The Part The Series Was Built For
Eleven parts of evidence have been gathered so that this part could be written properly. Everything here is drawn from what came before, and every claim in it has a source somewhere in Parts 1 to 11.
The two cases in Chapters 2 and 3 are written as advocacy. Not as summaries of what people say — as the argument itself, made as well as I am able to make it, in the voice of someone who believes it.
That is harder than it sounds and I want to be honest about the difficulty. Writing the case you hold is easy. Writing the case you do not hold — and writing it better than its own advocates usually manage, using the strongest evidence available including evidence they have never found — is genuinely uncomfortable. Twice while drafting Chapter 2 I had to stop, because the argument was working.
Three notes on method.
Each case is allowed its best evidence, not its most popular. The abolition case here does not claim that reserved candidates have taken over the government, because Part 11 showed they have not. It makes the arguments that survive. The same discipline applies to the other side.
Then they are cross-examined. Chapters 4 and 5 put ten questions to each, with the strongest reply each can give. Some replies are good. Some are visibly weak, and I have not strengthened them beyond what the evidence supports.
And there is no verdict here. Chapter 7 marks every contested claim as fact, interpretation or unsupported, and stops. My own judgement is Part 14, clearly labelled, after Part 13 has looked at what other countries did.
One last thing, and it is the most important sentence in this book. Chapter 8 is a list of things neither case addresses. It is longer than I expected when I started, and by the time I finished writing it I was no longer sure that the argument between these two positions is the argument that matters.
— Lovepreet Singh
July 2026
1The Rules Of This Part
What a steelman actually is, why it is the opposite of what social media does, and the four rules I have followed.
A strawman is the weakest version of an opponent's argument, built so it can be knocked down. Almost all public argument about reservation consists of two strawmen fighting.
A steelman is the opposite: the strongest version of the argument, stated so well that its own supporters would say "yes, that is what I mean, only better."
There is a practical reason to do this, and it is not politeness. If you can only defeat the weak version, you have not defeated anything. The person you are arguing with will simply say that is not what they meant — and they will be right.
The four rules
- Best evidence, not most popular evidence. Where a side's favourite claim has been shown false earlier in this series, I have dropped it and used something that survives. The abolition case in Chapter 2 does not say reserved categories dominate the services, because Part 11 showed they do not. It says something harder to answer instead.
- Equal length, equal force. Ten arguments each. Written in the same register, with the same access to the evidence.
- No sneering. No "of course", no "obviously", no rhetorical questions designed to embarrass. Each case is written by someone who means it.
- Cross-examination with real replies. Chapters 4 and 5 put the hard questions. Where a side has a good answer, it is given. Where it does not, I say so rather than inventing one.
Read both before forming a view on either. If you read Chapter 2 and stop, you will finish convinced. If you read Chapter 3 and stop, the same. That is what a working steelman does, and it is why they are printed consecutively with no commentary in between.
Notice also what happens to your own reaction. Most readers find one chapter obviously right and the other suspiciously clever. Which one you find clever tells you where you already stood.
2Case A — Abolish It
The case for ending caste-based reservation, written as its best advocate would write it.
Let us begin by conceding everything that is true, because the case does not need the falsehoods.
Caste discrimination was real, systematic and brutal. It is still measurable. Twenty-seven per cent of Indian households admit to practising untouchability. Ninety-seven per cent of manual scavengers come from one category. Identical CVs with different surnames produce ten interview calls against six and three. None of that is in dispute, and anyone who disputes it is not making this argument — they are making a worse one.
The case that follows accepts all of it, and argues that caste-based reservation is nevertheless the wrong instrument and should be replaced.
1. A remedy with no exit condition is not a remedy
Article 16(4) permits reservation for a class "not adequately represented in the services." That is a condition. Conditions end when they are satisfied.
In seventy-six years, no government, no court and no commission has ever defined what adequate representation would look like. There is no number, no threshold, no test. Which means the condition on which the entire exception rests cannot be met, not because the goal is distant but because it has never been specified.
A measure that cannot be satisfied is not a remedy. It is a permanent structure wearing a remedy's clothes. And the framers knew the difference — they wrote an expiry into Article 334 and Ambedkar told the Assembly that reservation must be confined to a minority of seats. Both limits are gone.
2. The state is the last institution keeping caste on file
Consider what the Indian state actually requires. Caste registered at birth. Declared at school admission. Certified for a college application. Carried into employment records. Recorded again for a child's application a generation later.
Every family in India therefore has a standing administrative and financial reason to preserve, document and transmit its caste identity. We ask people to prove their caste at every threshold of their lives, and then express surprise that caste has not faded.
