The Question That Started This
This is the part I began the series in order to write.
Everything before it was preparation. What reservation is. Where caste came from. How it hardened. Whether outsiders brought it. Those are interesting questions and they are all, in the end, secondary to this one:
Did it actually happen? And if it did, was it as bad as people say, or has it been inflated for political purposes?
That question deserves a straight answer rather than a lecture, so this part is built differently from the others. There is very little interpretation in it. It is mostly evidence, laid out in the order a reasonable person would want to see it, with sources attached.
That sentence is Chapter 6, and it is the single most important piece of evidence in this book. I will not spoil it here beyond saying that when I found it I stopped writing for a day.
Three commitments before we start.
First, I have gone looking for exaggeration, and I found some. Chapter 4 takes one of the most widely circulated stories in Indian anti-caste politics and shows what the record actually supports and what has been added to it. I am not going to pretend that everything told on my side of an argument is true, because that is exactly the habit this series exists to break.
Second, the "it is exaggerated" case gets a full chapter of its own, written properly, with the points where it genuinely wins clearly marked. It has real ones. They are not the ones its supporters usually make.
Third, I have preferred evidence that is hard to fake. Court records over recollections. Occupational censuses over surveys. Experiments over testimony. Perpetrator self-reports over victim self-reports. Where the only evidence is testimony, I say so.
One last thing. Some of this is difficult reading — Chapter 9 in particular. I have not made it more graphic than the facts require, and I have not made it less. People have died in those conditions this year.
— Lovepreet Singh
July 2026
1The Charge
The case that caste oppression has been exaggerated — written by its best advocate, not its worst.
In July 2026, one of the most prominent voices in the Reservation Hatao Andolan described historical caste atrocities as "propaganda pushed by leftist historians to induce guilt."
That is the crude version. Here is the serious one — the case as a careful, intelligent sceptic would actually make it. It deserves to be answered rather than dismissed, because a great many people believe some version of it.
1. The sources are contaminated. Almost everything written about Indian social conditions before 1947 was recorded by colonial officials or Christian missionaries. Both had powerful reasons to portray Hindu society at its worst: the first to justify continued rule, the second to justify conversion. We are reading the prosecution's file and calling it the record.
2. Demonstrable embellishment exists. Several of the most emotionally powerful stories in Indian anti-caste politics turn out, on examination, to have grown in the telling. If some are inflated, a reasonable person should ask how many.
3. Crime statistics measure reporting, not crime. A rise in registered cases can mean more offences or better registration, more awareness, more police stations, or more incentive to file. And a law with mandatory compensation attached creates an incentive that any honest analyst must account for.
4. Aggregation smuggles in the conclusion. When two people of different castes fight over land or a woman or a debt, and one is Scheduled Caste, the case is registered as an atrocity. That does not make caste the motive. National totals silently convert ordinary human conflict into evidence of a system.
5. Surveys measure question wording. Ask about "untouchability" in a country where the word carries legal and social weight and you will get answers shaped by what people think you mean and what they think is safe to say.
6. Enormous change has happened, and the language has not moved. India in 2026 is not India in 1926. Describing the present in the vocabulary of the past is not a small error; it is the whole error, because policy is made for the present.
7. Every pre-modern society was brutal. Slavery, serfdom, guilds, hereditary aristocracies. Singling out one society's hierarchy as uniquely monstrous is a political choice, not a historical finding.
How this part answers it
Not by asserting the opposite. By producing evidence of kinds that the seven objections above cannot touch.
| If you doubt… | Then look at… |
|---|---|
| Colonial and missionary sources | Court records, princely-state proclamations, and Indian-authored accounts — Chapter 3 |
| Embellished stories | Chapter 4, where I take one apart myself |
| Victim testimony | Chapter 6 — a survey of what people say they themselves do |
| Crime statistics | Chapter 8 — controlled experiments with no reporting step at all |
| Survey wording | Chapter 9 — an occupational headcount, not an opinion |
| Historical distance | Chapters 6 to 9 are all from the last two decades |
And Chapter 10 goes back through all seven and marks which ones survive. Three of them do.
2What Counts As Evidence
A short chapter on how to judge any claim about caste — including the ones in this book.
Before any of the evidence, a method. Not all evidence is equal, and the differences are not about how upsetting it is.
| Type | Why it is strong or weak | Used in |
|---|---|---|
| Controlled experiment | No reporting step, no memory, no politics. The only variable is the one deliberately changed. | Chapter 8 |
| Perpetrator self-report | People do not usually over-report their own bad behaviour. Bias runs one way — downward. | Chapter 6 |
| Occupational headcount | Counting who does a job. Not an opinion about anything. | Chapter 9 |
| Contemporary legal record | Court cases, proclamations, gazettes. Written at the time, for other purposes, with consequences attached. | Chapter 3 |
| Administrative statistics | Real but noisy. Measure the reporting system as much as the phenomenon. | Chapter 7 |
| Contemporary testimony | Valuable, especially when detailed and consistent, but a single account is a single account. | Chapter 5 |
| Oral tradition | Preserves real memory and also grows in transmission. Needs corroboration. | Chapter 4 |
For every claim: who recorded this, when, and what did they gain by recording it?
