The Part That Should Have Been First
This part does two jobs, and by the end of writing it I thought both should have come earlier.
The first three chapters are thirty questions that almost nobody in this argument asks. They have been accumulating across twelve parts. Some are uncomfortable for one side, some for the other, and several are uncomfortable for everybody.
The rest is the comparison. Six countries have run some version of this policy, and one of them stopped very recently.
That last point is why Chapter 5 matters more than anything else in this part. The abolition case rests on a prediction — that removing preferences would not much change outcomes because the barriers are elsewhere. In June 2023 the United States Supreme Court removed race-conscious admissions. We now have three admissions cycles of data.
It is the closest thing to a controlled experiment on removal that exists anywhere, it is recent, and neither side in India has mentioned it once.
Two warnings about comparison, because it is easily abused.
First, no other country has India's problem. Nowhere else is the disadvantaged group defined by a two-thousand-year-old endogamy system with no equivalent anywhere. Comparisons illuminate; they do not transfer.
Second, both sides cherry-pick. The abolition side cites Malaysia, which has run preferences for fifty-five years without resolution. The retention side cites Brazil, which produced fast measurable gains. Both examples are real. This part gives you all six, including the ones that damage the position I am closer to.
One last thing. Question 30 asks whether there is any version of this that ends well for everyone. I answer it honestly, and the answer is not the one I wanted.
— Lovepreet Singh
July 2026
1Ten Questions About The Argument Itself
Not about reservation. About why the argument has the shape it has, and whom that shape suits.
1. Why does removing reservation delete the record of caste but not caste?
The proposal is that the state should stop recording caste. But caste is sustained by marriage, by neighbours, by land and by employers — none of which consults a government form.
What would end is the ability to see it. We know 97% of manual scavengers are Scheduled Caste because somebody counted. Delete the category and the pattern continues, unmeasurable and therefore uncontestable.
2. Is this actually a fight between two elites?
Consider who is arguing. On one side, urban students competing for graduate entry to medical and engineering colleges. On the other, the organised, salaried, educated section of reserved communities — the people reservation has already reached.
Neither group is poor by Indian standards. The landless labourer and the unemployed rural general-category youth appear in this argument mainly as characters in other people's rhetoric.
3. Why is ninety per cent of the job market outside this argument?
Reservation applies to government employment. Most Indians work in the private or informal economy, where it has never applied — and where the cleanest measurement of discrimination was taken.
Both movements are fighting over the public remainder.
4. Government jobs are shrinking. Is this a war over a dying prize?
Public employment has been a falling share of Indian jobs for three decades. Mandal was implemented in 1990 into an economy about to be liberalised.
Both sides are contesting a share of something that has been leaving since before most of the current participants were born.
5. Why is the argument always about admissions and never about marriage?
Ambedkar said in 1936 that intermarriage was the only solvent of caste. The genetics in Part 2 confirmed that endogamy is the mechanism. Around 5% of Indian marriages are inter-caste.
Because admissions can be legislated and weddings cannot. The argument goes where policy can reach, not where the cause is.
6. Who benefits from this argument being unresolvable?
Anyone whose interests are served by private-sector hiring remaining unexamined, land ownership remaining undisturbed, and the two central constitutional conditions remaining undefined.
Not a conspiracy. Simply what a loud permanent argument does — it occupies the space a resolvable one might have used.
7. Why does neither side demand the data that would settle its own claims?
An audit of "none found suitable." A published time series of Group A composition. Beneficiary data for EWS. A metric for administrative efficiency. Each is cheap; each would settle a live dispute.
Neither movement has asked for any of them, which suggests neither is confident of the answer.
8. Why does the strongest version of each case never appear in public?
Part 12 built both, and both are more formidable than what is said aloud — because each requires conceding something. The abolition case is strongest when it accepts that discrimination is real. The retention case is strongest when it accepts that the design is indefensible.
Movements are built on certainty, and certainty is what makes both arguments weaker.
9. If both sides agree the design is broken, why is nobody proposing a redesign?
Because a redesign requires naming a number, and naming a number means someone loses. Abolition and retention are the only two positions in this field that never require anyone to specify anything.
10. Why is "expand the pie" absent from an argument about scarcity?
It worked in 2006–08 and again in 2019, and a proposal to raise the Lok Sabha from 543 to 850 seats is before Parliament now.
It is the only method that has ever defused this conflict, and it appears in neither movement's demands — because expansion costs money and neither movement is asking anyone to spend any.
2Ten Questions About Who Is Missing
Every group in this chapter has a stake in the outcome and no voice in the argument.
