Where We Left Off
Chapter Six compares 1984 with four later episodes under four different governments and finds the pattern is not partisan. That is the most uncomfortable finding in this part and it is not an exoneration of anybody.
Part Ten established what happened and what it produced. This part is about the machinery that has processed it since, and about what that machinery is for.
Four findings carry into this part.
The arithmetic. Part Ten’s Chapter Seven: 2,733 killed in Delhi and about 3,325 nationally, on the state’s own figures. 587 first information reports. About 240 closed by police as untraced. Twenty-five convictions, of which twelve are for murder. Four commissions, nine committees, two investigation teams.
The failure was at the beginning, not the end. Part Ten’s finding was that the courts were handed almost nothing, because a case is built in the first week by the officers who take the complaints, and those were the same force that had stood by. Chapter Three follows that further than Part Ten could.
Organisation is established; central direction is not. Part Ten’s Chapter Six declined to assert that the killing was directed from the top, because forty-one years of inquiry has not established it, and rested instead on the narrower point that the police not acting was not a local decision. That distinction governs Chapter Two.
The word. Part One promised to name the concealing word every time. Part Nine named “excesses” over the sterilisation programme. Part Ten named “riot”. Chapter Six here takes it properly, because it is still the official word.
One note on scope, because this part could easily be a repetition. It does not re-describe the five days; Part Ten did that and this part assumes it. What it examines is the forty-one years since — which is a different subject, is much less written about, and is where the thing that has actually shaped Indian public life happened.
How to read this book
The Six Boxes
Six coloured boxes run through the series, each doing one job. Here they are with examples from the material ahead.
A Word Box explains a hard word the moment it first appears.
Ex gratia: a payment made without admitting any legal liability. From the Latin for “out of grace” — a gift rather than a debt.
The phrase does real work. A state paying compensation is settling an obligation and admitting it had one. A state paying ex gratia is being generous about something it denies owing.
Every rupee paid to a family bereaved in November 1984 has been paid on that basis, in four tranches over thirty years, and the framing has never changed.
Why it matters here: Chapter Five is about what was paid, and the more revealing question is who had to ask.
An In Real Terms box turns a number too big to picture into something with a body.
The first payment made to a family whose father or son was killed in Delhi in November 1984 was twenty thousand rupees.
A person injured received between five hundred and two thousand. A house destroyed completely was worth five thousand.
It took a Delhi High Court judgment in 1996 to raise the figure for a death to three and a half lakh — and most families did not receive the balance until a package was announced in 2006, ten years after the court ordered it and twenty-two years after the killing.
A How We Actually Know This box shows the physical evidence and then says what it cannot prove.
This part rests almost entirely on the Indian state’s own paperwork, and that is deliberate.
Commission and committee reports. Court judgments, including several High Court decisions with unusually direct language about the administration. Government orders on relief, with dates and amounts. Parliamentary answers on disbursement. Conviction records.
Nothing here depends on a survivor’s account, not because those are unreliable but because they are not necessary and because the standard rebuttal to any account of 1984 is that it comes from an aggrieved community. It does not. It comes from the state.
What it cannot show: what was said and not written. The most consequential decisions in this part — that the police would not act, that a committee’s recommendation would not be pursued, that a file would be marked untraced — do not appear as decisions anywhere. They appear as absences.
An Argument box appears where serious people disagree, each side at its strongest.
The framing question, and the defence of the Indian record here is better than it is usually allowed to be.
Twenty-five convictions is not nothing. A former Member of Parliament is serving life. Special investigation teams reopened files that had been closed for decades and secured convictions from them. Compensation rose from twenty thousand rupees to well over eight lakh across four tranches, paid by governments of both major parties. India did not do what many states do with a massacre by its own supporters, which is to legislate an amnesty and move on. The system worked badly and it worked.
Four commissions, nine committees and two investigation teams over four decades is not a system straining towards justice; it is a system that produces reports instead of prosecutions. Every enhancement of compensation followed a court order, which means it was extracted rather than granted. The first senior conviction came thirty-four years afterwards, when most witnesses and several accused were dead. A process that reliably arrives after the participants have died is not slow justice. It is a different thing that resembles it.
What would settle it: the Delhi Police records of case disposal in 1984 and 1985, which have been sought repeatedly.