A missionary describing untouchability gains a conversion story. A colonial official gains a justification for rule. A caste association describing its own high status gains rank. An activist describing an atrocity gains attention and possibly compensation. A political party describing anything gains votes.
This is not cynicism, and it does not mean any of them are lying. It means the strongest evidence is the kind where the recorder had nothing to gain — or, better still, something to lose.
Apply that test to the two directions of error.
Over-reporting requires someone to invent or inflate a harm. It has obvious motives — sympathy, compensation, political advantage — and it certainly happens.
Under-reporting requires only that a person in a village where the dominant community controls their employment, their water and their safety decides not to walk to a police station. It requires nothing to be invented at all.
Both errors are real. But one takes effort and the other takes silence, and the second is always more common. Any honest reading of Indian caste data has to start there.
3The Documented Events
Five episodes with dates, places, participants and paper trails. Not stories. Records.
These are chosen deliberately. Each was led by Indians, reported in Indian newspapers, and in several cases litigated in Indian courts. None depends on a missionary's account or a colonial officer's report.
1. The Upper Cloth Revolt — Travancore, 1813–1859
In the princely state of Travancore, women of lower-ranked communities — the Nadar or Channar community above all — were forbidden to wear an upper cloth in the presence of those of higher rank. The prohibition was a marker of status, and it was enforced.
Converts to Christianity began covering themselves, which produced decades of intermittent violence, market disturbances, and repeated royal orders — some granting partial rights, some withdrawing them. The conflict is known as the Channar Lahala, or Upper Cloth Revolt.
It was substantially resolved by a royal proclamation in 1859, which permitted the women of the community to cover themselves.
Note what kind of evidence this is. It is a dispute recorded in the administrative and legal record of an Indian princely state, running across nearly five decades, requiring multiple royal proclamations. That is not the sort of thing a historian invents later.
2. Vaikom Satyagraha — 30 March 1924 to 23 November 1925
Not to enter the temple. To use the roads around the Vaikom Mahadeva temple in Travancore, which lower-caste people were prohibited from walking on.
The satyagraha ran for roughly twenty months. Volunteers were arrested repeatedly. E. V. Ramasamy — Periyar — came from Tamil country to participate and was imprisoned, earning the title Vaikom Veerar. Gandhi visited in 1925. It ended with the opening of three of the four roads.
Twenty months, of organised non-violent protest, to walk down a street.
3. Mahad Satyagraha — 20 March 1927
On 20 March 1927, at Mahad in the Bombay Presidency, Ambedkar led a large gathering — accounts put it in the thousands — to the Chavdar tank, a public water source, to drink from it. Access was legally permitted; it was socially prohibited.
After they drank, caste Hindus ritually purified the tank using cow dung, cow urine, milk and curds — the standard purification of a polluted object.
Litigation over the tank followed and continued for a decade, ending in favour of the Dalit claimants in the Bombay High Court in 1937.
On 25 December 1927, at the Mahad conference, Ambedkar publicly burned a copy of the Manusmriti.
A protest can be dismissed as politics. The purification cannot.
Nobody performs an elaborate, expensive, publicly visible ritual to cleanse a body of water unless they sincerely believe it has been defiled — and unless the community around them agrees. It was not done to make a point to historians. It was done because those involved thought the water had become unusable.
That single act tells you more about how untouchability actually worked than a hundred pages of testimony. And it happened in 1927, in the presence of a legal system that had already declared the tank public.
4. Kalaram Temple Satyagraha — from 2 March 1930, Nashik
Beginning 2 March 1930, Ambedkar led a satyagraha for the right of Dalits to enter the Kalaram temple at Nashik. It continued, with interruptions, for over five years, and involved large processions, prolonged confrontations and arrests.
A related campaign at the Guruvayur temple in Kerala, led by K. Kelappan, ran through 1931–32.
5. The Temple Entry Proclamation — Travancore, 12 November 1936
On 12 November 1936, the Maharaja of Travancore issued a proclamation opening state-controlled temples to all Hindus regardless of caste.
Each of them is a negative-space proof. You do not campaign for twenty months to walk on a road that is open. You do not purify a tank that was never polluted. You do not proclaim a right that already exists. You do not litigate for ten years over water you were freely drinking.
The existence of the campaign proves the existence of the prohibition, independently of anybody's description of it. This is why these events are so much stronger as evidence than any account of how bad things were.
What they establish: in the first half of the twentieth century, in multiple regions of India, lower-caste people were prohibited from using public roads, public water and public temples — and these prohibitions were enforced strongly enough to require years of organised resistance and formal legal action to overturn.
What they do not establish: how universal any of it was, or how it compares with earlier centuries. Those remain open.
4The Breast Tax
One of the most repeated stories in Indian anti-caste politics, taken apart honestly — including the parts that do not survive.
If this book only produced evidence that supported one conclusion, you would be right not to trust it. So here is a chapter that goes the other way.