11. Which Scheduled Caste sub-groups actually received the benefit?
The Supreme Court permitted sub-classification in 2024 precisely because the answer was "not all of them." Telangana implemented in 2025; Karnataka approved a formula in 2026.
Everyone arguing about the 15% is arguing as though it reaches the category evenly. On the highest court's own finding, it does not.
12. Dalit Muslims and Dalit Christians
Excluded from Scheduled Caste status by a paragraph written in 1950, while Sikhs were added in 1956 and Buddhists in 1990. Millions of people. A commission examining it since October 2022, extended repeatedly.
No mass campaign has ever formed around this, from any direction, in seventy-six years.
13. Denotified tribes
Communities declared criminal by birth under an 1871 law, repealed in 1952, still administratively named after the repeal. Roughly 150 communities, tens of millions of people, among the most deprived in India and frequently classified into whichever category has the most competition.
They are absent from this argument entirely.
14. Women, in every group
The mobility study in Part 11 found daughters' upward mobility lower than sons' across every social group, with less variation between groups than among men.
Which means gender operates as a constraint that cuts across the whole system — and reservation, as designed, does almost nothing about it. Women's reservation in legislatures was enacted in 2023 and is contingent on a census and delimitation that will not conclude before 2029.
15. The 3,743 castes sharing one quota
If the case for splitting the Scheduled Caste category was strong enough for a seven-judge bench, the case for examining a category of nearly four thousand communities is stronger.
It has not happened because the dominant communities inside the OBC category have the votes to prevent it.
16. Adivasis are not a variant of Dalits
Scheduled Tribes have a distinct history — communities largely outside the caste system rather than at the bottom of it, displaced by mining, dams and forest law rather than by ritual exclusion.
They are routinely folded into an argument built around a different problem, and their representation figure — 6.54% in Group A against a 7.5% quota — is the worst shortfall of the three.
17. The poorest of the general category
They exist. Parliament acknowledged it in 2019. They are also used as a rhetorical device by a movement whose actual constituency is urban and educated, and the EWS scheme built for them has a threshold so high it captures almost everyone eligible.
Being real and being used as a prop are not mutually exclusive.
18. The sanitation workers on the government's own payroll
Over 66% of central government safai karamcharis come from Scheduled Caste, Scheduled Tribe and Other Backward Class communities. Around 97% of identified manual scavengers nationally are Scheduled Caste. People die in sewers every few days, and the state and the activists cannot agree on the count.
No hashtag.
19. Migrants, whose caste does not travel with the paperwork
A Scheduled Caste certificate is issued by a state. Move to another state for work and the entitlement frequently does not follow, because the lists differ.
In a country with enormous internal migration, this affects millions, and neither side has a position on it.
20. Who speaks for any of the above?
Every group in this chapter has a stronger claim on attention than exam cut-offs, and none has a movement.
The pattern is not random. Attention flows to groups that are urban, literate, online and organised — which is a description of who is arguing, not of who is affected.
3Ten Questions About The Future
Including the two scenarios both sides refuse to write down.
21. What does caste do under automation?
The prize being fought over is clerical and administrative public employment — precisely the category most exposed to automation over the next two decades.
If those posts contract sharply, a quota over them delivers progressively less, and the argument will have been about a category that shrank underneath it. Neither side has a position on what replaces the ladder.
22. Inter-caste marriage predicts the future better than any quota
Around 5%. If that number began rising steeply, caste would dissolve within a few generations regardless of any policy. If it stays flat, no policy in this book will end caste.
It is the single most predictive statistic in the argument and neither movement tracks it.
23. North and south are running different experiments
Tamil Nadu has operated 69% reservation for three decades with strong social indicators. Several northern states are at or below 50% with weaker ones. The southern political settlement was made in the 1920s; the northern one in 1990.
Two large natural experiments inside one country, and almost nobody treats the comparison as data.
24. The collision nobody is discussing
Three things land at once. Census 2027 collects caste data in February 2027. Delimitation proposes raising the Lok Sabha from 543 to 850. Women's reservation is contingent on both.
Together they redistribute political power by caste, by region and by gender simultaneously — and delimitation by population shifts weight from states that controlled population growth to those that did not. That is a north–south confrontation waiting to happen, and it is barely in the public argument.
25. What if the caste census surprises everyone?
Both sides' positions rest on an OBC estimate derived from 1931. If the real figure is well below 52%, the case for 27% weakens sharply. If it is well above, the pressure to breach the ceiling becomes irresistible.
Nobody is preparing publicly for either outcome.