A society can forget a category the state has stopped writing down. It cannot forget one the state demands in triplicate.
3. Elite capture is not an accusation. It is a judicial finding.
This is no longer contested. The Supreme Court warned about it in 1963. It built the creamy layer around it in 1992. It extended that principle to Scheduled Castes in 2018. And in 2024 a seven-judge bench, by six to one, permitted the Scheduled Caste category to be split — because the benefits inside it had concentrated so unevenly that a single category could no longer be justified.
The child of a senior officer and the child of a landless labourer hold the same certificate and compete in the same pool. Telangana and Karnataka have now legislated to address this within the Scheduled Castes. Nothing has been done for the 3,743 castes inside the OBC quota — because the dominant communities within it have the political power to prevent it.
A policy whose benefits reliably concentrate at the top of each protected category is not reaching the people it names.
4. The categories are incoherent, and demonstrably so
The EWS threshold of ₹8 lakh was copied from the OBC creamy layer figure. When the Supreme Court asked what study supported it, the government could not point to one. At that level the overwhelming majority of eligible households qualify.
Jat communities were added to the central OBC list in 2014 and removed by the Court in 2015. Total reservation is 59.5% centrally and about 62% in Maharashtra, both past a ceiling described as constitutional. Bihar's 65% was struck down; the centre's 59.5% stands — because one came with an amendment and the other did not.
This is not a system with imperfections. It is a set of categories that are arbitrary at the threshold, unstable in membership, uncapped in total, and unevenly distributed inside.
5. The cost falls on people who did nothing
A poor general-category family in Bihar inherits no land, no capital and no coaching money — only a surname. The correction for a historical wrong is levied on individuals selected by birth, which is the same principle the wrong itself rested on.
This is not a hypothetical grievance. Parliament acknowledged those families exist when it created EWS in 2019, and the Supreme Court upheld it. The question is not whether they exist. It is why the burden of repair should fall on them rather than on the state's revenue.
6. Seventy-six years, and the top is still unrepresented
Group A of the central government, January 2024: Scheduled Castes 14.2%, Scheduled Tribes 6.54%, Other Backward Classes 19.14%. Every one below quota.
Read that as a defender must. If the instrument worked as claimed, three generations would have produced representation at senior level. It has not. Either quotas cannot reach the level where decisions are made, or something else is blocking advancement that a quota at entry does not touch.
Either way, continuing an instrument that has failed at the level that matters, on the grounds that it must eventually work, is faith rather than policy.
7. The prize is shrinking, and the fight is over the wrong thing
Reservation applies to government employment. Government employment is a small and falling share of Indian jobs. The correspondence study that measured discrimination most cleanly measured it in the private sector, where reservation has never applied and where most Indians work.
A policy that captured a share of a shrinking market, while the measured harm concentrates in the market it never entered, is not a solution to the problem it is defended by.
8. Permanent labelling entrenches the identity it aims to dissolve
The stated goal of the anti-caste movement, from Ambedkar onward, was the annihilation of caste — not a better position within it.
A system that assigns every citizen a caste category at birth and attaches material consequences to it does the opposite. It makes caste the most consequential fact about a person's public life. Whatever this achieves, it is not annihilation.
9. India already runs the alternative, and it is lawful
An economic quota was struck down in 1992 because poverty alone could not constitute a backward class. In 2019 Parliament amended the Constitution, and in 2022 the Supreme Court upheld it.
The constitutional obstacle no longer exists. A caste-blind, need-based system is available, has been running for seven years, and can be built properly rather than by copying a number from another scheme.
10. And the strongest argument is Ambedkar's own
He told the Constituent Assembly on 30 November 1948 that reservation must be confined to a minority of seats, or the principle of equality would be destroyed. Central reservation is 59.5%. Maharashtra is at 62%. Tamil Nadu at 69%.
The limit set by the architect of the policy has been crossed, and the people who invoke his authority most often are past his line. If his ceiling is not binding, his authority cannot be either.
The injury was real and remains measurable. But the instrument has no exit condition, keeps caste administratively alive, concentrates benefit at the top of each category, rests on incoherent thresholds, has not delivered where power sits, operates on a shrinking sector, and has crossed the limit its own architect set.
The claim is not that nothing is wrong. It is that this is the wrong tool, that a need-based system is now lawful, and that seventy-six years is long enough to conclude that a measure which cannot define its own success will never end.
3Case B — Keep It
The case for retaining caste-based reservation, written as its best advocate would write it.
Let us also begin by conceding everything that is true.