The story as it is usually told
In Travancore, lower-caste women were forbidden to cover their breasts, and were taxed if they did — a "breast tax," assessed according to the size of the woman's breasts by officials who came to measure them. A woman named Nangeli of Cherthala refused to pay. When the tax collector arrived, she cut off her own breasts and presented them to him on a plantain leaf, and bled to death. Her husband threw himself on her funeral pyre. The village came to be called Mulachiparambu.
It is a devastating story. It appears in articles, in artwork, in school discussions and in political speeches.
What the record supports
Travancore levied a range of poll taxes on lower-caste subjects, including one recorded as mulakkaram. Taxation that fell specifically on lower-caste people, and did not fall on higher-caste people, is documented in the state's own fiscal records.
The prohibition on lower-caste women covering their upper bodies is separately and thoroughly documented, as Chapter 3 set out — it produced forty-six years of conflict and required royal proclamations to resolve.
What the record does not support
Historians have raised serious questions about the popular version. The principal difficulties:
The specific incident rests on oral tradition. The story of Nangeli is transmitted locally and became widely known relatively recently. Contemporary documentary corroboration for the event as described has not been produced.
The nature of the tax is disputed. Several historians read mulakkaram as a poll tax — a head tax levied per adult member of certain communities — rather than a levy assessed on the body. On that reading it was a tax on being a member of a low-caste community, collected per person, which is a different thing from what the popular story describes.
The vivid details are exactly what oral transmission produces. The plantain leaf, the measurement, the husband on the pyre. These are the elements that make a story memorable, and memorability is precisely how details get added.
None of this proves the story false. It means the specific incident is not established, and it should not be presented as established.
What it costs the anti-caste side: a story it uses a great deal, which should be presented as tradition rather than documented history. Using contested material weakens a case that does not need it — and hands the other side a genuine catch.
What it costs the sceptical side, which is more: notice what remains standing after the story is set aside. The discriminatory taxation is documented. The clothing prohibition is documented. The forty-six-year conflict is documented. The royal proclamations exist.
Removing Nangeli removes an illustration. It removes nothing from the underlying record, because the underlying record was never resting on her.
And this is the pattern for the whole of Chapter 10. The exaggeration case scores real hits on the illustrations. It has almost nothing to say about the documents.
When any side of any argument offers you a story that is unusually vivid, unusually cruel, and unusually neat — with a named individual, a dramatic gesture and a perfect ending — be careful. Not because it is necessarily false, but because those are the properties that make stories survive retelling, whether or not they happened.
The evidence that actually holds is almost always duller. Tax registers. Proclamations. Court dates. Occupational tables. Nobody makes artwork about a proclamation, which is exactly why you can rely on it.
5Ambedkar's Own Testimony
What happened, in his own account, to a man with degrees from Columbia and the London School of Economics.
Ambedkar wrote a short set of autobiographical notes, published as Waiting for a Visa. They are among the most useful documents in this entire subject, for a reason that has nothing to do with sympathy.
They allow you to hold everything else constant.
The journey to Masur, 1901
Ambedkar, aged about nine, travelled with his brothers to meet their father. At the railway station the stationmaster, who had been helpful, asked who they were. On being told they were Mahars, his manner changed and assistance stopped.
No cart driver would take them. Eventually one agreed — for double the normal fare, and on condition that the boys drove the cart themselves, so that he would not have to sit with them. He walked alongside.
The school at Satara
- He was not permitted to sit on the classroom bench. He brought a gunny sack from home, sat on it, and carried it home again each day so that the school servant would not have to touch it.
- He was not permitted to touch the water tap. Water had to be poured for him by the school peon, from a height, without contact. In his own summary: no peon, no water.
- The village barber would not cut his hair. His sister cut it.
Baroda, 1917
By this point he held a doctorate from Columbia University and had studied at the London School of Economics. He had been invited to take up a senior post in the service of the Baroda state.
He could not find anywhere to live. No Hindu lodging would take him. He eventually obtained a room at a Parsi inn by giving a false name and pretending to be a Parsi. When the other residents discovered who he was, they confronted him and he was forced to leave. He sat in a public garden with his luggage and then left the city.
Its strength is that it is a controlled comparison. Almost every argument about caste disadvantage runs into the objection that poverty, not caste, explains the outcome. Baroda in 1917 removes that objection entirely. Ambedkar had qualifications almost nobody in India possessed, an offer of senior employment, and money. He was still refused a room.
That is the same logical structure as the experiments in Chapter 8, conducted eighty years earlier by accident.
Its limit is that it is one man's account. A single testimony, however detailed and however credible, cannot establish how widespread anything was. It establishes that these things happened to him.
Which is why this chapter sits here rather than at the end. Testimony shows you what the thing looked like. It cannot tell you how much of it there was. For that you need Chapters 6 to 9.
6The Confession
In 2011–12, researchers asked more than forty thousand Indian households a question almost nobody asks. The answers are the most important evidence in this book.