26. If reservation ended tomorrow, what happens?
Years 1–3. Composition of new recruitment at senior entry levels shifts within one or two cycles. On the American evidence in Chapter 5, the change is fast and visible. Existing employees are unaffected; the stock changes only through retirement.
Years 3–10. Group A composition, already below quota, falls further. The pipeline into senior administration narrows. Reserved-category presence in central institutions declines from a base that was not yet at target.
What does not change. The 27% of households practising untouchability. The 10 : 6 : 3 callback gap in private hiring. Who cleans the sewer. Land at 8.5%. None of these has any connection to the quota.
And politically. A policy with a constituency of several hundred million would have been removed. Part 9 explained why no government has attempted it: after 1990 the beneficiaries became an electorate. Any government that tried would face the reverse of 1990, from a far larger group.
27. If reservation never ends, what happens?
Years 1–10. The pattern of the last decade continues. More categories are added rather than removed — Marathas, and whoever follows. Total reservation rises past 60% in more states. Sub-classification spreads, producing quotas within quotas.
Years 10–25. The government-employment base continues shrinking. The quota covers a larger share of a smaller pool. Group A representation improves slowly through promotion. Land and capital ownership remain roughly where they are, because nothing in the policy touches them.
The risk. Legitimacy. A measure that never ends, whose conditions are undefined, in a shrinking sector, generates exactly the resentment that produced July 2026 — and the next version of that movement will be larger.
And the deeper problem. If the policy runs indefinitely without ever meeting a stated target, its defenders lose the ability to describe it as a remedy at all. It becomes a permanent settlement, which is a different thing and much harder to justify.
28. Does reservation reduce prejudice or manufacture it?
Genuinely open. Contact theory suggests that working alongside people erodes prejudice; the resentment literature suggests that visible preference creates it.
The Indian evidence points weakly against the resentment story: the sharpest measured discrimination is in the private sector, where reservation has never applied. But nobody has run the study properly, in either direction.
29. What would Ambedkar say in 2026?
Both sides claim him and neither can. He designed the provisions and set the ceiling that has been crossed. He said intermarriage was the only solvent and nobody acts on it. He warned in 1949 of a life of contradictions and it has lasted seventy-six years.
The honest answer is that he would be dissatisfied with everyone, which is what he was in life.
30. Is there any version of this that ends well for everyone?
I have thought about this for the length of the series, and the honest answer is no — not on the terms currently being argued.
Every outcome available from the abolish-or-keep binary leaves a large group with a legitimate grievance. Abolition removes a working instrument from people the evidence shows are still being discriminated against. Indefinite continuation asks a shrinking sector to carry a permanent redistribution with no stated end.
The only versions that end tolerably require doing several things that neither side is asking for at once — expanding the base, defining a target, splitting the categories, extending measurement to the private sector, and funding what happens after admission. That is Part 14, and it is not a compromise between the two positions. It is a different question.
4Why Compare At All
What international evidence can tell India, what it cannot, and the three ways this comparison is usually abused.
India is not the first country to reserve positions for a disadvantaged group, and will not be the last to argue about ending it. Six countries have run substantial programmes long enough to generate evidence.
Comparison cannot tell India what to do. It can answer four narrower questions that Indian data cannot: what happens when preferences are removed, whether benefits concentrate over time, whether such programmes ever end, and whether alternatives to quotas exist that work. Those are what the next five chapters address, in that order.
| Country | Group | Running since | Status |
|---|---|---|---|
| United States | Race, in university admissions | 1960s | Ended, 2023 |
| Malaysia | Bumiputera — the majority | 1971 | Running |
| Brazil | Race and public schooling | 2012 | Running, revised 2023 |
| South Africa | Black South Africans — the majority | 1998 | Running |
| Nepal | Caste, ethnicity, region, gender | 2007 | Running |
| Northern Ireland | Religious community | 1976 / 1989 | Largely wound down |
Three ways this comparison gets abused
- Selecting the convenient case. Malaysia is cited by opponents because fifty-five years have not resolved it; Brazil by supporters because a decade produced measurable gains. Both are real; neither alone is evidence.
- Ignoring what the group is. Malaysia and South Africa give preferences to the majority population — a fundamentally different political economy from a minority preference. Conclusions do not carry across.
- Ignoring the mechanism. American race discrimination has no equivalent of endogamy. Nowhere else is the disadvantaged group defined by a marriage system two thousand years old. That is the limit on all of it, and Chapter 11 is about it.