The categories are badly built. The ₹8 lakh line is indefensible. Benefits have concentrated inside protected groups, and the Supreme Court was right to permit sub-classification. Thousands of reserved posts sit empty. Representation at senior levels has not been achieved. And the policy has done nothing about land, capital or ownership.
All of that is accepted. The case that follows argues that these are reasons to rebuild the instrument, not to remove it.
1. It works, and there is causal evidence
This is the strongest single finding in the entire argument and it is rarely stated. Caste groups newly added to the Scheduled Caste list experienced, on average, a rise of seven to eight rank points in upward mobility over the following twenty years — measured through a natural experiment, published in a leading economics journal in 2024.
That is not correlation. It is a comparison between similar groups, one of which received the policy. Against a national background in which intergenerational mobility has been flat and low for decades, the groups that received reservation moved.
Any case for abolition must therefore argue that a measure which demonstrably works should be ended for other reasons. That is a much heavier burden than the one usually carried.
2. The injury is current, not historical
Twenty-seven per cent of Indian households told researchers that their own family practises untouchability — a self-report of unlawful conduct, with no benefit for saying yes, which means the true figure is higher. Fifty-two per cent among Brahmin households.
Around five per cent of marriages are inter-caste. Roughly ninety-seven per cent of identified manual scavengers are Scheduled Caste. Identical CVs produce ten interview calls for an upper-caste name, six for a Dalit name and three for a Muslim name.
Every one of those measures the present. This is not repair for something that ended in 1950.
3. The condition for ending has not been met — on its own terms
Article 16(4) is triggered by inadequate representation in the services. Seventy-six years on, every reserved category remains below quota in Group A: Scheduled Castes 14.2%, Scheduled Tribes 6.54%, Other Backward Classes 19.14%.
The abolition case treats this as proof of failure. Read it again. It is proof that the condition still obtains. A remedy is withdrawn when the problem it addresses has been solved, not when it has proved harder than expected.
4. Removing belief has been tried four times and failed
Buddhism rejected birth-based worth twenty-five centuries ago. Islam and Christianity entered India with founding texts that reject it. Sikhism was born in Punjab specifically to break it — langar, common surnames, initiation from a common bowl, scripture containing the hymns of a leather-worker and a butcher.
Caste is inside all four of them today. Ashraf, Ajlaf and Arzal among Muslims, documented by the government's own Sachar Committee. Separate pews and cemeteries among Christians. Mazhabi and Ravidasia Sikhs.
Four religions changed what millions of Indians believed, and the structure survived. The proposal that removing a column from a government form will succeed where that failed requires an argument nobody has made.
5. Deleting the record deletes the evidence, not the practice
We know that ninety-seven per cent of manual scavengers are Scheduled Caste because somebody counted by caste. We know the Group A shortfall because the government publishes composition by category. Telangana and Karnataka could sub-classify only because the data existed.
Remove caste from official records and none of those figures can ever be produced again. The pattern would continue and become invisible.
Note also what the state already does not measure. Seven years into EWS, there is no published figure for who has received it. That is what caste-blind administration looks like in practice: not neutrality, but the absence of any way to check.
6. Reserved posts sit empty, which destroys the central grievance
More than 2,750 reserved teaching posts were vacant in central universities as of November 2024. A further 4,297 reserved academic positions were closed after recruitment through a declaration that no suitable candidate had been found.
Those places went to nobody. The "they took my seat" grievance requires an occupant, and in thousands of cases there was none.
And the same discretion does not produce the same outcome for unreserved posts at the same institutions. Whatever explains that, it is not an excess of reserved appointments.
7. The economic alternative has been tried, twice, and is worse
An economic quota was introduced by executive order in 1991 and struck down in 1992. It was enacted again by constitutional amendment in 2019 and upheld in 2022 by three votes to two.
Look at what that scheme actually is. A threshold copied from another quota with no study behind it. A line so high that the overwhelming majority of eligible households fall below it. No published data on beneficiaries. And an exclusion the dissenting judges called the first time the Court had approved a discriminatory principle — a poverty quota that bars the poorest communities in the country.
This is the model being proposed as the replacement. It is not a hypothetical improvement. It exists, and it is worse.
8. Merit is substantially purchased
An entrance score measures performance on a test after a quantity of preparation, and preparation is sold at scale by an enormous industry at prices beyond most Indian households.
Two candidates of identical ability and different family resources do not score identically. That is arithmetic. A system that allocates by score therefore allocates partly by inherited advantage — which means "pure merit" is not an alternative to inherited privilege but a mechanism for laundering it.
9. Those who want it abolished for others want it for themselves
Marathas, Jats, Patidars, Kapus and Gujjars have spent a decade blockading roads and railways to be officially classified as backward. Maharashtra passed an Act in 2024 taking its total to about 62% to accommodate one of them.