Almost all argument about caste discrimination is an argument about victims' accounts — how reliable they are, how representative, how politically motivated.
The India Human Development Survey did something different. It asked the other side.
The second India Human Development Survey (IHDS-II), conducted in 2011–12 by the National Council of Applied Economic Research and the University of Maryland, covered over 42,000 households, nationally representative across class and social group.
It asked respondents directly: does anyone in your family practise untouchability? — and followed up with a concrete test: would it be acceptable for a Scheduled Caste person to enter your kitchen or use your utensils?
The answers
By community, the reported practice was highest among Brahmin households (52%), followed by OBC households (33%) and other forward-caste households (24%). Scheduled Tribe households reported 22% and Scheduled Caste households 15%.
By religion, roughly 30% of Hindu households, 23% of Sikh, 18% of Muslim and 5% of Christian households reported the practice — with Jain households reporting the highest figure of all at 35%.
The same survey found that around 5% of Indian marriages were inter-caste.
Go back to the seven objections in Chapter 1 and try them against this.
"Colonial and missionary sources are biased." This is an Indian research institution and an American university, in 2011–12. No colonial officials, no missionaries.
"Stories get embellished." There are no stories here. There is a question and a percentage.
"Crime data measures reporting." Nobody reported anything. No police station, no complaint, no compensation.
"Victims exaggerate." These are not victims. These are the people doing it, describing what they do.
And that is the point on which the whole chapter turns. Consider which way the error runs. Untouchability has been unlawful in India since 1950. Admitting to it in a survey carries social and legal risk and no benefit whatsoever.
So the plausible bias is downward. Twenty-seven per cent is a floor, not a ceiling.
Against the simple narrative: 15% of Scheduled Caste households and 22% of Scheduled Tribe households also reported practising untouchability. Caste is not a two-team game with clean villains. The people at the bottom of one ladder are frequently standing on someone else. Part 3, Chapter 7 explained why.
Against the "it is over" narrative: the figure for Brahmin households is 52%, and the inter-caste marriage rate is around 5%. Both of those are measurements of the present, taken with modern methods, published by mainstream institutions.
If you want one number to carry out of this entire series, it is 27% — and the knowledge that it is the number people were willing to say out loud.
7The Crime Data
The numbers everyone quotes, with every honest caveat attached — including the ones the side quoting them leaves out.
The National Crime Records Bureau publishes annual figures for offences registered under the laws protecting Scheduled Castes and Scheduled Tribes. These numbers are cited constantly and understood rarely.
Cases of crime against Scheduled Castes rose from 42,793 in 2018 to 50,291 in 2020. Uttar Pradesh consistently reported the highest number, with 12,714 cases in 2020.
Crimes against Scheduled Tribes rose from 10,064 in 2022 to 12,960 in 2023 — an increase of nearly 29% in one year.
On the processing of those cases: of roughly 50,000 SC cases in 2020, charge sheets had been filed in about 39,000, while investigation remained pending in around 19,700.
Now the caveats — all of them
A rise can mean improvement. More police stations, more awareness of the law, more willingness to come forward, and better recording all raise the number without any change in underlying behaviour. A rising line can be a sign of a system working better.
Registration is not proof. A registered case is an allegation. Some are false; some are ordinary disputes filed under a stronger statute; some are genuine and fail for want of evidence. This is true of every crime category in every country.
The aggregation problem is real. A case is registered under these laws when the victim is SC or ST. That does not establish caste as the motive. In a country where a sixth of the population is SC, some proportion of ordinary crime will involve SC victims for reasons unconnected to caste.
And under-reporting is severe and runs the other way. The person deciding whether to file a complaint frequently depends on the accused's community for work, for water, for credit and for physical safety, and has to go to a police station that may be staffed by members of it. Every study of this subject finds substantial non-reporting.
I do not think NCRB atrocity totals prove very much on their own, and I think both sides misuse them.
People who cite the raw numbers as proof of a worsening crisis are ignoring that better reporting inflates the same line. People who cite false cases as proof the whole thing is manufactured are ignoring that under-reporting is far larger than over-reporting and always has been.
What the data does establish is narrower and still substantial: that tens of thousands of cases serious enough to be registered under these specific statutes are filed every year, that the number is not falling, and that the state's own machinery is processing them in volume.
That is a floor on the phenomenon, not a measurement of it. If you want a measurement, Chapters 6, 8 and 9 are far better evidence — which is precisely why I have not built this part on the crime figures.
8The Experiments
Identical CVs. Different surnames. Nothing else changed. This is the cleanest evidence that exists on the subject.
Everything so far has depended on somebody reporting something. This chapter does not.
A correspondence study works like this. Researchers write fictitious job applications that are identical in every respect — same degree, same marks, same experience, same language, same formatting. Then they change one thing: the name. And they send them to real employers advertising real jobs, and count who gets called for interview.
There is no victim to doubt, no memory to fade, no politics, no compensation, and no police station. The employers do not know they are in a study. The only variable is the one deliberately altered.