- Ignoring the arithmetic. In the United States the preferred group is roughly a seventh of the population. In India the reserved categories are a substantial majority. A policy that moves a small minority into elite institutions and a policy that reallocates most of a country's public employment are not the same kind of object, whatever they are called.
Before looking abroad, it is worth noticing that India contains its own comparison and nobody treats it as evidence.
Tamil Nadu has operated 69% reservation for three decades under Ninth Schedule protection. Several large northern states have operated at or below 50%, and implemented OBC reservation only from 1990. Two settlements, two starting points, two long runs, in one country, under one constitution, with the same data collection.
Any serious argument about what heavy reservation does to institutions, growth and social indicators has that comparison available and does not use it — because the answer is contested enough to be inconvenient for both sides. Part 1 set out both readings.
5The United States
The only country to have actually removed preferences recently. Three admissions cycles of data — and neither Indian side has mentioned it.
In June 2023 the United States Supreme Court decided the Students for Fair Admissions cases against Harvard and the University of North Carolina, holding that race-conscious admissions programmes — in place for around sixty years — were unconstitutional.
This is the experiment the Indian abolition case implicitly predicts the outcome of. Here is what happened.
Harvard: Black students were 18% of the entering class in 2023, 14% in 2024, and 11.5% in 2025.
Princeton: from 9% in 2023 to 5% in 2025.
Amherst: from 11% in 2023 to 6% in 2025.
Of 29 elite institutions reporting autumn 2025 enrolment, only two maintained Black enrolment of at least 10%, compared with nine of the same institutions in 2022–23.
High-achieving students who would previously have been admitted to the most selective institutions enrolled at less selective ones instead. In autumn 2024, the large majority of public state flagship universities — around 83% — reported gains in under-represented minority enrolment.
Students did not vanish from higher education. They moved down the prestige ladder.
Against the abolition case: the prediction that removing preferences would not much change composition is falsified. It changed composition immediately, substantially, and at exactly the institutions that produce a country's elite. Whatever "merit-based admissions" produced in the United States, it was not the previous distribution.
Against the retention case, and this is real: the students did not fall out of the system. Most went to good universities slightly further down. That is a serious cost — access to elite networks matters enormously — but it is not the catastrophe sometimes predicted, and honesty requires saying so.
And the observation that should trouble both: the effect appeared within one admissions cycle. Whatever is happening at the point of selection responds to policy immediately. That is evidence that these programmes are doing substantial work at the margin — and equally, that removing them does not require decades to show.
American race-conscious admissions were a discretionary consideration in a holistic file review, not a fixed quota. Indian reservation is a numerical set-aside. Removing a soft preference and removing a hard quota are different operations, and the American effect might understate or overstate what would happen here.
What does transfer is the direction and the speed.
6Malaysia
Fifty-five years of preferences for the majority population. The case the abolition side cites, and it is worth citing.
Malaysia's New Economic Policy, launched in 1971 after ethnic violence, had two goals: reduce poverty, and restructure society so that the Bumiputera — the Malay majority and indigenous peoples — participated proportionately in education, professional occupations, business management and wealth ownership.
It is the longest-running comprehensive affirmative action programme in the world, and it is still going.
Poverty. The target was a fall from 49% in Peninsular Malaysia in 1970 to 16% by 1990. The actual figure in 1990 was around 17%. On its own poverty target, the policy essentially delivered.
Corporate ownership. By 1990 foreign ownership of corporate equity had fallen from 69.3% in 1970 to 25.4%. Bumiputera ownership had risen to just over 19% — while the Chinese share had almost doubled, to 45.5%.
The wealth transferred out of foreign hands. Rather less of it reached the intended group than the design assumed.
Scholarly assessments identify a consistent set of outcomes: polarisation and stalemate after five decades; intra-ethnic inequality, with preferences for Bumiputera enterprise disproportionately used by those in urban and prosperous rural areas while rural poverty persisted; and business-sector outcomes described as poor, including the bypassing of competitive non-Bumiputera firms.
The programme has not been ended, and no mechanism for ending it has been established.
The pattern predicted by critics of preference programmes is visible here in full. Benefits concentrated among the already-advantaged inside the preferred group. The programme created a constituency that made ending it politically impossible. Fifty-five years produced neither resolution nor an exit.
Every one of those is a charge made against Indian reservation, and Malaysia is the case where it can be observed over the longest period.
The two qualifications that matter. First, the preferred group is the majority — around two-thirds of the population — which changes the political economy completely; a majority preference has no natural stopping point because the majority controls the legislature. Second, the poverty target was largely met, which is not nothing.