Why people care so much: because if it eventually worked, the Indian state is slow, and if the process was the outcome, it is something else.
A Hidden Assumption box digs out a belief sitting underneath an argument that both sides accept without discussing.
Everybody who writes about the aftermath of 1984 assumes that what was lacking was political will.
It is the universal explanation. There was no will to prosecute. Governments protected their own. Successive administrations lacked the courage. Critics say it and defenders answer within the same frame, arguing that the will existed and the evidence did not.
“Political will” is not an explanation. It is a name for the absence of one.
It cannot be measured, cannot be falsified, and identifies nothing that anybody could have done differently on a particular day. A country that concludes its institutions failed for want of will has learned nothing it can act on, which is why the phrase is so popular with everybody: it condemns without requiring anything.
Every failure in this part has a procedural name and a location. Who registers a first information report, and where. Which force investigates a crime its own members are accused of. Who may close a case as untraced, and who reviews it. Which court has jurisdiction, and whether a trial can be moved out of the state. Who supervises an investigation — an executive or a judge. Whether a commission’s recommendation binds anybody.
Each of those is a rule. Each can be changed. And the one thing in forty-one years that demonstrably produced results — Supreme Court-monitored investigation teams with the power to reopen closed files — is a procedural device, invented in 2008 for a different massacre under a different party, and applied to 1984 only from 2015.
That is the whole argument of this part. The failure was not a shortage of virtue in Indian politicians, of which there has never been an unusual abundance in any country. It was a set of rules about who investigates whom, and those rules are still mostly what they were in 1984.
And every chapter closes with a Remember This box: the chapter in the plainest words available.
This part is not the five days. It is the forty-one years since, which is less written about and is where the thing that shaped Indian public life happened.
It rests almost entirely on the state’s own paperwork — commissions, judgments, relief orders, parliamentary answers — because the standard rebuttal to any account of 1984 is that it comes from an aggrieved community, and this one does not.
“Political will” is not an explanation. It is a name for the absence of one. Every failure here has a procedural name, and every one of them is a rule that could be changed.
1What Has Actually Been Found
A great many names are repeated in connection with November 1984. Very few of them have been found against by anybody. The gap between those two facts is where this part begins.
1.1 — The distinction
Three different things get run together in every discussion of this subject, and separating them is the only way to say anything useful.
An allegation is a claim made by somebody. It may be true. It establishes nothing.
A commission finding is a conclusion reached by an inquiry with the power to summon and take evidence on oath. It is serious and it is not a verdict. A commission cannot convict, its findings are not binding, and a person named in one has usually had a limited opportunity to answer.
A conviction is a court’s determination after trial, with the accused defended, the evidence tested and an appeal available.
Part Ten’s arithmetic tells you the shape of what follows. Allegations against many. Commission findings against several. Convictions against twenty-five cases and twelve murders.
1.2 — What the commissions found
Across the inquiries listed in Part Ten’s Chapter Seven, a small number of political figures were named repeatedly as having led, directed or incited mobs, and recommendations were made that cases be registered against them.
Recommendations. Not convictions, and in most instances not even prosecutions, because Chapter Three describes what happened to the recommendations.
The Nanavati Commission, the most comprehensive, reported in 2005. It named individuals, and it found a colossal failure in the maintenance of law and order.
Beyond that this part will not go. It will not repeat lists of names attached to allegations that no body has found on, because doing so is exactly the thing that has allowed this subject to be dismissed for forty-one years as the grievance of an aggrieved community.
1.3 — What the courts have found
This is much shorter and much stronger.
In December 2018, the Delhi High Court reversed an acquittal and convicted a former Member of Parliament, sentencing him to life imprisonment for his part in the killing of a family. A second conviction and life sentence followed in 2025 in a separate case.
In November 2018, a Delhi court convicted two men of murders committed in Mahipalpur — the first conviction arising from the cases reopened by a special investigation team constituted in 2015. The case had been registered in 1993 and closed by the police in 1994 for want of evidence.
Across all cases and all years, roughly 440 individuals have been convicted of something, on twenty-five cases, of which twelve are murder cases.
That is the documentary record of who has been found responsible for two thousand seven hundred and thirty-three deaths, in a court, after trial.
Commission of inquiry: a body appointed by a government under the Commissions of Inquiry Act to investigate a matter of public importance and report.