Beginning in October 2005 and running for 66 weeks, the researchers responded to job advertisements in national and regional English-language newspapers. The posts were entry-level positions in the private sector requiring a university degree.
Sets of applications were sent that were equivalent in qualifications and experience and differed only in the applicant's name — signalling upper-caste Hindu, Dalit or Muslim identity.
Ten, six, three. Same degree, same marks, same experience. Different name.
1. It is in the private sector. Reservation does not operate there. This is the part of the economy that the abolition case describes as running on pure merit — and here, at the very first filter, before anyone has met anyone, merit is being overridden by a surname.
2. It is at the top of the labour market. These were graduate jobs, in English, in cities. If discrimination survives among educated urban employers hiring degree-holders, the claim that it is confined to backward villages does not hold.
3. Class cannot explain it. The applications were equally qualified. There was no income, no school ranking and no accent to differ on. The name was the only signal.
4. Muslim applicants did worst. Three calls for every ten. This connects directly to Part 4 — the population that faces the sharpest labour-market penalty here is also the one excluded from Scheduled Caste status by the 1950 Order.
Correspondence studies measure the first filter — the decision to call someone for interview. They cannot tell you what happens afterwards, and they cannot tell you the employer's reasoning. An employer might be acting on prejudice, or on a belief about which candidates will fit a team, or on an assumption about English fluency attached to a name.
That distinction matters morally. It matters much less to the applicant, who does not get the call either way.
Similar audit methods have been applied in India to rental housing, with comparable findings — landlords responding differently to identical enquiries under different names.
9The Sewers
One occupation. One statistic. It ends the argument about whether any of this is still happening.
Every argument in this book about the past can be contested. Sources have motives, memories drift, statistics measure systems. This chapter contains no such difficulty.
Manual scavenging is the manual handling of human excreta — cleaning dry latrines, and entering sewers and septic tanks by hand to clear blockages. It has been prohibited by law in India, most recently under legislation of 2013.
People are still doing it, and people are still dying doing it.
Figures reported as of 2024 give 42,594 Scheduled Caste persons identified as engaged in the practice, against 421 Scheduled Tribe and 431 Other Backward Class persons.
That is roughly ninety-seven per cent from one category comprising around a sixth of the population.
Stop and consider what kind of statistic that is.
It is not a survey. Nobody was asked their opinion. It is not a crime report, so there is no question of false complaints. It is not a historical account, so there are no missionaries and no colonial officials. It is a count of who does a particular job.
The deaths
Government-linked figures record 63 deaths in 2023 and 52 in 2024 of sanitation workers cleaning sewers or septic tanks. By mid-2025 the figure reported was 72, including 10 in June alone.
The National Commission for Safai Karamcharis has recorded 471 deaths between 2019 and 31 October 2025, and 1,313 sewer and septic tank deaths between 1993 and mid-2025.
Independent counts are substantially higher. The Safai Karamchari Andolan recorded 116 deaths in 2024 against an official figure of 55, and 121 in 2025 against 46 officially. Its convenor, Bezwada Wilson, has said the official count is severely under-reported.
The distinction relied upon is between "manual scavenging," defined narrowly, and "hazardous cleaning of sewers and septic tanks." Deaths are recorded under the second heading.
Whatever the legal merit of the distinction, its effect on the public record is to make a category of death disappear from the answer to a question about it.
The claim that caste discrimination is a historical matter, exaggerated by people with political motives, has to account for this.
Not a claim about the past. A job, being done now, in which one community out of thousands supplies about ninety-seven per cent of the workforce — and in which people suffocate to death every few days, in numbers the state and the activists cannot agree on because the state's own count is contested.
There is no reading of that which is compatible with a labour market allocating dangerous work by anything other than birth.
And there is a second point, which is the one I would put to the abolition side directly. Their proposal is that caste should be removed from official records. If caste is not recorded, this statistic cannot be produced. The 97% figure exists only because somebody counted by caste.
Deleting the category does not change who goes into the sewer. It changes whether anyone can prove it.
10The Exaggeration Case, Tested
The seven arguments from Chapter 1, taken one at a time. Three of them survive.
Correct as a general principle, and it is why this part avoids relying on them. But it does not touch the evidence presented here. The Travancore proclamations are the records of an Indian princely state. Mahad, Vaikom and Kalaram were Indian-led campaigns reported in the Indian press and litigated in Indian courts. The IHDS is an Indian research council with an American university. The correspondence study was run by Indian researchers on Indian employers.
VERDICT: Valid methodological warning. Does not apply to the case as built here.True, and Chapter 4 demonstrates it rather than denying it. Some widely circulated stories are not established, and should not be presented as established.
But note the scope of the win. It affects illustrations, not documents. Setting aside every contested story leaves the proclamations, the court cases, the tax records, the surveys, the experiments and the occupational counts untouched.
VERDICT: WINS. And wins less than it appears to, because the case never rested on the stories.Correct, as Chapter 7 concedes at length. Registered-case totals are weak evidence, and I have declined to build on them for exactly this reason.
The reply is that under-reporting is larger than over-reporting, for structural reasons — but this is an argument about direction, not a rescue of the data.
