in the series for the traditionalist case, and Chapter Three treats it at length.
Four. Of the harms predicted for a woman who breaks the rules, some travel and some do not. Infection risk travels completely — it belongs to the act, and it shows up wherever you look. Damage to her standing in the marriage market does not travel at all. It appears where a community punishes it and vanishes where one does not, which means it is a fact about the community rather than about her. Grade: Solid for the infection half. Solid for the non-travel, because it rests on the presence of a penalty in some settings and its absence in others, which is a comparison rather than a study.
Association: two things that show up together. Taller people earn more; that is an association.
Cause: one thing making the other happen. Height does not make money.
Why it matters here: almost every claim made in public about women’s conduct is an association being reported as a cause. The test in this series has been the travelling test — if an effect belongs to the act, it appears everywhere the act appears. If it appears only where a community punishes the act, the community is the cause and the act is the occasion.
Five. The wellbeing effect reverses with the community’s belief. Where casual sex is disapproved of, women who have it report worse wellbeing. Where it is not, they do not. The direction of the effect is set by the surrounding judgment, not by the behaviour. Grade: Single to Good — more than one study points the same way, and none is large. This is one to hold loosely.
Six. Banning the output fails, every time it has been tried here. The 1929 act raising the age of marriage produced a rush of marriages before it commenced. Dowry prohibition did not end dowry. The ban on using ultrasound to find out a baby’s sex was followed by a child sex ratio that fell from 962 girls per 1,000 boys to 918. Each time, the demand stayed and the behaviour moved. Grade: Solid. These are administrative counts and census figures, collected by offices with no stake in this argument.
Seven. Putting women visibly in authority changes what people think about women in authority. In 1993 India began reserving village council leadership for women, rotated by village serial number. Because the rotation was mechanical, the villages that got a woman leader did not differ beforehand from those that did not — which makes this a real experiment rather than a comparison. Measured bias against female leaders fell. More women subsequently won seats that were not reserved. The gap between what parents wanted for their sons and for their daughters closed by around a quarter, and the gap in what adolescent girls wanted for themselves closed by about a third. Grade: Solid. This is the only genuine randomisation in the entire subject.
Randomisation: deciding who gets something by lot rather than by choice. It sounds careless and it is the most powerful tool in research, because groups picked at random do not differ beforehand in any systematic way. So if they differ afterwards, the thing you gave one of them did it.
Why it matters here: almost nothing in this subject has been randomised, because you cannot assign people to a culture. India’s 1993 rotation of reserved seats by village serial number was not designed as an experiment. It simply happened to work like one, which is why a single administrative convenience is carrying more evidential weight than the rest of the literature combined.
Eight. The damage from son preference concentrated among the richer, urban and educated. Where a family had money, a city, and schooling, the child sex ratio was worse, not better. The people best placed to act on a preference did so most effectively. Grade: Solid. Census data.
Child sex ratio: the number of girls for every 1,000 boys among young children. Because roughly equal numbers of each are born everywhere, a figure well under 1,000 means girls are going missing — before birth, or through worse feeding and slower treatment afterwards.
Why it matters here: 962 falling to 918 sounds like a small movement. Across a population of India’s size it is millions of girls, and it is one of the few measurements in this whole subject that nobody disputes.
Nine. Persuasion campaigns, as actually run, spend most of their money on themselves. India’s flagship campaign for the girl child released ₹446.72 crore to states between 2016 and 2019. A parliamentary committee reported in December 2021 that 78.91 per cent of it went on media advocacy. Grade: Solid, and unusually so — it is a legislature auditing its own government’s spending and finding against it, which is the direction institutional evidence almost never runs. The caveat is in the text of Part 14: the sex ratio did improve over the period, and attribution is contested.
Ten. No country has reversed a fertility decline. Not one, anywhere, has brought a birth rate back up to replacement once it fell well below. South Korea has spent more than two hundred billion dollars and moved from 0.72 to 0.80. Hungary spent around 5 per cent of its national output, reached 1.59 in 2021, and is back to roughly 1.31. Grade: Solid. National vital statistics.
Fertility rate: the average number of children a woman would have across her life at current rates. It is not a count of anybody’s family — it is a rate, worked out from births in a single year.
Replacement level is about 2.1. Slightly more than two, because not every child reaches adulthood. Below that, a population shrinks over time unless people move in.
Why it matters here: South Korea at 0.72 means that for every hundred people in one generation there are about thirty-five in the next. That is not a slow decline. It is a halving every generation, and it is the strongest single fact the traditionalist has.
Eleven. Two independent sets of criminals have priced the double standard. Blackmail using intimate images splits into two patterns. Girls are extorted for more images or for compliance. Boys — around nine in ten victims of the financial version are boys aged fourteen to seventeen — are extorted for money. The reason is commercial: a threat against a boy loses its value within days, and a threat against a girl does not. Two sets of people who have never met, working only from what pays, have independently measured the asymmetry. Grade: Good. Law-enforcement and child-protection analyses, converging from more than one source.
Look at what the strongest items on that list have in common. Not one of them is a study of attitudes.
The Solid items are: a register of births, a census, a parliamentary audit of spending, national vital statistics, and one randomisation that happened because an administrator needed a rule for rotating seats and chose village serial number. Every one of them was produced by an institution doing something else.
Independent sources converging is the other thing to look for, and item eleven is the cleanest example in the series. Two separate criminal enterprises, with no contact and no ideology, arrived at the same pricing of the same social fact. When a farmer’s wallet, a government meter and a satellite all say the same thing, the thing is true — and when two unconnected groups of blackmailers price a norm identically, the norm is real.
What this cannot show: any of the reasons. Administrative records are excellent at what happened and silent about why. Every explanatory sentence in this series is softer than the counts it is built on, and Chapter Five is nothing but explanatory sentences.
2.2 — What did not survive
Now the shorter list, and it will annoy both sides.
That a woman’s number of partners before marriage causes her marriage to fail. The association exists in American survey data. It is not a straight line — the divorce rate does not simply rise as the number rises — and the researcher who assembled the best version of it has published that no proposed mechanism explains its shape. It has essentially not been tested outside the United States. And the partner count is self-reported, on a subject where reporting is known to bend. Grade: Soft. It cannot carry the weight the argument puts on it, in either direction.
That the rules exist because of a belief about women’s nature. Finding one is the rival to this, and while finding one is not proved, this one is in worse shape. If belief about women drove the rules, the rules should fall on women wherever they are. They fall on women who can be reached.
That enforcement is done by men to women. The lethal end is. The daily end mostly is not, and any campaign built on the assumption that the obstacle is fathers and brothers has misidentified who administers the thing it is trying to change.
That education alone dissolves it. Finding eight kills this. The most educated families produced the worst sex ratios. Education gives people better tools; it does not tell them what to want.
That awareness changes behaviour. Finding nine. Not weakly supported — actively contradicted by the largest test anybody has run.
That the rich Western countries show where this ends. Part 5 is the whole argument here. The countries that dismantled these rules did not produce the outcomes either side predicted. They are not a preview. They are a set of separate cases with their own histories, and using them as a forecast fails five of the six tests Part 13 set for a valid comparison.
Six beliefs, on that second list, that a very large number of people on both sides hold firmly.
Put it as a wager. If you had to bet your own money on those six, at even odds, on the evidence currently available, you would decline all six bets. Most public argument about this subject consists of people taking those bets loudly and for free, because nobody ever settles up.
2.3 — What the shape of the list tells you
Stand back from both lists and something becomes visible that no individual finding shows.
The claims that survived are about structures: who can reach whom, what an office does when a woman is visibly in it, what happens to a birth rate, what a ban does to a demand, what a blackmailer can charge. The claims that did not survive are about individuals and their beliefs: what a woman’s history does to her marriage, what people think about women’s nature, what happens when you educate someone or tell them something true.
That is not a coincidence and it is not only a fact about the world. It is partly a fact about measurement. Structures leave administrative records; beliefs leave surveys. Chapter One’s hidden assumption said the map of the known is also a map of who paid for the survey, and here is that assumption cashing out: the evidence is strongest exactly where states happen to count things.
So the honest version of the pattern is a double one. Structural explanations are doing better and structural explanations are easier to check. Both are true. Anyone who tells you they can separate those two cleanly, in either direction, is telling you something nobody currently knows.
Eleven findings survive. The strongest are that the rule falls where it can be reached, that daily enforcement is largely women on women, that the stated paternity function was solved thirty years ago and nothing was retired, that banning an output has failed every time, and that visible female authority changes what people believe — the only genuine randomisation in the subject.
Six widely held beliefs did not survive: that partner count causes divorce, that the rules come from a belief about women’s nature, that men do the enforcing, that education dissolves it, that awareness works, and that the West is a preview.
Almost every Solid item was produced by an institution doing something else — a birth register, a census, a spending audit, a rule for rotating seats. Nobody collected it to win this argument.
Claims about structures survived. Claims about individuals and their beliefs did not. That is partly the world and partly the fact that structures leave records while beliefs leave surveys, and nobody can currently separate the two.
A thousand pages produced eleven findings, and the shortness of that list is the most useful thing in this document.
3What the Traditionalist Won
Not advocacy this time — scoring. Four claims survived and four did not, and one of the four that did not costs the tradition more than the other seven put together.
3.1 — Four that survived
Part 6 of this series argued the traditionalist case as an advocate would. This chapter does something colder. It asks which of those arguments is still standing after nine more parts of evidence, and it does not do the tradition the discourtesy of grading it generously.
One. The family really is the insurance system where there is no other. Son preference has a function, and the function is not superstition. In a country where most people reach old age without a pension worth living on, an adult son is the retirement plan. Daughters marry out and their earnings and care go elsewhere. Everything unpleasant that follows — the dowry, the sex ratio, the differential feeding — sits on top of a real financial problem that nobody has solved. Grade: Solid. Pension coverage and old-age support patterns are administrative facts.
Two. The reforming state’s tools do not work, and the traditionalist said so first. Every case in this series where a state banned an output failed. Every persuasion campaign examined either failed or spent its money on itself. The traditionalist’s long-standing scepticism about the competence of the modernising state to reorganise family life by statute has been vindicated more thoroughly than any positive claim he makes. Grade: Solid. Administrative counts and a parliamentary audit.
Three. Nobody has reversed a fertility decline, and the counter-programmes have failed. The traditionalist warning that a society reorganised around individual choice stops replacing itself has not been refuted. It has been confirmed as a description, everywhere it has been tried, and no state has found a way out with money. Grade: Solid on the fact. Not established on the cause, which matters enormously and is taken up below.