What India should take from it: not that reservation fails, but that a programme without a defined exit condition and without internal sub-classification will run indefinitely and concentrate at the top. India has now begun sub-classifying. It has still not defined an exit.
7Brazil
The newest programme, the fastest results, and the design feature India has never tried.
In 2012 Brazil enacted the Lei de Cotas — the Quota Law — reserving 50% of places in federal universities for students from public schools, with sub-quotas within that for Afro-Brazilian, mixed-race and Indigenous students and for low-income households.
Note the design, because it is the thing India has never done.
The primary filter is schooling — did you attend a public school. Within that, sub-quotas apply by race and by household income.
This is close to Position 4 in Part 12: not caste or income, but a composite. And unlike the Indian debate, it was not framed as a choice between the two.
Afro-Brazilian graduation rates from federal universities rose from around 5% before the law to about 20% afterwards. The student profile of federal universities changed substantially, particularly among entrants from public schools and self-declared Black, Brown and Indigenous students.
On employment outcomes for former beneficiaries, over 75% were in work, with more than half earning above R$5,000 a month.
The law was revised in 2023, correcting several implementation problems rather than reversing the policy.
1. Speed. Measurable change within a decade. Indian reservation in education has run for seventy-six years with a Group A representation figure still below quota. The difference is partly that Brazil's is a much larger quota applied at a single point of entry, and partly that Brazil's programme was designed with the benefit of others' mistakes.
2. The composite design works. Brazil did not have to choose between class and race, and its programme does not carry the Indian problem of a category that is either caste-only or income-only. The Indian argument treats these as alternatives. Brazil treated them as layers.
3. And revision rather than abolition. When problems appeared — including fraudulent self-declaration of race — Brazil amended the law in 2023. That is the third option Part 12 identified and neither Indian movement is proposing.
The honest caveat: a decade is short. The concentration effects visible in Malaysia after fifty years have not had time to appear in Brazil. Anyone citing Brazil as proof should note that the Malaysian problems took thirty years to become undeniable.
8South Africa
What happens when your side wins completely — and the warning it contains for everybody.
South Africa is the case where the historically excluded group is the overwhelming majority, took political power completely, and legislated redress with no meaningful opposition. If preference programmes work, this is where they should work best.
The instruments are the Employment Equity Act of 1998 and Black Economic Empowerment, later Broad-Based Black Economic Empowerment — covering employment, ownership, management control, skills development and procurement.
The recurring scholarly finding is that the programme has produced a Black middle class and a small Black corporate elite, while failing to generate broad-based economic participation or substantially reduce structural inequality.
Black unemployment rose from around 28.1% in the third quarter of 2013 to around 36% in more recent figures.
Analyses point to benefits concentrating among a small cohort — including through corruption and nepotism in the allocation of empowerment deals — and note that measures helped individuals already positioned higher on the skills ladder, with effects too small and too concentrated to move the average.
South Africa had every advantage India lacks. An overwhelming majority. Total political control. Global legitimacy. A comprehensive legislative programme covering not just jobs but ownership and procurement — precisely the asset dimensions Part 11 showed Indian reservation never reaches.
And the benefit concentrated at the top of the beneficiary group anyway, while the position of the average person did not improve.
What this suggests: elite capture is not a flaw in the Indian design. It appears to be a property of preference programmes generally, in every country that has run one long enough to measure. Sub-classification is therefore not an Indian repair to an Indian problem — it is the necessary architecture for any such policy anywhere.
And the sharpest version of the warning: if the concern is the material condition of the poorest, redistributing positions to a group has repeatedly produced a new elite within that group rather than a general improvement. That finding is now visible in Malaysia and South Africa on two continents, over decades, and it deserves more weight in the Indian argument than it gets from either side.
9Nepal and Northern Ireland
The two cases nobody cites — one because it is too close to home, and one because it worked.
Nepal — the closest comparison in the world
Nepal is the only other country with a caste system of the Indian type, and it built an inclusion regime very recently — after the movement of 2006 and the Interim Constitution of 2007, with reservation in the civil service divided among women, Adivasi Janajati, Madhesi, Dalit, persons with disabilities and backward regions, and inclusion principles entrenched in the 2015 Constitution.
It is the one case where the underlying social structure is genuinely comparable — caste with endogamy, untouchability, and a hierarchy of the same lineage.
The recurring criticism is also the familiar one: within each reserved category, the relatively advantaged sub-groups have taken a disproportionate share, and Dalits in particular — the group with the least existing education — have benefited least.