Nobody agitates for years to obtain a worthless benefit, and nobody demands a label they believe is stigmatising. The revealed preference of some of India's most powerful communities is the strongest available evidence that reservation delivers something real.
And note where the resistance comes from: existing OBC organisations, defending their share. The conflict is between adjacent groups over a pool that is too small — which is an argument for enlarging the pool, not for emptying it.
10. And there is no plan for the day after
Suppose reservation ends tomorrow. The 27% of households still practise untouchability. The correspondence study still returns ten, six and three. Somebody still cleans the sewer, and on every piece of evidence available it is the same community.
What mechanism replaces it? Not a sentiment — a mechanism. No answer has been offered, because the abolition case is a case against something rather than a case for anything.
The instrument is badly built and has been asked to do more than it can. But it demonstrably works, the injury it addresses is measurable today, its own constitutional condition remains unmet, four religions have already failed to remove caste by changing belief, and the proposed replacement has been tried twice and is worse.
The claim is not that the present arrangement is correct. It is that a working instrument should be rebuilt rather than removed, and that nobody proposing removal has explained what happens on the following day.
4Cross-Examination: Case A
Ten questions to the abolition case, with the strongest answer available to each — and where there is no good answer, I say so.
Answer: enforcement of existing law, mechanisation of sanitation work, and rehabilitation funded from general revenue rather than from a quota. These do not require caste categories.
Assessment: partially adequate. It is a real answer and mechanisation is a genuine policy. But rehabilitation programmes have to identify who needs rehabilitating, and the only reason we know who does is that the state counts by caste. The answer relies on data the proposal would abolish.
Answer: those religions changed belief while the state continued to administer caste. Removing state recognition attacks the material incentive rather than the belief, which is a different intervention.
Assessment: this is the best answer available and it is not strong. Caste was administered by no state at all for most of its history and survived every empire that ruled India. The mechanism identified in Part 3 — endogamy, enforced by families — has never required state recognition.
Answer: sub-classification concedes the principle that benefits concentrate, and then proposes to solve it by adding another layer of categories. Each layer will concentrate in turn.
Assessment: a fair point about regress, and unproven. Telangana implemented in 2025 and Karnataka in 2026; there is no evidence yet either way.
Answer: the case as stated here does not claim over-representation. It claims the instrument has failed at senior level.
Assessment: honest, and it costs the movement its most popular argument. Almost all public anti-reservation sentiment rests on a belief that this data contradicts.
Answer: the threshold is indefensible and should be rebuilt with proper measurement. A badly calibrated example of a category does not invalidate the category.
Assessment: logically correct and practically weak. The one time India built a caste-blind quota, it copied a number and published no results. That is the available evidence about how the replacement would be administered.
Answer: through targeted research and audit studies rather than routine administrative collection.
Assessment: insufficient. The correspondence study depended on knowing which surnames signal which community. The manual scavenging figure is an official count. Audit studies can measure discrimination at a moment; they cannot replace a continuous statistical series, and no state funds them at scale.
Answer: extending it enlarges a system we regard as harmful. Removing the religious test by removing the category treats everyone alike.
Assessment: coherent. But note the asymmetry: the exclusion has stood for seventy-six years and became a talking point in July 2026. If the concern is genuine, extending status is available now and abolition is not.
Answer: nobody, in those instances — which is itself an argument that the system functions poorly.
Assessment: the reply is sound and it changes the character of the grievance. A policy that leaves posts unfilled is inefficient. It is not a policy that took anything from anybody.
Answer: the moment created an opening. Movements appear when they become possible, and the underlying grievance predates it by decades.
Assessment: true and incomplete. The exam-leak protest was directed at the state's own examination machinery. Within days the target had become other students, and the leaks disappeared from the conversation. That is a fact about the movement's effect, whatever its intent.
Answer: the evidence in Part 6 is accepted. The disagreement is about the remedy, not the diagnosis.
Assessment: this is the strongest position the abolition case can take, and it is not the one its public advocates take. The movement's most prominent voices have described historical caste atrocities as invented propaganda. A case that concedes the diagnosis is far more formidable than the one being made in public.
5Cross-Examination: Case B
Ten questions to the retention case. Same treatment, same standard.
Answer: proportionality to population, sustained across levels, in a society where inter-caste marriage rates and untouchability self-reports have converged towards zero.
Assessment: this is an answer, and it is far more than the movement usually offers — but it is not the answer anyone has ever officially adopted. No government, court or commission has defined the threshold in seventy-six years, and until one does, the abolition case's charge that the condition cannot be met stands.