Pronatalism: government policy aimed at getting people to have more children — cash payments, tax breaks, housing loans forgiven per child, sometimes restrictions on contraception or abortion.
Why it matters here: it is the policy response to the finding in this section, it has been tried at enormous expense in several countries, and Part 9 found it moves real numbers by small amounts that stop when the money stops.
Four. Community enforcement really does coordinate behaviour. This is the one reformers concede least and should concede most. A shared code that everybody expects everybody to follow makes behaviour predictable, lowers the cost of trusting a stranger inside the group, and lets families make long commitments — a marriage, a loan, a joint household — with people they cannot individually vet. That is a real service and it is not obviously replaceable. Grade: Single at best as a measured claim, but it is structural rather than behavioural, and Part 12 showed what happens when the coordination breaks: the enforcement does not stop, it goes somewhere worse.
Coordination problem: a situation where what is best for you depends on what everybody else does, so nobody can improve things by acting alone.
Everyone driving on the same side of the road solves one. Any single driver who switches, on the grounds that the other side is just as good, is not making a point. He is causing a crash.
Why it matters here: it is the strongest structure available to the traditionalist, and Chapter Five shows it is also the reason the rules survive long after their reasons have gone. A coordination system does not need to be right. It only needs everybody to expect everybody else to keep following it.
3.2 — Four that did not
One. The paternity argument. Certainty about fatherhood is the reason the tradition offers most often and most explicitly when asked for a reason. A cheap, conclusive test has existed for three decades. Nothing was retired anywhere. This is dealt with in full in section 3.3, because it is not simply a failed argument — it is a failed argument that changes the status of all the others.
Two. “She ruins her own life.” Part 10 unpacked this into four separate claims and tested them. Infection risk is real and belongs to the act. Damage to her standing in the marriage market is real and belongs to the community — it appears where a community punishes it and vanishes where one does not. The wellbeing effect runs in whichever direction the surrounding judgment runs. The divorce association is soft, unexplained and untested outside America. What is left of the warning is not a prediction about her. It is a description of what her neighbours will do, stated as though it were a law of nature.
Three. That the rules reflect a fact about women. If they did, they would fall on women wherever women are. They fall on women who can be reached, which is not the same population. A daughter at home is governed; a son in another city is not, and no one seriously claims he is more trustworthy. The rule is shaped like a reach map, not like a theory of the sexes.
Four. That this is one coherent moral order. Part 11 counted the tradition’s own stated goals and found that four of six point at male conduct — at what men would do if unrestrained, at what men demand in a bride, at what men are owed. One points at the rule itself. Exactly one points at her, and that is paternity, which is finding one. A moral order that names male conduct as the problem four times out of six, and then builds its entire enforcement apparatus around female movement, is not a coherent order. It is an order that fell where it could land.
3.3 — The concession that costs the most
Now the thing that does the real damage, and I want to be careful with it, because it is easy to overstate.
The claim is not that the tradition lied about paternity. Part 15 put the strongest reply to this on the page, and the reply is good: a practice held as a duty may still be able to offer a practical reason to a practical questioner, and being able to produce a reason is not the same as holding the practice for it. Ask a man why he does not read his wife’s letters and he may say it would damage trust. Solve the trust problem with a machine and he will still not read them.
That reply works. But it costs something enormous, and the tradition rarely notices the bill.
If the stated reason was never the operative reason, then the tradition’s public arguments are not its actual grounds. Every practical justification offered in a courtroom, a newspaper column, a school assembly or a family argument — she will be hurt, the family will suffer, society will fray — is now revealed to be the kind of thing offered rather than the kind of thing held. And once one stated function has been shown to be non-operative, the burden moves: which of the others is real, and how would anybody tell?
The evidence has an unusual shape. It is not a study. It is the absence of a response to a technology.
The facts are: reliable paternity testing became cheap and widely available in the 1990s; it is used routinely in courts and in disputed maintenance cases; and no society, community or family authority anywhere has relaxed any rule on female conduct on the grounds that the question it was said to answer is now answerable another way.
Treat a missing response the way this series treats a missing count. When an institution had every opportunity to act and did not, the absence is a finding rather than a gap.
What this cannot show: whether the reason was insincere. Sincerity is not observable. What is observable is that the reason was not load-bearing, and that is the only claim being made.
Which brings us to the surviving claim with the largest political weight, and the one that deserves the most careful handling in the whole chapter.
Of the four surviving claims, this is the one with the largest political stake, so it deserves the full treatment.
No society that reorganised itself around individual choice has replaced itself. Not one has climbed back. South Korea spent more than two hundred billion dollars for a move from 0.72 to 0.80. Hungary spent about 5 per cent of national output, got to 1.59, and slid back to roughly 1.31. If a set of arrangements produces a population that stops existing, no ledger of other benefits can settle the account, because the ledger runs out of people to hold it.
Iran reversed the rules completely — the reformers lost, comprehensively, and the fertility rate fell anyway, along with rising female education. India’s own figures show the same thing inside one country: the national rate is about 1.9, rural India is at 2.1, Delhi is 1.2 and Bihar is 2.8. The variation tracks urbanisation, income and housing rather than piety. If restoring the rules restored the birth rate, Iran and urban India would be the proof, and they are the opposite of it.
Everyone agrees the decline is real, universal among rich and urbanising societies, and unreversed. Nobody has established what causes it. The candidates — the cost of housing, the length of education, insecure work, the collapse of unpaid childcare from relatives, the rise of women’s options — are all present together everywhere and have never been separated. Both camps are reading a correlation they like out of the one dataset everybody shares.
What would settle it: a case where the material conditions changed sharply — housing, childcare, job security — while the cultural arrangement held still, or the reverse. No such case exists, which is exactly what Chapter Eight means by an empty shelf.
Why people care so much: because a demographic argument is the only one that lets a political position claim the future is on its side without having to win an argument in the present.
3.4 — What the tradition would have to do to hold
If I were advising the tradition on how to answer this series honestly — and the method requires me to be able to, or I have not understood the position — I would say three things.
Stop making outcome predictions. Every prediction the tradition has made about what happens to a woman who breaks the rules has now been tested and most did not survive. A position that claims something is owed is not damaged by any of this. A position that claims something works has entered a competition it is losing.
Own the insurance problem instead of the modesty problem. Finding one is the strongest thing the tradition has, and it is barely used. The family really is doing work that no institution has replaced.
And answer finding four. Four of six stated goals point at male conduct. Until the tradition has an account of why the apparatus is built entirely on female movement, its own list of reasons is the strongest case against it.
Four traditionalist claims survived: the family as insurance where no state provides it; that the reforming state’s tools do not work; that no fertility decline has been reversed; and that community enforcement really does coordinate behaviour.
Four did not: paternity, “she ruins her own life”, that the rules reflect a fact about women, and that this is one coherent moral order.
The paternity failure costs most because of what the best reply to it concedes: if the stated reason was never the operative reason, then the tradition’s public arguments are not its actual grounds, and the burden moves to showing which of the remaining reasons is real.
On fertility, the traditionalist owns the fact and not the cause. Iran reversed the rules completely and the rate fell anyway. Nobody has separated housing, education, work and women’s options, because no case exists where they moved apart.
A moral order that names male conduct as the problem four times out of six, and then builds its whole apparatus around female movement, is not an order. It is a rule that fell where it could land.
4What the Reformer Won
The same treatment, applied the other way. Four claims survived and four did not — and the reformer’s failures are concentrated in exactly the activities that receive the money.
4.1 — Four that survived
Part 7 audited what feminism actually claims. This chapter asks a narrower question: after nine more parts, which of the reforming position’s claims are still standing on evidence rather than on assertion?
One. In the narrow band, reform worked, and it worked unambiguously. There is a small set of outcomes almost nobody openly defends: girls dying of neglect, women dying in childbirth for want of a clinic, children married at twelve, acid, murder. Wherever reform aimed at that band and paired the aim with delivery — clinics, midwives, registration, prosecution — the numbers moved and stayed moved. Grade: Solid. Vital statistics and administrative records.
Two. Structural intervention works where argument does not. The single strongest result in this subject is the Indian village council reservation, rotated mechanically and therefore a real experiment: measured bias against female leaders fell, more women later won unreserved seats, and the gap between what parents wanted for sons and for daughters closed by around a quarter. Grade: Solid. The reformer has been saying for decades that you change conditions rather than minds. That claim is now the best-evidenced positive finding either side possesses.
Three. The asymmetry is not justified by any fact anybody has produced. This is the reformer’s strongest analytic win and it is worth stating carefully. The claim is not “men and women are identical” — Part 4 found measured average differences and said so. The claim is that nothing in those differences accounts for the shape of the rules. The rules do not track any measured difference. They track who can be reached. And the tradition’s own stated goals point at male conduct four times out of six. Grade: the underlying counts are Solid; the reading is an inference, and it is currently the best one available.
Four. The state’s own record convicts the old arrangement. In 2018 India’s Supreme Court struck down the adultery law and said, in its own judgment, that the provision had treated a wife as her husband’s property. The exception that prevents a husband being prosecuted for rape within marriage is still in the statute book, was carried into the new criminal code, and the Union government filed against its removal in October 2024; the case has been through a change of bench and had no verdict as of early 2026. Grade: Solid. Court judgments and filed government positions are documents, not interpretations. When the highest court in the country describes the old order in those words, “the reformers are importing foreign categories” stops being available as a reply.
4.2 — Four that did not
One. That awareness changes behaviour. Not merely unsupported — contradicted by the largest test anyone has run. India’s flagship campaign for the girl child put 78.91 per cent of the money it released to states into media advocacy.
Two. That prohibition works. Raising the marriage age produced a rush of marriages before commencement. Dowry prohibition did not end dowry. Banning the sex-determination scan was followed by a child sex ratio falling from 962 to 918. Each time the demand survived and the behaviour moved. The reformer keeps reaching for the ban because the ban is the lever a legislature actually controls.
Three. That the gains reached everyone. Part 8’s ledger had eighteen rows and one column that made the difference: who did this actually reach? The answer, repeatedly, was the urban, the educated and the already-comfortable. The reformer’s ledger is an average, and an average conceals exactly the distribution the argument is about.
Four. That education dissolves the problem. The families with the most money, the most schooling and the most access to cities produced the worst child sex ratios. Education distributes capability. It does not distribute preferences.
Line up those two lists against a budget and something uncomfortable appears.
The two things that work — building the clinic, and putting a woman visibly in the chair — are slow, need staff, and hand something identifiable to somebody. The two things that do not work — passing the ban and running the campaign — are fast, cheap, announceable, and cost nobody anything they can feel.