Which is the same finding as Malaysia and South Africa, in a caste society, arrived at within fifteen years. Three countries, three different social structures, one pattern.
Nobody in the Indian argument cites Nepal, which is strange given that it is the only genuinely comparable case on earth.
Northern Ireland — the one that worked, and stopped
From the 1970s, Northern Ireland faced a persistent employment gap between the Catholic and Protestant communities, with Catholic unemployment running at a substantial multiple of Protestant unemployment.
The strengthened regime required employers to register, to monitor and report the community composition of their workforce, to review recruitment and promotion practices periodically, and to take affirmative action where under-representation was found — with enforcement machinery behind it.
It did not set numerical quotas for hiring. It made composition visible, mandatory to report, and subject to review.
Over the following two decades the employment differential between the two communities narrowed very substantially, and the regime has since been progressively wound down.
It is the only example in this chapter of a programme that appears to have worked, achieved something close to its objective, and been reduced as the problem receded — the exit that Malaysia never found and India has never defined.
And the mechanism was not a quota. It was compulsory measurement: every substantial employer required to count, report and review, with consequences for failing to act on what the count showed.
Compare that with India. Part 11 showed the sharpest measured discrimination is in the private sector, where nothing is counted, nothing is reported and nothing is reviewed. India has quotas where it measures, and measures nothing where it has no quotas.
Northern Ireland did the opposite, and it is the case that ended.
Northern Ireland is small, wealthy, and its two communities were not separated by an endogamy system of the Indian kind or by anything resembling untouchability. The gap being closed was one of employment discrimination within a shared labour market, not of two thousand years of exclusion from literacy.
The mechanism may still transfer even though the situation does not. Mandatory composition reporting requires no quota, no certificate and no ceiling — and it is Position 6 from Part 12, which nobody in India is campaigning for.
10Six Countries, Six Lessons
What survives the comparison.
| Case | What it shows | Who it damages |
|---|---|---|
| United States | Removal changes composition immediately and substantially at elite institutions; students cascade downward rather than disappear | Mainly the abolition case |
| Malaysia | Fifty-five years, no exit, benefits concentrated in the already-advantaged, permanent political constituency | Mainly the retention case |
| Brazil | A nested class-plus-race design produced fast, measurable gains; revised rather than abolished when problems appeared | Mainly the abolition case |
| South Africa | Total political control plus asset-level redress still produced elite capture without moving the average | Mainly the retention case |
| Nepal | The same concentration pattern, in a caste society, within fifteen years | Mainly the retention case |
| Northern Ireland | Mandatory measurement without quotas narrowed the gap and was wound down | Both — neither is proposing it |
The six lessons
- 1. Removal has fast, visible effects. The American data settles the claim that preferences are doing little work. They are doing substantial work at the margin, and taking them away shows within one cycle.
- 2. Elite capture is universal, not Indian. Malaysia, South Africa and Nepal — three different social structures, three continents, same finding. Sub-classification is not a patch on a flawed Indian design; it is the necessary architecture everywhere.
- 3. Programmes without a defined exit do not end. Malaysia has run fifty-five years. The only case that wound down is the only case that set out to measure a gap and could see it closing.
- 4. Composite designs outperform either-or designs. Brazil layered schooling, race and income. The Indian argument insists on choosing between caste and income, and the one time India tried income alone it copied a threshold from another scheme.
- 5. Redistribution of positions does not redistribute assets. South Africa legislated for ownership, management control and procurement — and still produced a small elite rather than a general improvement. India, which never legislated for assets at all, should not expect what South Africa failed to achieve.
- 6. Measurement may matter more than quotas. The single case that closed its gap and stopped used mandatory monitoring rather than set-asides. India has quotas where it counts and counts nothing where it has no quotas — which is precisely backwards from the one working example.
11What India Is Uniquely
Where the comparisons stop working — and why that matters for what comes next.
Every comparison in this part has a limit, and it is the same limit each time.
1. The mechanism has no equivalent. American racial segregation was enforced by law and by violence, and both were removable. Caste is enforced by marriage, by millions of families, requiring no state at all. Part 3 established this and Part 4 demonstrated it: caste survived conversion into four religions that reject it. Nothing in the other five cases works this way.
2. The duration is different by an order of magnitude. The endogamy lock closed roughly two thousand years ago. American slavery lasted about two and a half centuries; apartheid about four decades. The others are shorter still.
3. The structure is not binary. Every other case is broadly two-sided — one advantaged group, one disadvantaged. India has thousands of ranked communities in which the same group is oppressed from above and oppressive to those below. Part 6 found that 15% of Scheduled Caste households and 22% of Scheduled Tribe households also reported practising untouchability.