Its powers are real: it can summon witnesses, require documents and take evidence on oath. Its limits are equally real and are the point. It cannot prosecute, its findings bind nobody, and a government is not required to act on them or even to publish them.
Part Two’s chapter on Hyderabad found a report unpublished for sixty-four years. Part Six found one classified for sixty-three. Part Nine found one withdrawn from circulation after the government changed.
Why it matters here: appointing a commission is the most common Indian governmental response to an atrocity, and it is compatible with doing nothing at all.
Before going further it is worth grading those three categories against what this part can actually assert.
The distinction in section 1.1 is not pedantry; it is the difference between what this part can assert and what it cannot.
Solid: the conviction record. Judgments are published, name the accused, set out the findings and record the sentence. Anybody can read them.
Documented: the commission findings. The reports exist, they name individuals, and their recommendations are on the record. What they are not is determinations of guilt, and the people named have generally not had the opportunity to defend themselves that a trial provides.
Alleged: everything else, and there is a great deal of it, repeated for four decades.
What none of it can show: who actually killed most of the 2,733. For the large majority there was never an investigation capable of answering that, which is the subject of the box below.
Set that record against the death toll and the shape of the problem changes.
Twenty-five cases have produced convictions. Two thousand seven hundred and thirty-three people were killed.
Turn that around. For roughly nine in every ten deaths there is not merely no conviction — there is no case. No charge sheet, no accused, no trial that failed. Nothing was ever brought.
A country arguing for forty-one years about whether the guilty escaped justice has been arguing about the wrong tenth. For the other nine, justice was never attempted.
Which is why the argument about names has never gone anywhere.
Everybody arguing about 1984 assumes that the guilty are known and unpunished.
The critical version: everyone knows who did it and they have been protected. The defensive version: names have been circulated for decades without proof, which is precisely why nothing has stuck. Both are arguments about a set of names, and both parties behave as though establishing the list would establish the wrong.
The list is not the wrong. The absence of a list is.
Consider what it would take, in an ordinary murder, to be able to say who was guilty. A complaint recorded within hours naming the people the complainant saw. Statements from neighbours taken while the street still remembered. An investigating officer assigned, a charge sheet filed, a trial. None of that is remarkable; it is the routine processing of a killing in any Indian district.
For 2,733 deaths there are 587 reports and about 240 of those were closed as untraced. Which means that for most of the people killed in Delhi in November 1984, there has never been a document naming anybody — not because the killers were unknown to the neighbourhood, but because nobody wrote the names down.
So the argument about who is guilty is being conducted in a space where guilt was never determinable, and that is not an accident of history. It is the direct product of the decisions Chapter Two describes.
Which is why this part refuses to trade in names. Repeating an allegation is the thing the system’s design makes available instead of proof, and doing it has substituted for forty-one years of asking why proof was never possible.
All of which raises a question about this book rather than about its subject, and it should be answered in the open.
This part does not, and that is a choice which victim organisations would dispute. The dispute deserves stating rather than settling by omission.
Commissions named individuals. Committees recommended cases against them. Those recommendations were blocked. If the only names anybody may repeat are the ones a court has convicted, then a system which prevented cases reaching court has also succeeded in preventing them being spoken of — which is impunity completing itself. Survivors have said the names for forty-one years precisely because nothing else was available to them, and asking them to stop is asking them to accept the outcome the obstruction produced.
An allegation repeated for four decades is not evidence and does not become evidence. It is also the thing that has allowed this entire subject to be characterised as a grievance rather than a finding, and to be answered with the observation that names have circulated without proof. A book that repeats them hands its opponents that answer for free. And there is a specific cost: the person named has no forum in which to be cleared, because there is no case against him to lose.
What would settle it: prosecutions. The argument exists only because they did not happen.
Why people care so much: because for a survivor the name is the whole of what was left, and being asked to withhold it is being asked to give up the last thing.
One consequence for how this part is written. It names no individual against whom no body has found. That restraint costs it the most quotable material available on the subject, and it is the price of the argument in the box above.
Three things get run together: an allegation establishes nothing, a commission finding binds nobody, and a conviction is a court’s determination after trial.
The court record for 2,733 deaths: 25 cases, 12 murder cases, about 440 individuals, with the first senior conviction in December 2018 and a second in 2025.