A reforming government choosing between them is not choosing between effectiveness and ineffectiveness. It is choosing between a policy with a visible loser and a policy with none. Which is why, in almost every country, the money goes where the evidence does not.
4.3 — The concession that costs the most
For the traditionalist it was paternity. For the reformer it is the reach column, and the reason is the same shape: it is not that a claim failed, but that the best reply to it concedes something structural.
The reformer’s honest reply to “the gains were concentrated” is that a gain concentrated among urban graduates is still a gain, that no reform reaches everyone at once, and that the first cohort makes the second cohort cheaper. All of that is true.
But accept it and you have accepted that the ledger is not a measurement of the project. It is a measurement of the project at its leading edge, and the tail is unmeasured. Part 15 put this precisely: a total is a decision about whose experience is allowed to cancel out whose. When a movement reports rising literacy and falling maternal deaths as its record, it is reporting a national average, and there is no row anywhere that says what happened to the woman in the village where nothing arrived.
The traditionalist has an easy version of this objection and it is not the serious one. The serious one is that nobody knows, because the instrument that would tell you was never built, and the movement had no incentive to build it.
Chapters Three and Four are two lists. Two lists is a format, and the format carries a premise that neither camp has ever examined: that a claim belongs to a side.
Look at what that does. “Banning an output fails” appears in Chapter Three as a traditionalist win and in this chapter as a reformer loss. It is the same finding. It is not about traditionalists or reformers at all — it is a finding about bans, and it applies with equal force to a traditionalist state banning something it dislikes. Iran ran that experiment and it failed there too.
Once findings are sorted into ours and theirs, a result that helps the other camp has to be resisted rather than absorbed, because absorbing it is a defection. That is why both sides can look at the same fertility figures for thirty years and neither moves. The figures are not being evaluated. They are being defended against.
And I did this too. I wrote these two chapters as two ledgers because it is legible and because a reader can follow it. The evidence is not organised that way. It is organised by mechanism — reach, coordination, insurance, demand, visibility — and none of those mechanisms respect the boundary between the camps.
The general form: sorting evidence by whose case it helps guarantees that no evidence can ever change anybody’s mind, because a finding’s first property has become its allegiance.
That box is the hinge of this part. Chapters Three and Four are the last time this document uses the two-camp frame. Chapter Five drops it and asks the mechanical question instead, and the mechanical question turns out to have an answer that neither camp has been arguing about.
Everyone agrees reform succeeded on deaths, child marriage and violence. The fight is about what that success predicts.
Every extension of a moral consensus in history began with the case nobody could defend and moved outward. Slavery, child labour, public smoking: each started as a narrow prohibition on the indefensible and became a general change in what people found tolerable. The narrow band is not the limit of agreement; it is the wedge.
The band is narrow for a reason. Inside it, both sides want the same outcome and are only arguing about method. Outside it, they want different outcomes, because they hold different views of what a life is for. No amount of evidence resolves a disagreement about ends. Everything beyond the band is a values dispute wearing the clothes of an evidence dispute, and treating it as evidential is why fifty years of data have moved nobody.
Both are describing different things. As a matter of persuasion the band is a ceiling — nothing outside it will be settled by another study. As a matter of what actually changes, the band is a floor, because the interventions that work at the bottom of it (a clinic, a school roll, a wage, a seat) reach inside households without anybody having to be persuaded of anything. Reform advances not by winning the argument but by installing things.
What would settle it: a case where a values dispute outside the band was resolved by evidence rather than by generational replacement. I do not know of one, and I looked.
Why people care so much: because if the band is a ceiling, then most of what both movements spend their money on — campaigning, arguing, awareness — is spent outside the zone where it can work.
4.4 — What the reformer would have to do to hold
The same courtesy as Chapter Three. Three things.
Stop funding persuasion. The evidence against it is now stronger than the evidence for almost anything else in this series, and it is the largest single line in most reform budgets.
Report the distribution, not the average. Until a ledger says who a gain reached, it is a claim about the leading edge and the tradition is entitled to say so.
And take the insurance problem seriously. Son preference is a pension. Seclusion is a substitute for policing. Dowry is a property transfer. Every one of those is a function, and every reform that attacks the practice without replacing the function has failed. The reformer’s own best finding — that you change conditions rather than minds — points directly at this, and almost no reform programme is built that way.
Four reformer claims survived: reform worked in the narrow band of deaths, child marriage and violence; structural intervention works where argument does not; no measured fact about the sexes accounts for the shape of the rules; and the state’s own courts convict the old arrangement in their own words.
Four did not: awareness, prohibition, that the gains reached everyone, and that education dissolves it.
The two approaches that work are slow and hand something identifiable to somebody. The two that fail are fast, cheap, announceable and cost nobody anything they can feel — which is why the money goes where the evidence does not.
Underneath the two-list format sits the assumption that a claim belongs to a side. “Banning an output fails” is scored as a traditionalist win and a reformer loss; it is neither, it is a finding about bans. Sorting evidence by whose case it helps guarantees that no evidence can change anyone’s mind.
The argument stalls at the edge of the narrow band. The delivery does not — which means reform advances by installing things rather than by winning.
5Why the Answers Outlive the Questions
If a rule was built to solve a problem, and the problem is solved, the rule should go. It does not. This chapter drops the two-camp frame and asks the mechanical question, which turns out to have four answers.
5.1 — The puzzle, stated plainly
Here is the thing this series has circled since Part 3 and never faced directly.
Certainty about fatherhood was the stated reason. It has been cheaply answerable for thirty years and nothing was retired. A dowry is a property transfer, and daughters now inherit under the law. Seclusion substitutes for policing, and there are police. Sons are a pension, and pensions exist in some places now.
In every case the stated problem has been at least partly solved by other means, and in no case did the arrangement stand down. If these rules were answers, the answers have outlived the questions by decades.
Both camps have a story about this and both stories are wrong.
The traditionalist says the rules survive because they still work — because the modern substitutes are inferior and people can tell. The reformer says they survive because people still believe in them — so change the beliefs and the practice follows.
The first cannot explain why nobody claims the benefit when asked. The second cannot explain the surveys, which repeatedly find majorities disapproving of dowry in places where dowry is universal. Both stories require the practice to be wanted. There is a fourth possibility, and it is the one this chapter is about.
5.2 — Functions migrate
The first mechanism is the simplest and it explains the paternity result on its own.
A rule built to do one job does not stay attached to that job. It sits in a society, and societies have other jobs going spare.
Keeping a woman out of public life was, historically, a substitute for policing — a household could not call anyone, so it managed its own risk. In a city with police, that function is redundant. But the same practice now does something else: it signals that a family can afford to keep an adult out of paid work. It has become a display of wealth. Nobody redesigned it. The old machine got a new job because it was standing there.
Dowry began as a way of moving property to a daughter at marriage. Under a marriage squeeze — with more men of marrying age than women, which India’s own projections show worsening for decades — the same transfer now functions as a price for a groom in a market where grooms are scarce. Same practice. Completely different economics.
So solving the original problem does nothing, because by the time you solve it the practice is not doing that any more. You have paid off a debt that was transferred years ago.
Sunk cost: money, effort or time already spent that cannot be got back. In economics it is the thing you are told to ignore when deciding what to do next.
A sunk asset is the mirror image, and it is the useful one here: something already built, which is expensive to create and cheap to keep running. A railway line is a sunk asset. So is a network of relatives who report on each other.
Why it matters here: a sunk asset gets used. Not because anybody decided it should, but because the alternative — leaving something expensive idle — feels like waste to everyone involved, and because the people operating it have no other role.
5.3 — The apparatus is a sunk asset
Building a system of watching is expensive. It takes generations to establish who reports to whom, whose word counts, which families exchange information, what a raised eyebrow at a wedding means, and who arranges marriages.
Once built, it costs almost nothing to keep. And machinery that exists gets used.
Part 12 is the proof, and it is the most striking thing in the series on this point. When the community lost its edges — when people moved to cities, when the neighbourhood stopped being a place where everyone knew everyone — the enforcement did not stop. It moved onto the phone, where it works better than it ever did in a village, because a village forgets and a screenshot does not.
The infrastructure outlived the community that built it. That is not what a practice sustained by current demand looks like. It is what a capability looks like.
5.4 — The beneficiary is not the enforcer
The third mechanism is the one Part 11 established and it is the reason nobody can switch this off.
The daily enforcement is administered largely by women on women — mothers-in-law, aunts, older sisters, neighbours. The people doing the work are not the people who gain most from the arrangement. And the people who gain most do not do the work, so they never feel its cost or handle its machinery.
A system where the operator and the beneficiary are different people has no natural off-switch. The beneficiary cannot retire it because he does not run it. The operator will not retire it because retiring it removes the only authority she has ever been given — which she acquired late, after decades of being on the receiving end, and which is the one currency her position pays in.
The evidence is a prediction that could have failed and did not.
If enforcement were fathers and brothers acting on belief, then a woman’s own view should be the best available predictor of her behaviour once the men are absent or indifferent. In the Indian household survey data it is not. A mother-in-law’s view of contraception predicts a couple’s use of it better than the wife’s own view does.
That is a specific, checkable, counter-intuitive result from a large national survey, and more than one South Asian dataset points the same way.
What it cannot show: motive. Nothing here establishes that an enforcing woman is protecting her own position rather than acting from conviction, or from fear, or from habit. The behaviour is measured. The reason is a story I am telling about the behaviour, and you should hold it more loosely than the number underneath it.
Three mechanisms so far, and each one weakens the case for the rules going away on their own. The fourth is different in kind. It does not describe machinery or incentives at all — it describes a trap that catches people who agree with each other.
Equilibrium: a situation that holds itself in place, because each person — given what everyone else is doing — has no reason to be the one who moves first.
It does not mean balanced, fair, or good. A traffic jam is an equilibrium. So is everybody staying seated at the end of a bad film.
Why it matters here: it is a third kind of explanation, alongside “it works” and “people believe it”. An equilibrium needs neither. It needs only that each person correctly expects the others not to move.
5.5 — Nobody can defect alone
The fourth mechanism is the strongest and it is what makes the other three durable.
Suppose a family becomes completely convinced that this is all nonsense. Their daughter should go where she likes, marry whom she likes, and no dowry will be paid.
What happens? Not a moral victory. The ranking is still in force everywhere else. Her marriage prospects narrow inside a market that has not changed its criteria. The family absorbs the entire cost of a reform that benefits everybody only if everybody does it.
That is the shape of a coordination problem, and it has a brutal property: the outcome does not depend on what people want. It depends on what each person expects everybody else to do. A population in which a clear majority privately disapproves of dowry will continue paying dowry, indefinitely, without anybody being a hypocrite, because each family is responding correctly to a world where the others have not stopped.