VERDICT: WINS. NCRB totals should not be used as a measurement by anyone.Also correct, and also conceded. A dispute between two people of different castes is not automatically a caste dispute, and national totals cannot separate the two.
This is a genuine and underrated methodological point, and the people who make it are usually more careful than the people who dismiss it.
VERDICT: WINS, on the crime data specifically. Does not reach Chapters 6, 8 or 9.Generally a fair concern, and fatal to a lot of published survey work. It runs into a specific problem here.
The IHDS asked people to admit to something unlawful, socially disapproved and personally discreditable, with no benefit for saying yes. Wording effects on that kind of question push responses down. To argue the 27% figure is inflated, you would have to explain why a substantial number of households falsely confessed to practising untouchability.
VERDICT: FAILS against this particular survey, and fails badly, because the incentive runs the wrong way.This is the strongest argument the sceptical side has, and it is right.
Untouchability is unlawful. Temple entry, road access and water access are legally settled questions. There are Dalit judges, chief ministers, industrialists and a President of India. Urban anonymity has dissolved enforcement mechanisms that were unbreakable in a village. Describing 2026 in the vocabulary of 1926 is inaccurate, and the inaccuracy matters because policy is made for now.
The reply is not that nothing changed. It is that 27%, 5% inter-caste marriage, ten-six-three, and 97% are all measurements of the present, taken with modern methods. Change is real and incomplete, and both halves of that sentence are load-bearing.
VERDICT: WINS on the framing. Loses on the conclusion, because the present-day numbers exist.The premise is true and Part 3 said so. Roman slavery, European serfdom, hereditary aristocracies everywhere. India's hierarchy was probably not uniquely cruel by pre-modern standards.
The problem is that this is an argument about the past being deployed against a claim about the present. The others are gone. Ninety-seven per cent of manual scavengers in 2024 being from one category is not a fact about the Roman Empire.
And Part 3, Chapter 9 asked the better version of this question: not "was India uniquely bad" — probably not — but "why did India's system uniquely survive." That is a real and interesting question. This argument is not asking it.
VERDICT: True premise, irrelevant conclusion. It answers a comparative question nobody is asking.11Nine Questions Nobody Asks
The blind spots in this stage of the argument.
1. Why does nobody quote the 27% figure?
It is the single strongest piece of evidence on the subject. It is from a mainstream survey, a large sample, and a perpetrator self-report. It should be in every discussion of caste in India.
It is barely cited by either side. The abolition side has an obvious reason. The retention side does not use it either — possibly because 15% of SC households and 22% of ST households also reported the practice, which complicates a simpler story.
2. Is a caste-blind system even measurable?
The 97% manual-scavenging figure exists because someone recorded caste. So does the ten-six-three result, which required knowing which names signalled what.
Remove caste from official records and both become impossible to produce. You would not have ended the pattern; you would have ended the ability to detect it. Any serious abolition proposal has to answer what replaces the measurement.
3. Why is discrimination worst in the sector with no reservation?
The correspondence study was conducted on private employers, where reservation has never applied. That is where the ten-six-three gap appears.
This is awkward for both sides. It undercuts the claim that reservation causes resentment which causes discrimination, since the effect is strongest where reservation is absent. And it undercuts the claim that reservation is a sufficient remedy, since it does not reach the sector where most jobs are.
4. Why do Muslim applicants do worse than Dalit applicants?
Three calls to six. The group facing the sharpest labour-market penalty in the best experiment we have is the one excluded from Scheduled Caste protection by the 1950 Order.
Almost no one in the 2026 argument has connected those two facts, though both were available.
5. What is the honest estimate of false cases?
Nobody knows, and both sides prefer it that way. The sceptical side implies the figure is large without producing it. The other side treats the question as illegitimate.
It is a perfectly legitimate question with a knowable answer, and the fact that no one has funded the study is itself informative.
6. Why is the strongest evidence the least emotional?
The material that moves people is Nangeli, the purified tank, the boy on the gunny sack. The material that actually establishes the case is a percentage from a household survey and a table of who works in sewers.
This is a general problem with public argument, not a problem with caste. The persuasive evidence and the reliable evidence are rarely the same evidence, and confusing them is how both sides end up defending things they did not need to defend.
7. What would falsify the claim that caste discrimination persists?
A fair question to put to my own side, so here is my answer. Inter-caste marriage rates rising to something like population-random. Correspondence studies returning no callback gap. The manual-scavenging workforce reflecting population shares. Untouchability self-reports near zero.
Those are measurable, and none of them is currently close. But naming them matters — a claim that nothing could disprove is not a claim.
8. And what would falsify the claim that it has ended?
Fairness requires the reverse question. The answer is that the four measures above are all currently in the wrong place, so the "it has ended" claim is already falsified on its own terms — unless one argues the measures are wrong, which requires an argument about measurement rather than about history.
9. Does establishing the injury settle the remedy?
No, and this is the most important limit on everything in this book.