Answer: no. The ceiling was breached by the EWS amendment, which the retention side did not seek. The breach was made for the general category and should be reversed.
Assessment: factually correct and politically empty. Nobody on this side opposed EWS on ceiling grounds at the time, because doing so meant defending a limit that also constrains their own claims. The position is available in principle and was abandoned in practice.
Answer: the data has not existed since 1931. Census 2027 will provide it.
Assessment: partially true and mostly evasive. Several states created most-backward sub-categories decades ago without a census. The obstacle at the centre is that dominant communities inside the OBC category benefit from the absence of sub-classification and have the votes to keep it.
Answer: private-sector reservation, which has been proposed for decades and never enacted; or at minimum mandatory diversity reporting.
Assessment: a genuine answer that nobody is fighting for. If the harm is concentrated where the policy does not reach, the campaigning energy should be there. It is not, on either side.
Answer: nothing directly, because the state cannot legislate whom people marry. Reservation creates the economic independence that makes such choices possible.
Assessment: honest and thin. Inter-caste marriage is around 5% after seventy-six years of the policy. If economic independence produced the effect claimed, the number should have moved.
Answer: no. Reservation was never designed to redistribute assets, and land reform, credit access and inheritance policy would be required to address them.
Assessment: the concession is correct and the implication is large. If the deepest inequality is in assets and the policy cannot reach it, then the argument being fought so bitterly is about a secondary front.
Answer: no. The distinction rests on the form of the instrument rather than on any principle about equality.
Assessment: correct, and it is an argument the abolition case can borrow intact. A ceiling that binds states and not Parliament is a test of parliamentary arithmetic, not of constitutional principle.
Answer: sustained proportional representation across all levels of employment, convergence in the untouchability self-report, and asset ownership approaching population share.
Assessment: a real answer, and note how demanding it is. On those criteria no reserved category is close, but nor would the criteria ever have been met by any group in any country. A threshold set that high is functionally a permanent one, which is the abolition case's charge restated.
Answer: senior posts are filled by promotion from cohorts recruited decades earlier, and reservation in promotion was contested until 1995 and litigated thereafter. The stock lags the flow.
Assessment: this is a good answer and it is testable. If correct, Group A composition should be improving steadily. Whether it is — and at what rate — is precisely the series nobody publishes prominently.
Answer: yes. The thresholds are arbitrary, the categories are too crude, elite capture is real, and the ceiling has been breached for the wrong reasons.
Assessment: and this is where the two cases quietly converge. Both agree the machine is badly built. They differ on whether the response is to rebuild it or remove it — which is a much narrower disagreement than either movement admits.
6The Positions Nobody Is Defending
Abolish or keep is a choice between two options. There are at least seven others, and every one of them is more specific than either.
The previous chapters treated this as a binary because that is how it is being fought. It is not a binary. Here are seven positions that exist in Indian policy discourse, each with a real constituency somewhere, and none of which has a movement behind it.
The case: the strongest charge against reservation is that it has no exit condition. That is a solvable problem. Parliament could define the threshold — proportionality sustained across all groups for a stated number of years — and provide that a category graduates out when it is met. Nothing in the Constitution prevents this, and Article 16(4) already implies it.
The hardest problem: the moment a number exists, every group has an incentive to dispute the measurement, and the government of the day owns the political cost of any category graduating.
The case: the Supreme Court permitted this for Scheduled Castes in 2024, and Telangana and Karnataka have implemented it. The same logic applies with far greater force to 3,743 OBC castes sharing one undivided 27%. Elite capture is real and this is the remedy the highest court chose.
The hardest problem: the dominant communities inside each category have the political power to prevent it, which is precisely why it has happened for SCs and not for OBCs.
The case: this attacks elite capture directly and at the household level rather than the community level. The child of a beneficiary competes openly; the benefit circulates to families that have never had it. It is the creamy layer principle taken to its logical conclusion, and it gives the policy a natural churn.
The hardest problem: one generation of access does not undo eighty generations of exclusion, and a family that has produced one clerk is not thereby equal. It also creates an enforcement problem of extraordinary complexity.
The case: this is what the Mandal Commission itself did — eleven weighted indicators, social, educational and economic, out of twenty-two. The framework exists and has been used. A points system that counts caste heavily, but alongside parental education, first-generation status, rural residence, school medium and household assets, targets far more precisely than either pure caste or pure income.
The hardest problem: complexity invites gaming and litigation, and every weight in the index becomes a political fight. It also requires household data India does not currently collect.