The list of guilty names is not the wrong. The absence of a list is. For most of those killed there has never been a document naming anybody, because nobody wrote the names down.
2The Investigators
The Delhi Police were found by commission after commission to have stood aside while it happened. They were then given the investigation of it. That is not an oversight and it is not a scandal — it is the ordinary rule.
2.1 — What was found about them
Every substantial inquiry into November 1984 reached findings about the police, and they are consistent across bodies appointed by different governments over twenty years.
That officers did not intervene where they could have. That complaints were refused registration. That in documented instances Sikhs who had armed themselves for defence were disarmed by police while the mobs were not. That the force’s response amounted to a colossal failure in the maintenance of law and order.
That last phrase is the Nanavati Commission’s, in 2005, on the conduct of the Delhi Police.
Committees were specifically appointed to examine police conduct and to recommend action against named officers. Recommendations were made.
2.2 — Who investigated the crime
The Delhi Police.
The first information reports of Chapter One were recorded by that force. The investigations were conducted by it. The decisions to close about two hundred and forty cases as untraced were taken by it. The case files that the special investigation teams reopened from 2015 onwards were files that force had made and closed.
This is not a hidden fact and nobody has ever concealed it. It is the ordinary operation of Indian criminal procedure: crimes are investigated by the police force with territorial jurisdiction, and there is no general rule removing that jurisdiction where the force itself is implicated.
Jurisdiction: the authority of a particular body to deal with a particular matter — which police force investigates, which court tries, which magistrate supervises.
It sounds like an administrative detail and it decides outcomes. Indian criminal procedure allocates investigation by territory: an offence is investigated by the police of the area where it occurred.
There is no general exception for the case where that force is itself implicated. Removing a case from the ordinary jurisdiction requires a specific order from a High Court or the Supreme Court, on an application somebody must make and sustain.
Why it matters here: this is the single rule from which everything in Chapters Two and Three follows, and it has not been changed.
The consequence of that rule is easier to see at the scale of one person than at the scale of a city.
Imagine reporting a burglary to a police station, and being asked to give your statement to the officer you say committed it.
Now add that he decides whether to write your complaint down at all; that if he writes it down he investigates it; that he decides what evidence to gather; and that he can close the file on the ground that the offender could not be identified.
Nobody would design that. It is not a design. It is the ordinary territorial rule of criminal procedure meeting a situation it was not written for — which is why it produces the same result every time it meets one.
2.3 — Was it addressed?
Partly, late, and by a court rather than by a government.
Committees recommended action against officers. In some cases departmental proceedings followed. Some officers were named in commission reports and continued in service.
Convicting a police officer of an offence committed during communal violence in India is close to unknown. It is difficult to convict a politician and it has been done. The record on officers is thinner still, and the reason is Chapter One’s: the investigation of a police officer’s conduct in a riot is conducted by police officers, using records made by police officers, about a period in which the record-keeping is the thing complained of.
The genuine change came only when investigation was removed from that force. Special investigation teams operating under Supreme Court supervision from 2015 reopened closed files and obtained convictions from them within three years — from cases the same force had marked untraced two decades earlier.
That is the single most useful fact in this part and Chapter Eight is about it.
The findings about police conduct come from commissions appointed by the Government of India — including by governments of the party whose local functionaries the same commissions were naming.
That matters for the standard rebuttal. An account of police failure in November 1984 sourced to survivors can be dismissed as partisan. An account sourced to the Misra Commission, the Kapoor–Mittal Committee and the Nanavati Commission cannot, because those bodies were the state examining itself.
The 2018 Delhi High Court judgment adds the strongest layer: a court, in a judgment, stating that the perpetrators enjoyed political patronage and were assisted by an indifferent law enforcement agency.
What it cannot show: how many officers acted well. Some did, and some saved lives, and the inquiries record instances. There has never been a systematic account of them, which means the honourable conduct of individual policemen in November 1984 is as poorly documented as the dishonourable conduct, and that is its own small injustice.
Which brings up the word every report on this subject uses.
Everybody who examines the police role in November 1984 assumes that the police failed.
The word is in every report. Failure of law and order. Dereliction. Deficiency. The inquiries use it, the courts use it, and the defenders of the force use it too, arguing that a small number of stations were overwhelmed by an unprecedented situation.