Part 6 answers the question "did it happen and is it still happening." It says nothing about whether reservation is the right response, whether it works, or who should receive it. A proven disease does not validate a particular medicine.
That is Part 11, and readers who take this part as settling the reservation argument will have misread it in exactly the way I have spent six books asking people not to.
12The Verdict
The scorecard first, as evidence. Then my judgement, clearly marked as judgement.
| Question | Status | Best current answer |
|---|---|---|
| Were lower castes barred from public roads, water and temples? | Settled | Vaikom, Mahad, Kalaram, 1936 Proclamation |
| Were there discriminatory taxes and clothing prohibitions? | Settled | Travancore records; 46-year Upper Cloth conflict |
| Is the Nangeli story established history? | No | Oral tradition; disputed by historians |
| Did education or wealth exempt a person? | No | Ambedkar, Baroda, 1917 |
| Is untouchability practised today? | Yes | 27% of households admit it (IHDS 2011–12) |
| Do NCRB totals measure the phenomenon? | No | They measure the reporting system |
| Does caste affect hiring where reservation does not apply? | Yes | 10 : 6 : 3, identical CVs, private sector |
| Is dangerous degrading work allocated by caste now? | Yes | ~97% of manual scavengers are SC (2024) |
| Has there been enormous improvement? | Yes | Legal, political and urban change is real |
| Does any of this prove reservation works? | No | Different question — Part 11 |
Did caste discrimination really happen? Yes, and it is not a close question. The evidence is Indian, contemporaneous, legal and in several cases produced by the perpetrating side. Twenty months of protest to walk on a road. A public tank purified with cow dung after Dalits drank from it. A ten-year court case about that tank. A proclamation required to let people into temples. A man with a Columbia doctorate unable to rent a room.
Has it been exaggerated? In places, yes — and I have shown one case in detail rather than waiting to be caught. Specific stories have grown in the telling. Crime totals are routinely misused. The vocabulary of 1926 is applied to 2026 in ways that are simply inaccurate.
Is the exaggeration the substance? No. And this is where the sceptical case fails. Remove every contested story, discard the NCRB data entirely, accept every methodological objection in Chapter 1, and you are still holding: 27% of households admitting to untouchability in 2011–12; around 5% inter-caste marriage; ten interview calls against six and three for identical CVs in the private sector; and 97% of manual scavengers from one category in 2024.
Those four numbers were produced by mainstream institutions using standard methods, and every one of them measures the present.
What I think the honest summary is: the historical injury is established beyond reasonable dispute. The improvement since 1950 is enormous and is routinely understated by the side that ought to be proudest of it. And the residue is not a memory — it is measurable, current, and concentrated exactly where you would predict: in marriage, in hiring, and in who cleans the sewer.
It does not tell you that reservation works, or that it is well designed, or that it reaches the right people, or that it should continue in its present form. I have not examined any of those questions yet and this part contains no evidence about them.
An injury being real is the first of five questions from Part 1. It is a necessary condition for a remedy and nowhere near a sufficient one.
13The Five-Line Nichod
Everything in this book, compressed. If you remember nothing else, remember these.
PART 6 — THE EVIDENCE FILE
- 27% of Indian households admitted practising untouchability. IHDS-II, 2011–12, over 42,000 households. Not victims — perpetrators, describing themselves, about something unlawful with no benefit for saying yes. The bias runs downward, so that is a floor. Brahmin households 52%; SC households 15%.
- The negative-space proofs cannot be argued with. You do not campaign twenty months to walk on a road that is open (Vaikom). You do not purify a tank with cow dung unless you believe it was defiled (Mahad, 1927). You do not proclaim a right that already exists (Travancore, 1936).
- Ten, six, three. Identical CVs sent to private employers over 66 weeks. Ten interview calls for an upper-caste name, six for a Dalit name, three for a Muslim name. Graduate jobs, English newspapers, no reservation in that sector at all.
- Around 97% of manual scavengers are Scheduled Caste. 42,594 SC against 421 ST and 431 OBC, reported 2024. Not a survey, not a crime report — a count of who does the job. Dozens die in sewers every year, and the state and the activists cannot agree on how many.
- Some of it is exaggerated, and the exaggeration is not the substance. The Nangeli story is not established history and should not be presented as such. Crime totals measure reporting. But strip out every contested story and every weak statistic, and the four numbers above are still standing — and all four measure the present, not the past.
Coming next
Real quotes, full context, no cropping. Ambedkar on the annihilation of caste and on why he came to distrust political safeguards without social change. Gandhi, and how his position moved across thirty years — including the parts his admirers leave out. Phule and Periyar. Vivekananda, quoted by everyone and read by few. Savarkar's written attacks on caste, and exactly where he stopped. Golwalkar and the RSS position, then and now. Nehru, Patel, Lohia, Kanshi Ram, Mayawati. And the modern voices on both sides — Kancha Ilaiah, Chandrabhan Prasad, Ashwini Deshpande, Arvind Panagariya, Yogendra Yadav, and Thomas Sowell on quotas worldwide.
Each figure gets a strongest point and a weakest point. Nobody is available to be borrowed cheaply.