The case: this is the only method that has ever been tried and worked. In 2006–08 seats were increased by roughly half to accommodate the OBC quota, and again for EWS in 2019. Both times the conflict was defused, because the arithmetic that produces panic was removed. In April 2026 Parliament took up a proposal to raise the Lok Sabha from 543 to 850 seats, which does the same thing at constitutional scale.
The hardest problem: it costs money, it requires faculty and infrastructure that were not provided last time, and it works only where capacity can actually be expanded. You cannot double the number of District Collectors.
The case: the cleanest measurement of discrimination in India was taken in private-sector hiring, where reservation has never applied and where most Indians work. Mandatory composition reporting — not quotas — would create the first continuous evidence base on the sector that matters most.
The hardest problem: it has been proposed for decades and never enacted, because the constituency opposing it is the one that funds political parties.
The case: more than 13,500 reserved-category students dropped out of central universities, the IITs and the IIMs. Whichever explanation you accept, the remedy is identical — bridge courses, language support, mentoring, hostels. It is cheap, it is uncontroversial, and it is the one intervention in this entire book that requires nobody to lose an argument.
The hardest problem: it is unglamorous, it produces no headline, and there is no constituency that gains politically from delivering it. Which is presumably why nobody has.
Every one of these positions is more specific than "abolish" or "keep". Several are already partly implemented. Two — sub-classification and expansion — have been demonstrated to work.
And not one of them has a movement, a hashtag or five million followers.
The reason is structural rather than sinister. Each of these positions requires naming a number, accepting a trade-off, and taking responsibility for an outcome. Abolition and retention require neither. They are positions you can hold entirely, at no cost, forever.
Which is worth remembering when a movement appears demanding one of the two options that requires nothing of anybody.
7The Scorecard
Every substantial claim from both cases, marked. No verdict — that is Part 14.
| Claim | Status | Where established |
|---|---|---|
| "Adequate representation" has never been defined | FACT | Part 8 |
| The state requires caste at every life threshold | FACT | Parts 5, 8 |
| Elite capture inside categories is real | FACT | 2024 judgment; Part 10 |
| The EWS threshold has no evidential basis | FACT | Part 10 |
| Reserved categories are below quota in Group A | FACT | Part 11 |
| Government employment is a shrinking share of jobs | FACT | Parts 1, 11 |
| Ambedkar's "minority of seats" limit has been crossed | FACT | Part 8 |
| Removing state recognition would weaken caste | CONTESTED | Four religions failed — Part 4 |
| Permanent labelling entrenches caste identity | INTERPRETATION | Plausible; unmeasured |
| A need-based system would work better | CONTESTED | EWS record — Part 10 |
| Claim | Status | Where established |
|---|---|---|
| Reservation causally improves mobility | FACT | Part 11 — 7–8 rank points |
| Untouchability is practised by 27% of households | FACT | Part 6 |
| Identical CVs produce 10 : 6 : 3 callbacks | FACT | Part 6 |
| ~97% of manual scavengers are Scheduled Caste | FACT | Part 6 |
| Caste survived conversion into four religions | FACT | Parts 3, 4 |
| Thousands of reserved posts sit unfilled | FACT | Part 11 |
| Article 16(4)'s condition remains unmet | FACT | Part 11 |
| Entrance scores partly reflect purchased preparation | FACT | Part 11 |
| Deleting the category would end measurement | FACT | Parts 6, 10, 11 |
| Rebuilding will succeed where the present design failed | CONTESTED | Untested |
Seven of Case A's ten claims are established fact. Nine of Case B's are. Neither side is arguing from fantasy, and both are substantially better than their public versions.
But look at where each side's contested claims sit. Case A's unproven claims are all about what would happen if the policy were removed — that caste would weaken, that labelling entrenches, that a need-based system would work. Case B's single unproven claim is about what would happen if it were rebuilt.
Both sides have solid ground about the present and speculation about the future. The whole disagreement lives in the speculation — which is why eleven parts of evidence have narrowed it far less than one might hope.
8What Neither Case Addresses
The list that changed how I see this argument. Every item is established somewhere in Parts 1 to 11, and neither movement has a position on any of them.
I assembled this list expecting four or five entries. It came to eight, and by the end of writing it I was no longer confident that the argument between abolition and retention is the argument that matters.
- Marriage. Endogamy is the mechanism that creates and sustains caste — Ambedkar said so in 1936 and the genetics confirmed it. Around 5% of Indian marriages are inter-caste. Neither side has a policy, a campaign or a sentence about it.
- Land, credit and capital. Scheduled Castes: about 18% of households and 8.5% of land; 2.6% of billionaire wealth. This is where the durable inequality sits, no reservation policy reaches it, and no movement is campaigning on it.