This is why the answer to “why don’t they just stop” is not stubbornness, and not false consciousness, and not secret enthusiasm. It is that stopping alone is not a smaller version of everybody stopping. It is a different act with a different cost, paid by one family.
The traditionalist and the reformer agree on almost nothing, and they agree completely on this: that a practice persists because it is still wanted.
Everything both movements do rests on it. The traditionalist points at survival as evidence of value — it lasted, so it must be serving something. The reformer points at survival as evidence of belief — they still think this way, so change how they think. Both are reading continuation as a signal about preference.
A coordination system does not work like that. It persists in a population where nobody in particular wants it, because wanting is not the mechanism. Expectation is. What holds a rule in place is not that people approve of it but that each person expects the others to enforce it — and being right about that is enough.
Two things follow, and they are large. Surveys of approval tell you nothing about durability — you can measure majority disapproval of a practice that is universal, and both facts can be stable for fifty years. And the entire visible surface of this argument, on both sides, is aimed at the layer that responds least. Persuading people is working on wanting. The rule is held by expecting.
The general form: continuation is evidence about coordination, not about consent — and every movement that reads it as consent will keep attacking a thing that is not holding the door shut.
That box, if it is right, is the most useful sentence in sixteen parts. It also explains the two things this series found hardest to reconcile: that persuasion fails so completely, and that visibility works so well. Persuasion changes what one person wants. Visibility changes what everyone expects everyone else to have seen.
The mechanisms above can be read two ways, and the reading decides what anybody should do about it.
Insurance where there is no pension. Coordination where there is no contract enforcement. Risk management where the police do not come. Every one of those needs is currently unmet for hundreds of millions of people, and the arrangement is meeting them badly rather than not at all. Remove it without replacing it and you do not get freedom, you get exposure — which is precisely what happened to the women in Part 12 when the community’s boundaries dissolved and the enforcement did not.
The stated functions have been solved or superseded one by one and nothing was retired. That is not a system responding to needs; that is machinery running because it was switched on. The migration in section 5.2 is the giveaway: a practice that changes jobs whenever its old job disappears is not serving a need, it is looking for one.
An equilibrium is a third kind of thing. It needs no current function, so the traditionalist cannot defend it by pointing at needs. It needs no inertia and no belief, so the reformer cannot dissolve it by changing minds. It needs only that each person expects the others to keep enforcing. That is why it survives prosperity, education, law and argument — none of those change the expectation, and several of them make defection more expensive rather than less.
What would settle it: a case where the function was fully replaced — a real pension, real policing, real contract enforcement — for a whole community at once, with the practice then measured over a generation. Parts of urban East Asia come closest and the evidence there is mixed and confounded. This is the largest empty shelf in the subject.
Why people care so much: because the three accounts imply three completely different programmes. If it is a living system, replace the function. If it is a leftover, wait. If it is an equilibrium, the only thing that works is a coordinated, visible, simultaneous shift — which is the hardest thing to organise and the one nobody is funding.
Notice what has happened over this chapter. We started with a puzzle that looked like a question about motives — why do people keep doing this? — and ended somewhere with no motives in it at all. That is usually a sign of progress in an argument, and it is also the point at which most readers lose interest, because a mechanism is far less satisfying than a villain.
Four mechanisms keep an answer alive after its question is solved. Functions migrate — seclusion became a display of wealth, dowry became a price for a scarce groom, so solving the original problem changes nothing. The apparatus is a sunk asset — expensive to build, cheap to keep, and machinery that exists gets used, which is why enforcement moved onto the phone when the community dissolved. The beneficiary is not the enforcer, so nobody is positioned to switch it off. And nobody can defect alone: the first family to stop pays the whole cost of a reform that only pays if everyone does it.
Underneath both camps sits the assumption that a practice persists because it is wanted. A coordination system persists on expectation, not on wanting — which is why you can measure majority disapproval of a universal practice and have both facts hold for fifty years.
This explains the two findings hardest to reconcile. Persuasion changes what one person wants. Visibility changes what everyone expects everyone else to have seen.
The three accounts of honour — living system, leftover, equilibrium — imply three different programmes: replace the function, wait, or organise a simultaneous visible shift. Only the third fits the evidence, and it is the one nobody is funding.
The rules survive law, prosperity, education and argument because none of those touch the thing holding the door shut.
6The Four Commitments — A Report
Part 15 ended with four checkable promises instead of a conclusion, and said that if this part arrived and none of them existed you would know exactly what that chapter had been. Two exist. Two do not.
6.1 — The scoreboard
The last part put four things on the page and named this one as the auditor. Here is the result, before the excuses.
| The commitment | Status | What actually exists |
|---|---|---|
| An evidence grade on every empirical claim across all fifteen parts | Part-kept | The four-grade scale is defined in Chapter One and used on every claim from Chapter Two onwards. The back matter of this part grades twenty-two load-bearing claims from across the series. The full pass, claim by claim through all fifteen parts, does not exist. |
| Part 6 republished so its scoring chapter cannot be detached from it | Not done | Nothing. Part 6 is still a standalone sixty-five-page brief for the old rules, with its own cover and its own download, and Chapter Nine of it still does not travel with it. |
| A genuinely short version | Kept | The Rules About Women — The Short Version. About four thousand words. Twenty minutes. Published alongside this part, as its own document, not as a summary that points back at the long one. |
| The missing part: what these rules are costing right now, and who is paying | Not done, and I am not going to do it | Nothing. Section 6.4 gives the reason and then says why the reason is not good enough. |
Two of four. That is a worse result than I expected when I wrote the promises, and a better one than the base rate for promises writers make about their own future work.
6.2 — What the short version is for
Of the two kept, one matters much more than the other.
Part 15 found that this series had made its prose accessible and left its object inaccessible. Every rule in the method is aimed at the sentence — short sentences, everyday words, nothing assumed. Nothing in the method is aimed at the total. So a thousand pages of scrupulously simple English produced something with an entry requirement of about twenty-five hours, and the person with the least spare time in the country is the person the series is about.
That objection landed and there is exactly one repair for it, which is to write the short one. It is published with this part.
The long series: about twenty-six hours. The short version: about twenty minutes.
Twenty minutes is a bus ride. It is the time between putting rice on and eating. It fits inside the gap that a person with two shifts actually has, which twenty-six hours does not, and no amount of simplifying the sentences was ever going to close that gap.
The short version contains eleven findings and four mechanisms. The long series contains eleven findings, four mechanisms, and roughly 320,000 words of showing the work. Whether the showing was worth 320,000 words is a question I cannot answer from inside it.
I want to be honest about what a short version costs, because the method in this series is built on the claim that simplification is not compression — that the simple version is longer, and that shortening drops content while pretending to clarify.
The short version breaks that rule. It is compression. It states the findings and does not show the work, which means a reader cannot check it and has to take my word. That is precisely the thing sixteen parts were written to avoid. It is still right to publish, because a document nobody can reach protects its reader from nothing, but the trade is real and I am not going to pretend it is free.
6.3 — What was not done, and why the reasons are thin
The full evidence grade is a large job and it is not finished. Twenty-two claims are graded in the back matter. There are several hundred empirical statements across fifteen parts. I am not going to give a date, because Part 15 established that naming a deadline converts a promise into a schedule and a schedule produces the feeling of progress without any. It is open. It has no date. Judge it by whether it appears.
Part 6 is the interesting failure, because it is the easy one.
Republishing Part 6 with its scoring chapter attached is not a writing task. It is a decision and about an hour of work: put the brief and its audit into one object, and take down the version that can be forwarded without the answer. There is no research involved, no evidence to gather, nothing to think through. It is the cheapest of the four by an enormous margin.
It is also the only one that requires me to remove something I made. Part 6 is, I think, the best-argued document in this series. It is the one that took most skill to write, because writing a case you disagree with at full strength is harder than writing one you believe. Taking it down as a standalone object means the best thing I made stops existing in the form that shows it off.
I do not have a defence of this. I have an explanation, and an explanation is not a defence.
6.4 — The one I am not going to write
The fourth commitment was the missing part: what these rules are costing right now, and who is paying, at the depth Part 10 gave to the other direction.
I am not going to write it, and Part 15 said that saying so honestly was an acceptable outcome. So here is the reason, followed by the reason it is not good enough.
The reason: I do not think I am the right person. The material would be, overwhelmingly, accounts from women of what daily management feels like — being asked where you are going, having a phone checked, being accompanied, being discussed. That is testimony, not published record, and this series has been built entirely from the published record. Writing it properly would mean going and asking, which is a different kind of work that I have not done and am not trained for. Writing it badly would mean assembling other people’s testimony from whatever fragments reached English-language print, which is the thinnest archive in the whole subject.
Now the problem with that reason. It is the same argument that would have prevented any part of this series. Part 11 was built almost from scratch because the material barely existed. Part 12 assembled a picture from law, police statistics and child-protection analysis because nothing better was there. Thin archives did not stop me before, and the difference this time is not evidential.
The honest description is that the missing part is the one where the discomfort is highest and the payoff for me is lowest, and I have found a methodological reason that happens to point where I was already going. That is what a rationalisation looks like from the inside, and I have no way of establishing that this is not one.
So the position on the record is: the objection from Part 15, Chapter Two, stands unanswered. The series asked what these rules are for and never asked at the same depth what they cost. That was the first objection to land and it is the one that remains open at the end. A reader who concludes that this makes the whole project lopsided is drawing the correct conclusion from the evidence I have given them.
Since I have declined, the reader is entitled to the strongest case on each side rather than just my decision.
A missing chapter is not neutral. In a series this long, absence reads as insignificance. Fifteen parts examining what the rules are for, with no part on what they cost, tells a reader that the second question is smaller — and that is a claim, made by the shape of the thing rather than by any sentence in it. An imperfect part that is clearly labelled as imperfect does less damage than a silence that is labelled as nothing.
Everywhere else in this series the author’s distance from the subject was a limitation. Here it is the subject. A part on what daily constraint feels like, written by a man who has never been constrained, from fragments of testimony filtered into English, would be a man narrating women’s experience while citing women’s experience. That is not a thin version of the right document. It is a different and worse kind of object, and its existence would make the real one less likely to be written by anybody else.
The choice is not write-it-badly or leave-it-blank. A third option exists and is standard in research: publish the specification. What the part would have to contain, what evidence it would need, who would be able to gather it, and what questions it would have to answer. That converts a silence into a commission. It is also, conveniently, much less work than writing it — which is a reason to distrust how attractive it sounds to me.