4. And the numbers are of a different scale. The reserved categories in India comprise a substantial majority of the population. This is not a minority preference. It is closer to Malaysia and South Africa in scale — and both of those are the cautionary cases.
For the abolition case: the international evidence that preferences are removable without much consequence does not exist. The one country that removed them saw immediate substantial change, and it had a far shallower problem to begin with.
For the retention case: the international evidence that preferences resolve the underlying inequality also does not exist. Fifty-five years in Malaysia, twenty-eight in South Africa, and in both the benefits pooled at the top of the preferred group.
And for anyone who wants a design rather than a slogan: the three features associated with the better outcomes are visible across the cases. A composite rather than a single criterion, as in Brazil. Internal sub-classification, which every long-running case shows is necessary. And mandatory measurement, which is the only mechanism in this part that closed a gap and then stopped.
India has begun the second. It has never attempted the first or the third.
12What The Evidence Can And Cannot Settle
The honest scorecard for Part 13.
| Question | Status | Best current answer |
|---|---|---|
| Does removing preferences change composition? | Yes | Harvard 18% → 11.5% in two years |
| Do removed students leave education? | No | They cascade to less selective institutions |
| Is elite capture specific to India? | No | Malaysia, South Africa, Nepal — same pattern |
| Do such programmes end on their own? | Rarely | One case in six wound down |
| Does asset-level redress work better? | Not shown | South Africa legislated it and still concentrated |
| Do composite designs work? | Promising | Brazil — but only a decade of data |
| Is there an alternative to quotas? | Yes, one | Mandatory monitoring — Northern Ireland |
| Is India's case comparable? | Partly | Nepal only; the mechanism is unique |
| Does reservation reduce or create prejudice? | Unknown | No adequate study in either direction |
| Is there an outcome good for everyone? | Not on these terms | See Question 30 — and Part 14 |
The three things nobody can tell you
- Whether the American cascade is benign. Three cycles is short. Whether attending a less selective institution changes life outcomes materially will take a decade to establish.
- Whether Brazil will follow Malaysia. The concentration problems in Malaysia took thirty years to become undeniable. Brazil is twelve years old.
- Whether Northern Ireland's mechanism transfers. A small wealthy society with no endogamy system is a long way from India, however elegant the instrument.
Six countries tried this; the one that removed preferences saw composition change within a year, the ones that kept them saw benefits pool at the top, and the only one that finished the job used compulsory measurement rather than quotas — which is the single thing India has never tried and neither side is asking for.
13The Five-Line Nichod
Everything in this book, compressed. If you remember nothing else, remember these.
PART 13 — THE BLIND SPOTS
- One country actually removed preferences, and we can see what happened. After the 2023 US Supreme Court ruling, Black enrolment fell at Harvard from 18% to 11.5%, at Princeton from 9% to 5%, at Amherst from 11% to 6%. Students did not leave education — they cascaded to less selective institutions. Neither Indian side has mentioned it once.
- Elite capture is not an Indian flaw. It is universal. Malaysia after fifty-five years, South Africa after twenty-eight, Nepal within fifteen — three continents, three social structures, one finding: benefits pool among the already-advantaged inside the preferred group. Sub-classification is not a patch; it is the necessary architecture anywhere.
- South Africa is the harshest lesson, and it lands on the retention side. An overwhelming majority, total political power, and legislation covering ownership and procurement — the asset dimensions Indian reservation never reaches. It still produced a small elite while Black unemployment rose from 28% to 36%.
- The only programme that finished used measurement, not quotas. Northern Ireland required every substantial employer to count, report and review workforce composition, with enforcement. The gap narrowed and the regime wound down. India has quotas where it counts, and counts nothing in the private sector where the measured discrimination is worst — exactly backwards.
- And the answer to Question 30 is no. There is no outcome from the abolish-or-keep binary that ends well for everyone. The versions that end tolerably require doing five things at once that neither movement is asking for — which means the question being fought over is not the question that matters.
Coming next
The Reservation Hatao Andolan's demands, taken seriously and assessed one at a time for feasibility. What is genuinely true in the anti-reservation case, and what is genuinely true in the case for keeping it. What each side is lying about, or lying to itself about.
Then my verdict — boxed, labelled as opinion rather than evidence, with the reasoning shown so you can disagree with it precisely rather than in general. What a fair system for 2047 India could actually look like, in concrete detail. The one thing each side would have to give up. And a closing note.