A closing thought: the most useful thing I found in this part was not any single number. It was that the reliable evidence and the moving evidence are almost never the same evidence. Nobody makes artwork about an occupational table. Which is exactly why you can rely on it. — L.S.
Sources & further reading — Part 6
Every factual claim in Part 6, and where it comes from.
The documented events
Channar Lahala / Upper Cloth Revolt, Travancore, c. 1813–1859, resolved substantially by royal proclamation in 1859. · Vaikom Satyagraha, 30 March 1924 – 23 November 1925; participation and imprisonment of E. V. Ramasamy (Periyar); Gandhi's visit in 1925. · Mahad Satyagraha, 20 March 1927 — Ambedkar leads thousands to the Chavdar tank; subsequent ritual purification of the tank by caste Hindus; litigation concluding in favour of the Dalit claimants in the Bombay High Court in 1937; public burning of the Manusmriti at Mahad on 25 December 1927. · Kalaram Temple Satyagraha, Nashik, from 2 March 1930, continuing more than five years. · Guruvayur Satyagraha, 1931–32, led by K. Kelappan. · Temple Entry Proclamation, Travancore, 12 November 1936.
The breast tax
Discriminatory poll taxation in Travancore, including the levy recorded as mulakkaram, appears in the state's fiscal records; several historians read it as a head tax on members of certain communities rather than a levy assessed on the body. The Nangeli narrative is transmitted through local oral tradition and has been questioned by historians on grounds of documentary corroboration. This book treats the taxation and the clothing prohibition as established and the specific incident as contested.
Ambedkar's testimony
B. R. Ambedkar, Waiting for a Visa — autobiographical notes covering the journey to Masur (1901), the school at Satara (the gunny sack; water poured by the peon; the barber), and Baroda (1917), where he was unable to obtain lodging and was forced to leave a Parsi inn after his identity became known.
The survey
India Human Development Survey II (IHDS-II), 2011–12, conducted by the National Council of Applied Economic Research with the University of Maryland; over 42,000 households, nationally representative. Question on family practice of untouchability with a follow-up on kitchen and utensil access. Combined responses: 27% of households nationally; 30% rural, 20% urban. By community: Brahmin 52%, OBC 33%, other forward castes 24%, ST 22%, SC 15%. By religion: Hindu 30%, Sikh 23%, Muslim 18%, Christian 5%, Jain 35%. The same survey found approximately 5% of marriages to be inter-caste.
Crime data
National Crime Records Bureau. Crimes against Scheduled Castes: 42,793 registered cases in 2018 rising to 50,291 in 2020; Uttar Pradesh highest with 12,714 in 2020; of roughly 50,200 cases in 2020, charge sheets filed in about 39,075 with investigation pending in about 19,696. Crimes against Scheduled Tribes: 10,064 in 2022 rising to 12,960 in 2023, an increase of 28.8%. This book treats registered-case totals as a floor on the phenomenon rather than a measurement of it.
The correspondence study
Sukhadeo Thorat, Paul Attewell and Firdaus Rizvi, "The Legacy of Social Exclusion: A Correspondence Study of Job Discrimination in India," Economic and Political Weekly, 13 October 2007. Fieldwork from October 2005 over 66 weeks; responses to advertisements in national and regional English-language newspapers for entry-level private-sector posts requiring a university degree; equivalently qualified applications differing only in the name. Result: for every ten interview invitations to an upper-caste Hindu name, six to a Dalit name and three to a Muslim name. Comparable audit methods have been applied to rental housing in India with similar findings.
Manual scavenging
Composition as reported for 2024: 42,594 Scheduled Caste, 421 Scheduled Tribe and 431 Other Backward Class persons identified as engaged in the practice — approximately 97% Scheduled Caste. · Deaths from hazardous cleaning of sewers and septic tanks: 63 in 2023 and 52 in 2024 on government-linked figures, with 72 reported by mid-2025 including 10 in June. National Commission for Safai Karamcharis: 471 deaths between 2019 and 31 October 2025, and 1,313 between 1993 and mid-2025. · Safai Karamchari Andolan independent counts: 116 deaths in 2024 against an official 55, and 121 in 2025 against an official 46; Bezwada Wilson has stated that official counts are severely under-reported. · The government's position in Parliament has distinguished "manual scavenging" from "hazardous cleaning," with the consequence noted in Chapter 9. · Prohibition of Employment as Manual Scavengers and their Rehabilitation Act, 2013.
A note on what is missing — and on corrections
Part 6 establishes only that the injury occurred and that measurable discrimination persists. It makes no claim about whether reservation is an effective or appropriate remedy, who should receive it, or how it has performed. Those are Parts 11, 12 and 14. Anything marked CONTESTED CLAIM stays contested until evidence is produced — including claims made by people I might agree with.
Where official and independent counts of sewer deaths differ, both are given rather than a preferred figure. NCRB series are cited to the most recent year located for each category and are not directly comparable across categories. If you find an error, I want to know. Corrections will be carried in later parts.