What would settle it: somebody writing it. If the part exists in five years and was written by a woman who lived it, this was the right call. If it does not exist, the silence was mine and specifying it changed nothing.
6.5 — What the pattern tells you
Look at which two were kept.
The evidence grade and the short version are both more writing. They cost time, which I was going to spend anyway, and they cost nothing else. Both were kept.
The other two are different. One required removing something I am proud of. The other required doing work that is uncomfortable and unrewarding for the person doing it. Neither was kept.
So the pattern is: I kept the promises that could be discharged by producing more of what I already like producing, and broke the two that required giving something up. That is not a special failure of mine. It is the ordinary shape of a promise kept by the person who made it, and it is why Part 15 said an author’s own audit is worth less than one line of correction actually applied.
It is also, in miniature, the finding of Chapter Five. A commitment I could discharge alone got discharged. A commitment that required me to bear a real cost, alone, while nothing else changed, did not.
Four commitments, two kept. The evidence grade is part-kept — the four-grade scale is in use from Chapter One and twenty-two load-bearing claims are graded in the back matter, but the full pass does not exist and has no date. The short version is kept and published alongside this part.
Two were not kept. Part 6 is still a detachable brief for the old rules, and it is the cheapest of the four — about an hour’s work — but the only one that requires taking down something I am proud of. The missing part on what the rules cost right now will not be written by me.
My reason for the fourth is that I am the wrong author. The problem with that reason is that it would have prevented every other part of this series. Thin archives never stopped me before, and the difference this time is not evidential.
So on the record: the first objection to land in Part 15 is the one still standing at the end.
I kept the two promises that could be discharged by writing more, and broke the two that required giving something up — which is the finding of Chapter Five, happening to the author.
7What You Could Do On Monday Morning
Part 15 found that in fifteen parts, not one section was addressed to a reader with an actual decision in front of her. This chapter is that section. It does not tell anybody what to do. It says which questions decide a case.
7.1 — The distinction this chapter runs on
There is a difference between telling someone what to do and telling them what their case turns on, and the whole of this chapter sits in the gap.
A doctor who says “have the operation” is advising a decision. A doctor who says “this operation is worth it if your main problem is pain and not worth it if your main problem is movement — which is it?” is advising the reasoning. The second is more useful and it is also the honest one when the person advising will not live with the result.
Fifteen parts refused to advise at all, on the ground that a man who has never been subject to these rules should not be spending someone else’s safety. That refusal was not neutral. Where the default is one-sided, declining to advise votes quietly for whatever was going to happen anyway.
So: four readers, four sets of questions. Nothing here is an instruction and everything here is drawn from a graded finding in Chapter Two.
7.2 — If you have a decision in front of you
By which I mean the reader this series is actually about — a woman weighing something specific against a warning she has been given.
Ask which half of the warning travels. The predicted harms are not one thing. Infection risk belongs to the act and shows up anywhere. Damage to your standing in the marriage market belongs to the community and shows up only where a community enforces it. So the useful question is not “what happens to a woman who does this?” It is “what does my community actually do, and to whom, and how recently?”
Treat the warning as a prediction and check it locally. You have been told something will happen. Has it happened, near you, to someone you can name? If it has, that is real information about your situation and it outweighs anything in this series. If nobody can produce a case, you are being told about a population you may not live in.
Notice who bears the cost of the advice and who bears the cost of the outcome. They are almost never the same person. This applies to the advice in this chapter too — I carry none of the risk of anything I have written.
Understand that the price of going first is not a smaller version of the price of everyone going. A family that defects alone absorbs the whole cost of a change that only pays if others follow. That is not an argument against doing it. It is an argument for knowing the actual price rather than the imagined one, and for looking hard at whether anyone near you is moving at the same time — because moving with three other families is a different act from moving alone, and costs a fraction as much.
Everything in the section above, and everything in every “informed choice” campaign ever run, rests on a premise nobody examines: that knowing more improves a decision.
Consider what this series can actually offer the woman in section 7.2. Population-level findings. What happens on average, across countries, in datasets mostly gathered elsewhere. But Chapter Two established that the effect she cares about most does not travel — it is determined by what her particular community does.
She already knows what her community does. She knows it in far more detail than any researcher, with a resolution no survey reaches. On the one question that decides her case, she is the expert and I am the amateur.
Which means the information in this series is least useful to the person with the most at stake, and most useful to the person with none — the observer, the policymaker, the reader at a distance. That is not a flaw in the writing. It is what population-level knowledge is, and no amount of care changes it.
There is a second half, and it is worse. “Informed choice” assumes information is the binding constraint — that she would act differently if only she knew more. In a coordination system the binding constraint is not knowledge. It is the price of going first. Telling someone the truth about a situation she cannot afford to change is not empowerment. It is a description delivered to a person who is already inside it.
The general form: information helps whoever has room to act on it, so campaigns built on informing people transfer help to the already-mobile and call it universal.
That box does not cancel the section above it. Local checking, the travelling test and knowing the real price of going first are all things a person can use. But it sets the ceiling on what any document like this can do for the reader it is most about, and the ceiling is low.
7.3 — If you are deciding for somebody else
A parent, usually. Sometimes a brother, an aunt, a husband.
Ask whether you are applying the rule to the person who needs it or the person you can reach. This is the single most useful thing in the series for anyone in this position. If you have a daughter at home and a son who has moved away, and the rules in your household fall almost entirely on the daughter, that is worth thirty seconds of honest thought — because on the evidence, that pattern is what enforcement looks like, not what protection looks like.
Notice who your worry is actually about. Of the six goals the tradition itself states for these rules, four concern what men will do. If your fear is about men, look at which child you are managing.
Do not expect education to do the work. The families with the most money and the most schooling produced the worst child sex ratios. Sending someone to a good school changes what she can do. It does not change what you want for her.
What measurably changed parents’ aspirations was seeing. In the village council experiment, the gap between what parents wanted for sons and for daughters closed by around a quarter — and it closed for people who had a woman visibly running things, not for people who were told women can run things. If you want your own view to move, the evidence says exposure does it and argument does not.
7.4 — If you want to change something
An organiser, an official, a funder, a person on a panchayat or a school committee.
Do not fund persuasion. It is the largest line in most reform budgets and the worst-evidenced activity in this entire series.
Do not reach for a ban. Every ban examined here failed and moved the behaviour somewhere less visible. A ban is attractive because it is the lever a legislature actually controls, which is a fact about legislatures rather than about the problem.
Put women where they are seen holding authority, and make sure they are seen. This is the one tier-one result in the subject. The active ingredient is the visibility, not the office — which predicts large effects in councils, schools, clinics and buses, and small ones in boardrooms, where almost nobody watches the person work.
Ask what function the rule is serving here, and supply that function another way. Insurance where there is no pension. Policing where nobody comes. Pricing in a marriage market. Attack the practice without replacing the function and you get the ultrasound ban: the demand survives and the behaviour moves.
Aim for simultaneous and visible rather than individual and private. If Chapter Five is right, the thing holding the rule in place is mutual expectation. Twenty families changing at once, publicly, in one place, is not twenty times one family changing quietly. It is a different intervention, and it is the one nobody funds because it has no national launch and no advertisement.
The advice in 7.4 rests on the sturdiest evidence in the series, which is worth saying plainly because advice usually rests on the softest.
“Do not fund persuasion” comes from a parliamentary committee auditing its own government’s spending. “Do not reach for a ban” comes from census counts before and after three separate prohibitions. “Put women where they are seen” comes from the only genuine randomisation in the subject, where villages were selected by serial number.
What this cannot show: that these transfer. Every one of them is Indian, and one of the standing findings of this series is that effects which look general often belong to their setting. The village council result has been replicated in its own country and not seriously tested elsewhere. Treat 7.4 as well-established for India and as a hypothesis anywhere else.
7.5 — If you are simply reading about this
Four habits, and they are worth more than any conclusion in this document.
Ask for the grade. When you meet a number, ask what it is made of. A register or a survey? Replicated or a single study? Self-reported, on something people are punished for? Most public argument on this subject consists of Soft numbers being carried like Solid ones.
Watch for the association-as-cause move. Two things appearing together is not one causing the other. Almost every claim made in public about women’s conduct performs this substitution, in both directions.
Apply the travelling test. Does this effect appear where nobody punishes the behaviour? If not, the punishment is the cause and the behaviour is the occasion.
Notice which findings you resisted. Go back through Chapters Three and Four and find the ones that irritated you. Then check whether the irritation tracked the evidence or tracked which side the finding helped. That is the test for whether you are reading or defending, and it is the only one that works from the inside.
Advice given to a person with little power has a well-known second effect, and it deserves stating at full strength.
Once she has been informed, every outcome becomes her choice. The family that constrained her, the market that prices her, the code that everyone else is still enforcing — all of that recedes, and what remains is a woman who was told the facts and decided. “Informed choice” is the mechanism by which a structural problem is converted into a series of personal ones, and the conversion always runs downhill.
The alternative is a thousand pages of analysis addressed to nobody. Silence does not protect her from having the decision; it only means she makes it without the two or three things that are actually known. And the person choosing silence is invariably the person with nothing at stake, who gets to feel scrupulous at no cost. Refusing to advise is not a lower-risk act. It is a lower-risk act for the writer.
The blame-shifting happens when advice is given only to the constrained party. This chapter has four sections and only one is addressed to her; the other three are addressed to the people doing the constraining, the people funding the response, and the people watching. That is not a complete answer, but it is the difference between “here are your options” and “here is what each of you is doing”.
What would settle it: nothing evidential. This is a question about what a writer owes, and the honest position is that I have chosen the arrangement that seems least bad to me while carrying none of the consequences of being wrong about it.
One last thing about this chapter. It is the only one in sixteen parts that tells anybody anything, and it took a prosecution of my own work to produce it. That is worth knowing about how writing like this normally goes: the practical section is not the hard part, and it is still the part that gets left out.
This chapter advises the reasoning, not the decision — what your case turns on, not what to do.
If you have a decision: ask which half of the warning travels, check the prediction locally against someone you can name, notice who bears the cost of the advice versus the outcome, and know that going first costs far more than going together.
If you are deciding for someone else: ask whether the rule falls on the person who needs it or the person you can reach, notice that four of the tradition’s six stated goals concern male conduct, and expect seeing to change your view where argument will not.
If you want to change something: do not fund persuasion, do not reach for a ban, make women visible in authority, supply the function the rule was serving, and move many households at once rather than one at a time.
Underneath all of it: knowing more only helps whoever has room to act. On the question that decides her case, the woman in section 7.2 is the expert and I am the amateur — which means this series is least useful to the person with the most at stake.