A closing thought: I went looking for international evidence expecting it to favour one side. It does not. It favours a set of design features — composite criteria, internal sub-classification, and compulsory measurement — that cut across the argument entirely. The countries that did better were not the ones that argued harder. They were the ones that measured more carefully. — L.S.
Sources & further reading — Part 13
Every factual claim in Part 13, and where it comes from.
United States
Students for Fair Admissions v. President and Fellows of Harvard College and Students for Fair Admissions v. University of North Carolina, decided June 2023, holding race-conscious admissions programmes unconstitutional after roughly sixty years of operation. · Reported freshman enrolment: Harvard 18% Black in autumn 2023, 14% in autumn 2024 and 11.5% in autumn 2025; Princeton from 9% in 2023 to 5% in 2025; Amherst from 11% in 2023 to 6% in 2025. Of 29 elite institutions reporting autumn 2025 data, two maintained Black enrolment of at least 10%, against nine of the same institutions in 2022–23. · On the cascade effect: high-achieving students of colour enrolling at less selective institutions, with approximately 83% of public state flagship institutions reporting gains in under-represented minority enrolment in autumn 2024.
Malaysia
New Economic Policy, 1971, with its twin objectives of poverty reduction and social restructuring through pro-Bumiputera affirmative action in education, high-level occupations, enterprise management and wealth ownership. · Poverty in Peninsular Malaysia: target of a fall from 49% in 1970 to 16% by 1990, with an actual figure of approximately 17% in 1990. · Corporate equity ownership: foreign ownership falling from 69.3% in 1970 to 25.4% in 1990; Bumiputera ownership rising to just over 19%; the Chinese share rising to 45.5%. · Scholarly assessment of fifty years of the NEP identifying polarisation and stalemate, intra-ethnic inequality with preferences disproportionately used by those in urban and prosperous rural areas, and poor outcomes in the business sector including the bypassing of competitive non-Bumiputera firms.
Brazil
Lei de Cotas, 2012, reserving 50% of federal university admissions for public-school students with sub-quotas by race and household income; revised in 2023. · Afro-Brazilian graduation rates from federal universities rising from approximately 5% before the law to approximately 20% afterwards; substantial change in the student profile of federal universities, particularly among public-school entrants and self-declared Black, Brown and Indigenous students. · Employment outcomes for former beneficiaries: over 75% employed, with more than half earning above R$5,000 monthly.
South Africa
Employment Equity Act, 1998, and Black Economic Empowerment, subsequently Broad-Based Black Economic Empowerment, covering employment, ownership, management control, skills development and procurement. · Assessments finding growth of a Black middle class and corporate elite without broad-based economic participation or substantial reduction in structural inequality; concentration of benefits among a small cohort including through corruption and nepotism; and effects concentrated among individuals already higher on the skills ladder. · Black unemployment rising from approximately 28.1% in the third quarter of 2013 to approximately 36% in more recent figures.
Nepal and Northern Ireland
Nepal: inclusion and reservation provisions following the movement of 2006 and the Interim Constitution of 2007, with civil service reservation divided among women, Adivasi Janajati, Madhesi, Dalit, persons with disabilities and backward regions; inclusion principles entrenched in the 2015 Constitution. Criticism centres on disproportionate capture by relatively advantaged sub-groups within each category, with Dalits benefiting least. · Northern Ireland: Fair Employment (Northern Ireland) Act 1976 and the strengthened Fair Employment Act 1989, requiring registration, monitoring and reporting of workforce composition by community background, periodic review of recruitment and promotion practices, and affirmative action where under-representation was found, with enforcement machinery — but without numerical hiring quotas. The employment differential between the two communities narrowed substantially over the following two decades and the regime has since been progressively wound down.
Carried forward from earlier parts
Indian figures cited throughout this part — the 27% untouchability self-report, the 10 : 6 : 3 correspondence study, the ~97% manual scavenging composition, the Group A representation figures, the land and wealth data, the 2024 sub-classification judgment, Census 2027 and the delimitation proposal — are sourced in full in Parts 6, 10 and 11.
A note on what is missing — and on corrections
International comparisons in this part are drawn from published scholarship and reporting and are given in outline. Enrolment figures for United States institutions are as reported and definitions of racial categories differ between institutions and years, which affects comparability at the margin. Malaysian and South African assessments are summarised from academic literature in which findings are contested in detail though broadly consistent in direction. Northern Ireland's mechanism is described from the structure of the legislation; attributing the narrowing of the gap to it specifically involves a causal claim that the literature supports but does not prove.
Part 13 offers no verdict. That is Part 14. If you find an error, I want to know.