- The private sector. Where most Indians work, where reservation has never applied, and where the cleanest measurement of discrimination was taken. Both movements are arguing about the small public remainder.
- Support after admission. More than 13,500 reserved-category students dropped out of central universities, IITs and IIMs. Both readings of that figure — they were unprepared, or they were unsupported — imply the same remedy: bridge courses, language support, mentoring, hostels. It is cheap and uncontroversial, and nobody is asking for it.
- Dalit Muslims and Christians. Excluded from Scheduled Caste status by a paragraph written in 1950, while Sikhs and Buddhists are included. Millions of people, a commission running since 2022, and no mass campaign from any direction in seventy-six years.
- Denotified tribes. Communities declared criminal by birth under an 1871 law repealed in 1952, still bearing the administrative name, still among the most deprived and least represented people in India. Absent from this argument entirely.
- Expansion. The one demonstrated method of implementing reservation without crisis: in 2006–08 total seats were increased by roughly half so that unreserved places did not fall, and again for EWS in 2019. Both times it defused the conflict. And in April 2026 Parliament took up a proposal to raise the Lok Sabha from 543 to 850 seats — the largest expansion of reserved political representation in Indian history. Neither movement has mentioned it.
- Measurement. Article 335 conditions reservation on "efficiency of administration" and no metric has ever been created. Article 16(4) turns on "adequate representation" and no threshold has ever been set. The two central conditions of the entire policy are undefined, seventy-six years in, and neither side is demanding definitions.
Every item shares one property: none of them can be resolved by winning the argument about admissions.
Notice also who would have to do the work. Land reform, credit access, private-sector regulation, extending the 1950 Order, funding student support, defining measurement standards — these are administrative and legislative projects requiring money, expertise and years. The abolition-versus-retention argument requires none of those. It requires only a position.
That is not a cynical observation about either movement. It is an observation about what kinds of argument are available to people with phones and no institutional power. The argument goes where the argument can be had.
But it does mean something for anyone deciding what to believe. If you were designing a programme to reduce caste inequality in India from scratch, and you had the evidence in this series in front of you, the reservation question would not be the first item on the list. It would not be the second either.
9Nine Questions Nobody Asks
The blind spots in this stage of the argument.
1. Do the two cases actually disagree about much?
Both accept the injury is real. Both accept the categories are badly built. Both accept elite capture. Both accept the ceiling has been breached incoherently. Both accept the top is unrepresented.
The disagreement is confined to one question: rebuild or remove. That is a narrower dispute than either movement's rhetoric suggests, and neither has an interest in saying so.
2. Why does the public version of each case differ so much from its best version?
The strongest abolition argument concedes that discrimination is real and current. The movement's prominent voices have called historical atrocities invented propaganda.
The strongest retention argument concedes that the design is indefensible. Its defenders in public rarely concede anything.
Both movements are fighting with weaker weapons than they possess, because the strong ones require admitting something.
3. What would each side accept as defeat?
Chapter 4, Q10 and Chapter 5, Q8 asked exactly this. Both produced an answer, which is more than most participants can do — and both answers set thresholds that no evidence is likely to reach in either direction.
4. Why is "expand the pie" absent from a debate about scarcity?
It has worked twice, in 2008 and 2019, and it is before Parliament right now at constitutional scale. It is the only method that has ever defused this conflict, and it appears in neither case.
5. If both sides agree the design is broken, why is nobody proposing a redesign?
Because a redesign requires specifying thresholds, and specifying thresholds means someone loses. Abolition and retention are both positions that avoid ever naming a number.
6. Who benefits from the argument staying where it is?
Anyone whose position depends on the private sector remaining unexamined, on land ownership remaining unexamined, and on measurement standards remaining undefined.
That is not a conspiracy. It is simply what a long, loud, unresolvable argument does: it occupies the space where a resolvable one might have been.
7. Why has neither side asked for the data that would settle their claims?
An audit of "none found suitable." A published series of Group A composition over time. Beneficiary data for EWS. A study of institutional efficiency. Each is cheap and each would settle a live dispute.
Neither movement has demanded any of them.
8. What happens to this argument in eight months?
Census 2027 collects caste data in February 2027. Both cases in this part rest partly on an OBC population estimate derived from 1931. Neither side is preparing publicly for the result.
9. Is "abolish or keep" even the right question?
Chapter 8 is my answer, and it is the most useful thing I found while writing this part. The question is available, urgent-feeling and cheap to have opinions about. On the evidence assembled here, it is not the question with the most at stake.
10The Five-Line Nichod
Everything in this book, compressed. If you remember nothing else, remember these.