8The Empty Shelves
A map of what is known is also a map of where somebody sent a surveyor. These are the nine studies that would settle the largest open questions in this subject, and nobody has run any of them.
8.1 — Why this chapter exists
Chapter One said the list of established findings is a list of what has been checked, not a list of what is true, and that the blank areas on the map are not featureless ground but places nobody visited.
This chapter is the inventory of those blanks. Each item is a study that could actually be done, that would settle something specific, and that does not exist. I have put the cost next to each one, roughly, because the usual explanation for missing research is that it is hard, and for most of these it is not hard. It is unfunded, which is a different word.
8.2 — The nine
One. What the rules cost right now, and who pays. The missing part from Chapter Six, specified rather than written. It would need structured testimony gathered at scale in local languages, and time-use data covering the hours spent being asked, accompanied, checked and negotiated with — hours that no time-use survey currently codes because no category exists for them. Settles: the burden-of-proof asymmetry that has run through this whole series, where the cost of changing was measured and the cost of staying was assumed to be zero. Cost: a genuine fieldwork programme. Expensive by the standards of this list and trivial by the standards of a government campaign.
Two. The full replacement. Take a set of communities and supply, at once, the things the rules are substituting for: a pension worth living on, policing that arrives, contract enforcement that works. Then measure the practices over twenty years. Settles: the central question of Chapter Five — whether this is a living system meeting real needs, machinery running on after its purpose, or an equilibrium held by mutual expectation. Cost: very high, and it is the only item here that is genuinely a research programme rather than a study.
Three. Coordinated defection. The cheapest high-value study on this list. In matched villages, arrange for twenty families to change a specific practice at the same time and publicly, and compare with villages where the same twenty families are approached one at a time and privately. Same families, same message, different coordination. Settles: whether the equilibrium account of Chapter Five implies a workable intervention or is merely a good description. Cost: small. This could be run for less than a district-level advertising budget.
Four. Men’s constraint, measured outside the household. Part 11’s central claim is that the rule falls where it can be reached. But household surveys are conducted in households, so a method that visits homes will find the control that happens in homes. Measure what actually constrains migrant men: remittance expectations, marriage brokerage, the threat of being cut off. Settles: whether the reach thesis is a finding about the world or an artefact of where researchers knock. Cost: moderate. It needs a new sampling frame, not new money.
Five. The long-lag persuasion study. Follow a campaign cohort for twenty-five years. Every finding in this series about persuasion rests on measurements taken within five to ten years, and every large moral change in history took two or three generations, with the persuasion phase looking useless throughout. Settles: whether Part 14’s verdict on persuasion is right or is an artefact of how long grants last. Cost: low per year and impossible under current funding structures, which is the whole problem.
Six. The behavioural findings, run outside the West. The partner-count association, the regret asymmetry, the wellbeing reversal — all of it, replicated properly in India, Nigeria and Indonesia. At present roughly 96 per cent of behavioural research subjects come from countries holding about 12 per cent of the world’s people, and those are the countries where these rules have already been dismantled. Settles: most of Part 10, in whichever direction the results fall. Cost: moderate, and lower every year.
Seven. A shared scoreboard, negotiated in public. Not the scoreboard — the negotiation. Put serious people from both positions in a room and have them agree, in advance and before seeing any data, on what would count as a good outcome. Publish the transcript. Settles: nothing empirical, and that is the point. Where the negotiation breaks down is the most informative document anybody could produce on this subject, because it would show exactly where the disagreement stops being about evidence. Cost: almost nothing. Nobody has done it in a century of arguing.
Eight. Who reads documents like this one, and what they do with them. Where a piece is linked, what gets quoted from it, whether a scoring chapter travels with the brief it scores. Settles: whether even-handed writing informs people or supplies them. Cost: the least of anything here. It is a weekend’s work with public data and I have not done it, which Part 15 said was not a neutral omission and I am not going to pretend otherwise now.
Nine. Visibility, isolated from office. Part 14 claimed the active ingredient in the village council result was being seen holding authority rather than holding it. That is an inference from the result, not a separate measurement. Compare roles where the holder is watched working — a council, a clinic, a classroom, a bus — against roles where she is not, like a board seat. Settles: whether the single strongest positive finding in the subject generalises or is about one kind of job. Cost: moderate.
India’s flagship campaign for the girl child released ₹446.72 crore to states over three years and put 78.91 per cent of it into media advocacy. That is roughly ₹352 crore — about 3.5 billion rupees — spent on advertising a campaign.
Items three, seven and eight on that list, together, would cost a small fraction of it. Item eight would cost roughly one person’s fortnight.
So the money to answer several of the largest open questions in this subject has already been spent. It went on telling people the questions matter.
8.3 — Why the shelves are empty
Four reasons, and none of them is that the questions are hard.
Funding runs in three-year cycles and social change runs in generations. Item five is impossible not because anybody objects to it but because no institution exists whose planning horizon is twenty-five years and whose interest is this.
A finding of nothing does not get published. So researchers avoid designs likely to produce one, which removes exactly the studies that would tell you a popular claim is empty.
Most of these serve nobody’s campaign. Item seven would embarrass both sides by showing where their disagreement stops being evidential. Item eight would tell a writer unwelcome things about his own work. Item four could weaken a finding that reformers currently find useful. Research gets funded by people who expect to like the answer.
And the subjects are not the customers. Nobody commissions research on their own behalf here. The women whose lives these questions are about have no budget, no institution and no journal, so the questions that matter most to them are the ones nobody is paying to ask.
The evidence for an absence is awkward, so here is exactly what it consists of.
For items two, three, five, seven and nine: a search of the published literature turns up nothing of the described design. That is weak evidence — an unpublished study is invisible and an unsuccessful search is not a proof.
For item six the evidence is much stronger and it is a count rather than a search: the survey of research subjects by country, showing 96 per cent from Western industrialised nations and 68 per cent from the United States alone. That is a measurement of where the work was done, not a failure to find it.
For item eight the evidence is that I know, because it is my own work and I have not looked.
Treat a missing count the way this series has treated one throughout. When institutions had the power to fund an answer and did not, the absence is a finding rather than a gap — and this chapter’s argument is that the pattern of what is missing is too consistent to be random.
Section 8.3 stated four reasons as though they were established. They are not, and the strongest reply to them deserves the floor.
Section 8.3 makes a strong claim, and the strong version is worth testing against its best rival.
Look at the pattern. Every missing study is one that would either embarrass a funder, produce a null result, take longer than a career, or serve people with no money. That is not a random distribution of gaps. It is the shape you get when the questions are filtered by who pays, and the filter is invisible because nobody ever announces a study they decided not to fund.
Research is opportunistic everywhere, on every subject. Physics has enormous unexplored areas too, and nobody calls that political. Studies happen where a graduate student, an existing dataset and a supervisor’s interest happen to meet. A list of nine unrun studies assembled by somebody who already believes the field is skewed will always look damning, because he selected them. The null cases — obvious studies that were run and found nothing interesting — are invisible to him for exactly the reason he complains about.
“Design” implies somebody decided and “accident” implies nobody did. What actually operates is a filter: journals prefer findings, funders prefer answers they can use, careers prefer results inside a decade, and subjects with no budget generate no demand. Nobody has to intend anything. Run those four preferences for fifty years and you get precisely this pattern of shelves, with no meeting having taken place.
What would settle it: registries of proposed studies that were not funded. Some funders publish these. Nobody has analysed them for this subject.
Why people care so much: because “nobody has studied it” is currently used as a knockdown argument by whichever side the absence favours, and it is not one. An absence is evidence about the field, not about the world.
An empty shelf is the cheapest thing in research to point at and the most expensive to fill. That asymmetry is why this chapter exists and why it is the second-shortest in the part.
Nine studies would settle most of the open questions here and none has been run. The cheapest and most valuable is coordinated defection — twenty families changing at once and publicly, against twenty approached one at a time — which would test whether Chapter Five’s account implies a real intervention.
Others: what the rules cost right now, the full replacement of the functions they serve, men’s constraint measured outside the household, a twenty-five-year persuasion study, the behavioural findings run outside the West, a published scoreboard negotiation, who actually reads this kind of work, and visibility isolated from office.
They are missing because funding runs in three-year cycles, null results do not publish, most of them serve nobody’s campaign, and the subjects are not the customers — the women these questions concern have no budget, no institution and no journal.
Nobody designed this and nobody caused it. It is a filter with no operator: four ordinary preferences, run for fifty years, with no meeting ever taking place.
The money to answer several of these questions has already been spent — on advertising a campaign about the questions.
9What I Actually Think
Fifteen parts refused to say. This one says it, states plainly how little weight it deserves, and then turns the last hidden assumption on the whole series.
9.1 — Why this is here at all
The standing reason for refusing was good. I am a man who has never been subject to these rules, writing for people who would bear every consequence of taking my advice. Advice under that arrangement is a stranger spending somebody else’s safety.
That reason has not stopped being true. Here is why I am overriding it.
A thousand pages plainly imply a view. Anybody who has read this far knows roughly what I think, because the choice of what to examine, what to test hardest and what to leave alone has been telling them for sixteen parts. Refusing to state it while broadcasting it is not restraint. It is having it both ways: the influence of a position with none of the accountability of holding one.
So it goes on the page, in one section, labelled, where it can be argued with.
9.2 — What I think
First, on what this is. I think the rules about women are not a moral order and not a conspiracy. They are an equilibrium — a set of arrangements each person is right to keep following as long as everyone else does, held in place by expectation rather than by belief or by benefit. That is why they survive prosperity, education, law and argument. None of those touch the thing holding the door.
Second, on whether they are wrong. I think they are, and I want to be careful about the ground, because the outcome evidence in this series does not settle it and I am not going to pretend it does.
The ground is not that the rules produce bad results. Some of them produce mixed results and one or two produce real protection. The ground is that the person carrying the burden did not agree to it and cannot leave it, and that the tradition’s own accounting says four of its six stated worries concern what men will do. An arrangement that identifies male conduct as the problem four times in six, and then builds its entire enforcement around female movement, has stopped being a moral order and become a distribution of costs onto whoever is in reach. I think that is wrong for reasons that have nothing to do with any table in this book, and I would think it if every effect size came back favourable.
Third, on the reformers. I think most of what the reforming movement does is ineffective and some of it is worse than ineffective. Persuasion does not work and consumes the largest share of the budget. Bans do not work and move the harm somewhere less visible. Reporting national averages while the gains sit with the urban and educated is not a communication problem, it is a measurement choice that conceals exactly what the argument is about. An honest reformer would stop persuading and start installing: clinics, wages, records, school rolls, seats, and women visibly running things where people can watch them do it.
Fourth, on the traditionalists. I think the fertility argument is the strongest thing anybody has said in this entire debate and that nobody has an answer to it, including me. No society has replaced itself after reorganising around individual choice, and none has climbed back. The habit on my side of the argument — changing the subject when this comes up, or treating it as a bad-faith talking point — is intellectually disgraceful and it is what people do when they have nothing. I do not know what the answer is. I am fairly confident it is not more advertising, and fairly confident it is not restoring the rules, because Iran ran that experiment and the rate fell anyway.
Fifth, the concrete test. The honest way to state a view about other people’s lives is to say what you would do in your own. If I had a daughter, I would not manage her movements, and I would expect to pay for that, and the payment would be real rather than symbolic — narrower marriage prospects inside a market that has not changed its criteria, and a family that has spent something on a change that only pays if others make it too. I would want to make that move alongside other households rather than alone, because Chapter Five says that is not a preference but an enormous difference in price. And I would not pretend the cost was somebody else’s superstition.
9.3 — How much weight this deserves
Very little, and I can be specific about why.
Part 15 measured this author against his own first expectations on six questions and found him wrong on all six — always in the direction of the ordinary, individual-level, belief-driven answer. That is a measured error rate, on this subject, for this person, and it applies to everything above, because the claims above have not been tested by anything.
Nothing in section 9.2 carries a grade. It cannot. It is not that kind of statement.
And the one question that decides an actual case — what your community will actually do — is one on which the woman deciding is the expert and I am the amateur. Chapter Seven’s hidden assumption established that and it applies with full force here.
The whole justification for reading somebody after a long study is that the study improved them. It is worth checking.
A person who has read the literature knows which claims have been tested and which have not, which effects travel, and where the evidence stops. That is not a matter of opinion improving; it is a matter of having information the first impression lacked. Anyone who denies this is denying that reading does anything, which nobody believes when it is their own field.
The literature is not a sample of what is true. It is a sample of what got funded, published and written in English. Reading it thoroughly moves you towards the field’s centre of gravity, not towards the truth, and it moves you there with much greater confidence than you started with. Part 15’s six reversals were all in one direction: the direction of the popular reading. That is not a personal defect, it is what happens to anyone who reads carefully in a skewed field.
The two claims are different. On how something works — what a ban does to a demand, what visibility does to expectation — a considered view is straightforwardly better, because those are checkable and the checking has been done. On whether the arrangement is right, the study adds nothing at all. Sixteen parts of evidence gave me no standing whatsoever on section 9.2’s second paragraph, which is a claim about consent and burden that a person could reach in an afternoon.
What would settle it: tracking whether people who study a contested subject converge on truth or on their field’s consensus. This has been examined in some sciences and not here.
Why people care so much: because “I have studied this” is the strongest move available in a public argument, and it works even when the study bore only on a question nobody is fighting about.
9.4 — The assumption underneath all sixteen parts
One thing left, and I have been avoiding it since about Part 4.
Sixteen parts, a thousand pages, one title. Every one of them assumed something nobody in this argument ever states: that “the rules about women” is one thing. A single object, with a single history, a single explanation, and therefore a single answer.
Suppose it is five things wearing the same face. A property transfer at marriage. A paternity device. A coordination code that lets strangers within a group trust each other. A labour allocation that assigns unpaid work. A status display that signals a family can afford to keep an adult out of paid employment.
Those five have different origins, different beneficiaries, different economics and different repairs. What they share is an enforcement apparatus — the same aunts, the same reputation, the same phone — which is precisely why they look like one system from the outside. But a shared delivery mechanism is not a shared cause, any more than the postal service is the reason for the letters.
If this is right, then almost every general statement in this series is an average across five unrelated machines, and the flat results are exactly what you would expect from averaging things that do not belong together. It would also explain the most stubborn pattern in the whole subject: that interventions which work well on one practice do nothing to another, in ways nobody has been able to predict.
And I know why I assumed the single object. Not because the evidence pointed there. Because a series needs a subject, and “five loosely related arrangements that happen to be enforced by the same people” is not a title. The structure of the thing I was making decided the shape of what I could find in it.
The general form: the unit of analysis is chosen before the analysis and is never itself analysed. It arrives with the project, disguised as the topic.
9.5 — What that leaves
Not much, and I would rather end honestly than tidily.
Eleven findings hold. Four mechanisms explain a great deal of what the findings show, and the fourth of them — that nobody can defect alone — is the one I would keep if I could keep only one. Six widely held beliefs do not survive, three of them mine at the start.
Beyond that: an argument that stalls at the edge of a narrow band, a set of empty shelves too consistently shaped to be accidental, a first objection that is still standing at the end because I declined to answer it, and a subject that may not be a subject.
The series set out to find out why the rules about women exist and what happens when they change. It found that the second question has better answers than the first, that the first may have been malformed, and that the thing most likely to move anything is not writing at all — which is an awkward result for a thousand pages, and the correct one to report.
What I think: the rules are an equilibrium, not a moral order and not a conspiracy. They are wrong, on grounds of consent and of who carries the burden — not on outcome evidence, which does not settle it. The reformers’ programme is mostly ineffective and should stop persuading and start installing. And the fertility argument is the strongest thing anyone has said in this debate, nobody has an answer, and changing the subject when it comes up is disgraceful.
Weight it accordingly: none of that carries a grade, and Part 15 measured this author wrong on six out of six of his own first expectations, always in the direction of the popular reading.
A considered view is worth more on mechanisms and no more on values. Sixteen parts gave me real standing on what a ban does to a demand and no standing at all on whether the arrangement is right.
The last assumption, underneath everything: that “the rules about women” is one thing. It may be five — a property transfer, a paternity device, a coordination code, a labour allocation and a status display — sharing an enforcement apparatus but not a cause. I assumed one object because a series needs a subject.
The unit of analysis is chosen before the analysis and never itself analysed. It arrives with the project, disguised as the topic.
10An Honest List of What We Do Not Know
Every part of this series ends here. This time the list covers all sixteen — what is genuinely unknown across the whole subject, why, and the shorter list of what can be built on.
10.1 — Genuinely unknown
The reason for a gap is nearly always more useful than the missing fact would have been. That has been true in every part of this series and it is most true here.
What causes the fertility decline. The decline is beyond dispute and universal among rich and urbanising societies. The candidate causes — housing costs, the length of education, insecure work, the disappearance of unpaid childcare from relatives, the expansion of women’s options — arrive together everywhere and have never been separated. Why it is unknown: there is no case where one moved and the others held still. Not one. This is the largest open question in the subject and both camps are reading their preferred cause out of the single dataset everybody shares.
Whether the reach thesis is a fact about the world. The claim that rules fall where they can be enforced rather than where belief points is the most important thing in this series and it is not proved. Why it is unknown: household surveys are conducted in households, so a method that visits homes will find the control that happens in homes. Nobody has measured what actually constrains a man who has moved away.
Whether this is one thing or five. Chapter Nine’s last box asks whether “the rules about women” is a single arrangement or a family resemblance across a property transfer, a paternity device, a coordination code, a labour allocation and a status display. Why it is unknown: the unit of analysis is chosen before any analysis and is never itself examined — by me or by anyone else writing on this.
What the rules cost right now, and who pays. Why it is unknown: nobody counts it. There is no category in any time-use survey for hours spent being asked, accompanied, checked and negotiated with, so the largest ongoing cost in the subject is invisible to every instrument pointed at it. Absence of a category is not absence of a thing.
Whether persuasion works on a long lag. Every measurement runs five to ten years. Every large moral change in history took two or three generations and looked useless throughout. Why it is unknown: research funding runs in three-year cycles. The gap between those two numbers is a fact about institutions, not about persuasion.
Whether the visibility effect is about being seen or about holding office. Why it is unknown: the strongest positive finding in the subject came from one experiment that did not separate the two, and nobody has run the comparison since.
Whether any of the Indian results transfer. The village council result, the sex ratio gradient, the campaign spending, the ban failures — all Indian. Why it is unknown: roughly 96 per cent of behavioural research subjects come from Western industrialised countries holding about 12 per cent of the world’s people, and the middle of the world has been studied by almost nobody. Indian results are unusual in existing at all, which is a poor reason to assume they generalise.
What the people with no organised position would say. The woman who thinks the rules are foolish and complies because complying is cheaper. The mother-in-law who enforces and would not call it enforcing. The unmarried man at the bottom of the class structure, who appears in this series only as a risk to others. Why it is unknown: they have no literature and no budget, this series was built entirely from the published record, and I never went and asked.
Whether this series did more good than harm. Why it is unknown: there is no comparison world. I cannot observe the version of the last year in which these thousand pages were not written and see what those readers did instead. It is the only question that finally matters and there is no instrument for it at all.
Look at what those nine have in common, because the pattern is the finding.
Not one is unknown because the evidence is contradictory. Not one is a case where careful people looked hard and came back with a muddle. Every single one is unknown because nobody built the instrument — no category in the survey, no funding past three years, no sampling frame outside the household, no researchers in the middle of the world, no budget belonging to the people concerned, and in the last case no possible measurement at all.
Chapter Eight’s argument box asked whether that pattern was designed or accidental and settled on a third answer: a filter with no operator. This list is what that filter produces.
What this cannot show: that the answers would be interesting. Some unstudied questions are unstudied because they are dull, and you cannot tell which from outside. That is the second-order version of the same problem and there is no way around it.
10.2 — Solid
Shorter, as always, and worth more than everything above it.
The eleven findings in Chapter Two carry their grades and stand. The strongest are the ones built out of records that institutions made while doing something else.
The stated function was solved and nothing was retired. Cheap paternity testing has existed for thirty years; no community anywhere relaxed a rule in response. Why it is solid: both halves are checkable by anybody, and an absence of response where every opportunity existed is a finding rather than a gap.
Every ban examined here failed. The 1929 marriage-age act, dowry prohibition, the ultrasound ban and the child sex ratio falling from 962 to 918. Why it is solid: census counts and administrative records, collected by offices with no stake in this argument.
Visible female authority changes what people believe about female authority. Village council leadership reserved and rotated by serial number; measured bias fell; more women later won unreserved seats; the gap in what parents wanted for daughters closed by around a quarter. Why it is solid: the rotation was mechanical, which makes it the only genuine randomisation in the subject.
Son preference did the most damage among the richest and best educated. Why it is solid: census data, and the direction is the opposite of what almost everyone predicts, which is a good sign in a finding.
India’s flagship campaign put 78.91 per cent of its released funds into media advocacy. Why it is solid: a parliamentary committee auditing its own government and finding against it — the direction institutional evidence almost never runs. The caveat stands: the sex ratio improved over the period and attribution is contested.