PART 12 — THE TWO CASES, STEELMANNED
- The strongest abolition case concedes the injury. It does not claim discrimination is invented or that reserved candidates dominate the services — Part 11 killed both. It argues that a remedy with no defined exit, resting on incoherent categories, which keeps caste administratively alive and has failed at senior level for seventy-six years, is the wrong instrument.
- The strongest retention case concedes the design is broken. It accepts arbitrary thresholds, elite capture and the unmet target. It argues that the policy causally works, that the injury is measurable today, that four religions already failed to remove caste by changing belief, and that the proposed replacement has been tried twice and is worse.
- Seven of Case A's ten claims are established fact. Nine of Case B's are. Both sides' unproven claims are about the future — what would happen if the policy were removed, or rebuilt. Eleven parts of evidence narrow the argument far less than one would hope, because the disagreement lives entirely in the speculation.
- Under cross-examination they quietly converge. Both accept the injury is real, the categories are crude, elite capture is genuine and the ceiling was breached incoherently. The whole dispute reduces to one question: rebuild or remove. Neither movement has an interest in admitting how narrow that is.
- And neither case addresses eight things that matter more. Marriage. Land and capital. The private sector. Support after admission. Dalit Muslims and Christians. Denotified tribes. Expansion. Measurement standards. None can be resolved by winning the argument about admissions — which is why both movements are having a different argument instead.
Coming next
Around thirty questions nobody puts, drawn together from the whole series: why removing reservation keeps caste but deletes the record of it; whether the real fight is between two elites using the poor of both sides as ammunition; whether a shrinking government job market makes this a war over a dying prize; intra-Dalit inequality; what caste does under automation; inter-caste marriage as the number that predicts the future; north against south; and the collision between delimitation and the caste census that nobody is discussing.
Then the comparison: the United States and what followed the 2023 ruling; Malaysia's Bumiputera policy, more than fifty years old and still running; South Africa; Brazil; Nepal; Northern Ireland. Including the places that ended preferences, and what happened next.
A closing thought: I set out to write the two best cases and expected to find them far apart. Under cross-examination they converge on almost everything except the final step — and both fall silent in front of the same list of eight things. If you finish this part less certain than you began, that is the correct outcome and I am not going to apologise for it. — L.S.
Sources & further reading — Part 12
Part 12 contains no new research. Every factual claim is carried forward from Parts 1 to 11.
Where each claim was established
The 27% untouchability self-report, 5% inter-caste marriage, the 10 : 6 : 3 correspondence study result, and the ~97% manual-scavenging composition — Part 6, with full sources including the India Human Development Survey 2011–12, Thorat, Attewell and Rizvi (EPW, 2007), and government figures on sanitation work.
Group A composition (SC 14.2%, ST 6.54%, OBC 19.14%), reserved faculty vacancies, the 4,297 "none found suitable" positions, the 7–8 rank point mobility finding, and the land, enterprise and wealth figures — Part 11, sourced to the Department of Personnel and Training Annual Report 2024–25, Education Ministry replies to Parliament, Asher, Novosad and Rafkin (AEJ: Applied, 2024), the All-India Debt and Investment Survey and the National Family Health Survey.
Article 16(4)'s undefined condition, Article 335's efficiency clause, Ambedkar's statement of 30 November 1948, and the amendment and case history — Part 8.
The EWS threshold and its derivation, the 2022 judgment and its dissents, the Maratha, Jat, Patidar, Kapu and Gujjar demands, the 2024 sub-classification judgment, the Telangana and Karnataka implementations, Census 2027 and the delimitation proposal — Part 10.
The 1991 economic quota struck down in 1992, the Kalelkar covering letter, the derivation of 27%, and the 2006–08 seat expansion — Part 9.
Caste inside Islam, Christianity and Sikhism, and the Constitution (Scheduled Castes) Order 1950 — Part 4. Endogamy as the mechanism — Part 3. The Criminal Tribes Act and denotified communities — Part 5.
A note on method — and on corrections
Chapters 2 and 3 are written as advocacy and should be read as such. They represent the strongest version of each position that the evidence in this series will support, not my own view. Where an argument commonly made by either side has been shown false in earlier parts, it has been excluded rather than included and refuted — which means both cases here are stronger than the ones being made in public.
Chapters 4 and 5 assess each answer. Those assessments are mine and are marked as such. The scorecard in Chapter 6 classifies claims by evidentiary status only and deliberately reaches no conclusion. My own verdict is Part 14, after the comparative evidence in Part 13.
If you believe I have made either case weaker than it deserves — particularly the one you hold — I want to know, and I would rather hear it before Part 14 than after.