No country has reversed a fertility decline. Why it is solid: national vital statistics, and the claim is an absence across every case, which is the easiest kind of claim to refute and nobody has.
The state’s own courts have described the old arrangement in the plainest possible terms. A 2018 judgment striking down the adultery law said the provision had treated a wife as property. Why it is solid: it is a document.
Two unconnected sets of criminals price the double standard identically. Why it is solid: independent convergence. When two groups with no contact, no ideology and only a profit motive arrive at the same valuation of the same social fact, the fact is real.
Nine things unknown. Nine things solid. Sixteen parts, a thousand pages, and roughly a year of evenings.
Put like that it sounds thin, and it is worth saying that this is a normal yield rather than a poor one. A working scientist can spend a decade and add two or three things to the second list. The difference is that nobody expects a scientist to also produce a conclusion, and every reader of a book expects one.
10.3 — And one thing that is certain without being empirical
Where nobody is persuaded, the arrangement already running continues. That needs no study. It is what a default is, and it means every argument in this series that ended without resolution ended with the rules still in force — decided by the structure of the situation rather than by anything either side said.
Add the finding from Chapter Five and it sharpens into something more useful: a practice continuing is evidence about coordination, not about consent. You may not infer from the survival of an arrangement that anybody wants it, and you may not infer from majority disapproval that it will end. Both inferences are made constantly, by both sides, and both are invalid.
Nine things are genuinely unknown, and not one of them is unknown because the evidence is contradictory. Every single one is unknown because nobody built the instrument — no category, no funding past three years, no sampling frame, no researchers, no budget belonging to the people concerned.
The largest is what causes the fertility decline: the candidates always arrive together and no case exists where one moved and the others held still.
What is solid is narrow and mostly built from records made by institutions doing something else — a census, a court judgment, a spending audit, a rule for rotating seats, and one accidental randomisation.
And one certainty that needs no study: where nobody is persuaded, the default runs. Every unresolved argument in sixteen parts ended with the rules in force.
A practice continuing is evidence about coordination, not about consent — and both sides make the opposite inference constantly.
Where This Ends
There is no Part 17. What is left is a set of questions I could not answer, handed over to anybody who wants them. They are stated as questions because questions create an obligation and topics do not.
- What do these rules cost, right now, in hours and in what those hours would otherwise have been — and who is paying? Nobody counts it, because no survey has a category for being asked, accompanied and checked.
- Does the rule really fall where it can be reached, or does it only look that way because researchers knock on doors?
- Is this one arrangement or five — a property transfer, a paternity device, a coordination code, a labour allocation and a status display — sharing an enforcement apparatus but not a cause?
- What happens when twenty households in one place change at the same time and in public, compared with twenty approached one at a time and quietly?
- What actually causes a fertility decline, and is there any case anywhere where the material conditions moved and the cultural arrangement held still?
- Does a campaign that shows nothing after ten years show something after twenty-five, and is there any institution on earth whose planning horizon could find out?
- And the one addressed to whoever reads this: which of the findings in Chapter Two irritated you, and did the irritation track the evidence or track which side the finding helped?
The short version of this series is published alongside this part. It is about four thousand words and it takes twenty minutes. If you are going to pass one thing to somebody, pass that.
Sources & further reading — Part 16
The Evidence Grade
This is the first instalment of the commitment made in Part 15 and reported on in Chapter Six. Twenty-two load-bearing claims from across the series, with what each one actually rests on and what weight it can carry. The full pass, claim by claim through all sixteen parts, does not exist.
The scale, once more. Solid is a count, a register, a court record or a genuine randomised experiment. Good is a finding an independent team has run again. Single is one study, never rechecked. Soft is self-report about punished behaviour, or a projection rather than a measurement. Soft does not mean false and Solid does not mean important.
| Claim | What it rests on | Grade |
|---|---|---|
| The rule falls where it can be enforced, not where belief points | Censuses and household surveys, plus an inference from the pattern | Solid counts, contested reading |
| Daily enforcement is largely administered by women on women | Large national household surveys; more than one South Asian dataset | Good |
| Cheap paternity testing has existed for thirty years and nothing was retired | The technology record, plus an absence of response where every opportunity existed | Solid |
| Of the predicted harms, infection travels and marriage-market damage does not | Epidemiology for the first; a comparison across settings for the second | Solid |
| The wellbeing effect reverses with the community’s belief | Several small studies pointing the same way; none large | Single to Good |
| The regret asymmetry travels | A US finding replicated by an independent Norwegian team | Good, but a modest effect |
| Partner count before marriage is associated with later divorce | One national US survey; self-reported; not a straight line; no mechanism; untested elsewhere | Soft |
| The 1929 act raising the marriage age produced a surge of marriages before it commenced | Administrative records | Solid |
| The child sex ratio fell from 962 to 918 girls per 1,000 boys after the scan ban | Census | Solid |
| Son preference did most damage among richer, urban and educated families | Census | Solid |
| Reserving village council seats reduced measured bias and led to more women winning unreserved seats | Reservation rotated by village serial number — a genuine randomisation | Solid |
| The parental aspiration gap closed by around a quarter, and adolescent girls’ own by about a third | The same experiment | Solid |
| 78.91 per cent of ₹446.72 crore released to states went on media advocacy | Parliamentary committee report, December 2021, auditing its own government | Solid |
| South Korea has spent over $200 billion; the rate moved from 0.72 to 0.80 | National vital statistics | Solid |
| Hungary spent around 5 per cent of national output, reached 1.59 in 2021 and is back near 1.31 | National vital statistics | Solid |
| India’s fertility rate is about 1.9, with rural India at 2.1, Delhi at 1.2 and Bihar at 2.8 | Sample Registration System | Solid |
| The marital rape exception was retained in the new criminal code; the Union opposed removal in October 2024; no verdict as of early 2026 | Statute and filed court positions | Solid |
| India’s Supreme Court, striking down the adultery law in 2018, said it had treated a wife as property | The judgment text | Solid |
| Around nine in ten victims of financial blackmail using intimate images are boys aged 14 to 17 | Child-protection and law-enforcement analyses, converging | Good |
| Never-married Indian men rise from about 4.5 per cent in 2010 to 12.6 per cent by 2040 | A demographic projection, not an observation | Soft |
| Of 100 psychology studies rerun, 36 replicated; of 21 in Nature and Science, 13; of 274 claims across 54 journals, about 55 per cent | Pre-registered replication projects designed to be checked | Solid |
| About 96 per cent of behavioural research subjects come from countries holding about 12 per cent of the world’s people | A count of published papers by country | Solid |
The Series, Part by Part
| Part | Title and what it established |
|---|---|
| One | The Question Underneath the Question — set the framing for all sixteen parts, and the travelling test used later to check whether an effect found in one place shows up in another. |
| Two | Where the Rules Came From — the historical origins, and the warning that where something came from does not settle what it is worth. |
| Three | The Indian Machine — how the rules are transmitted and enforced here; turns the where-it-came-from objection on the series itself. |
| Four | The Bodies Themselves — measured sex differences, with an effect-size table and the caveats about what an average difference licenses about any individual. |
| Five | The Paradox of the Free Countries — four competing accounts of why outcomes in liberal societies matched nobody’s predictions. |
| Six | The Case for the Old Rules — advocacy; seven arguments at full strength, then scored against four conditions. |
| Seven | What Feminism Actually Claims — an audit; a claims scorecard and an Indian timeline from 1848. |
| Eight | The Ledger of the Equality Project — eighteen rows of outcomes, with a column asking who each change actually reached. |
| Nine | The Arithmetic of Descendants — fertility, pronatalism, and the finding that no country has reversed a decline. |
| Ten | What the Evidence Says Happens to a Woman — the core warning unpacked into four claims and tested. |
| Eleven | The Half Nobody Studied — men as the missing comparison group; the rule falls where it can be enforced. |
| Twelve | The Community With No Edges — enforcement after the community loses its boundaries; two patterns of blackmail-by-photograph, priced differently by two independent sets of criminals. |
| Thirteen | When History Ran the Experiment — nine cases graded against six tests; one genuine randomisation exists in the whole subject. |
| Fourteen | What Would Actually Work — four intervention shapes scored; persuasion classified as a category error, visibility as the one tier-one success. |
| Fifteen | The Case Against This Series — fourteen objections, twelve of which land, and four commitments instead of a conclusion. |
| Sixteen | What Survived — this document. Eleven findings, four mechanisms, six beliefs that did not survive, and a report on the four commitments: two kept. |
| After | The series ends here. What remains is the short version, the nine empty shelves in Chapter Eight, and one objection still standing. |
Glossary
Every term this document treats as needing explanation, in plain English, in one place.
| Term | Meaning |
|---|---|
| Administrative record | Information collected by a government or institution in order to do its job — a birth register, a school roll, a court file. The most reliable evidence in this series, because it was made for another purpose. |
| Association | Two things that show up together. Not the same as one causing the other. |
| Cause | One thing making another happen. |
| Coordination problem | A situation where what is best for you depends on what everybody else does, so nobody can improve things by acting alone. |
| Crore | Ten million, in Indian counting. ₹1 crore is ten million rupees. |
| Default | What happens if nobody does anything. In a system of rules, the arrangement currently running. |
| Effect size | How big a difference something makes, as opposed to whether the difference is real at all. |
| Equilibrium | A situation that holds itself in place because each person, given what everyone else is doing, has no reason to move first. |
| Fertility rate | The average number of children a woman would have over her life at current rates. About 2.1 is replacement — the level at which a population holds steady without migration. |
| Functional explanation | Explaining something by what it does, rather than by how it arose or whether it is right. |
| Hidden assumption | A premise both sides of an argument accept without stating, and therefore never examine. |
| Marriage squeeze | A shortage of women of marrying age relative to men, produced by an unbalanced sex ratio at birth working its way up the age structure. |
| Null result | A study that finds nothing. Rarely published, which is why the published record is not a sample of what is true. |
| Randomisation | Assigning people or places by lot, so the groups differ only by chance and any later difference can be attributed to the treatment. |
| Replication | Running somebody else’s study again, on new people, to see whether the same thing happens. |
| Self-report | Evidence made of what people say about themselves. Reliable for some things, and not for anything they are punished for. |
| Sunk asset | Something already built which is expensive to create and cheap to keep running. Machinery that exists gets used. |
| Synthesis | Putting separate findings together to see what they add up to. Not a summary — a summary shortens each part, a synthesis asks what holds across all of them. |
| Travelling test | Checking whether an effect found in one place appears in another. If it does not travel, it belongs to the setting rather than to the act. |