A failure is a thing that was attempted and did not work. Ask what was attempted.
A force that does not deploy, then does not register complaints, then investigates the resulting crimes, then closes two-fifths of the cases it opened for want of information it did not gather, has not failed at a task. It has performed a coherent sequence in which each step follows from the last. The non-registration in November makes the untraced closure in 1985 inevitable, and the untraced closure makes the acquittals of the 1990s inevitable, and each stage is individually explicable as ordinary incompetence.
This is why the failure framing is so durable and so useless. It suggests the remedy is better policing — more training, more officers, faster response. None of that touches the actual mechanism, which is that the same institution occupied every position in the sequence: witness, accused, investigator and record-keeper.
And it explains the one thing that worked. The 2015 teams did not police better. They took the files away from the force that made them and put the investigation under a judge. The convictions followed within three years, from cases that had been sitting closed for twenty.
The general rule, which this series has met before in a different setting: an institution asked to investigate itself will produce a record, and the record will be the defence. Part Six found an inquiry forbidden from examining the people who gave the orders. This is the same problem without even the terms of reference.
Which raises a question with a less obvious answer than it first appears to have.
The obvious answer is no, and the objection to the obvious answer is real.
Every police force in the world investigates offences involving its own members, because there is no infinite regress of independent agencies and somebody has to hold the file. India has internal vigilance, magisterial oversight of custodial deaths, and appeal to the courts. A rule that removed investigation from the local force whenever it was accused would hand every accused person a method of derailing an inquiry: allege police misconduct and the case moves.
The objection is not to ordinary cases involving officers. It is to cases where the failure to act is itself the thing complained of, because then the investigator is being asked to document his own institution’s inaction, using records his colleagues made, about a period in which the record-keeping is the allegation. That is a specific and narrow category, it is identifiable in advance, and Chapter Eight shows what happens when investigation is removed in exactly that category.
What would settle it: comparing outcomes in cases transferred out and cases left in place. Chapter Six’s table is the beginning of that and the pattern is one-directional.
Why people care so much: because police reform in India has been recommended by commission after commission since the 1970s and the recommendations have never been enacted either.
Note what the second side of that argument is asking for. Not a new principle, not police reform in general, and nothing that requires anybody to admit anything about 1984. A rule about transfer, in a narrow and identifiable category, of the kind Chapter Nine sets out.
Commissions appointed by successive governments found that the Delhi Police did not intervene, refused to register complaints, and in documented instances disarmed Sikhs who were defending themselves.
That force then investigated the crime. It recorded the 587 reports, conducted the investigations and closed about 240 cases as untraced. This is not a scandal; it is the ordinary rule of Indian criminal procedure.
It is not a failure. It is a coherent sequence in which non-registration makes untraced closure inevitable and untraced closure makes acquittal inevitable — and the one thing that worked was taking the files away from that force.
3The Machinery of Delay
Fourteen bodies over thirty-four years. This chapter names the specific devices by which that produced twelve murder convictions, because “delay” is not a mechanism and each of these is.
3.1 — The devices
Delay is a description of an outcome. What follows are the mechanisms that produced it, each of which is a rule that operated as designed.
The commission that reports to the body it examines. A commission of inquiry submits to the government. The government decides whether to publish, whether to accept, and whether to act. None of those is compelled.
The committee that recommends registration. Several of the committees after 1984 had the specific function of recommending that cases be registered against named people. A recommendation to register is not a registration; it goes to the police, who are Chapter Two’s force.
The writ petition. When one such committee recommended registration against named individuals, those individuals went to court challenging the committee’s competence to do so. The proceedings took years. The committee’s work was stalled and it was eventually replaced by another committee, which was replaced by another.
That is worth stopping on. The legal system’s protections for the accused — the right to challenge a body’s jurisdiction — were available to people named in a massacre inquiry, and were not available to the families, who had no case to protect because no case existed.
The untraced closure. Chapter Two. Once a file is closed, reopening it requires somebody with authority to want to.
Jurisdiction. Every case remained in Delhi, tried under Delhi courts, investigated by Delhi Police, with witnesses living in Delhi neighbourhoods where the accused also lived.
Succession. Each new body’s appointment was itself an answer to demands for action, and each one restarted the clock.
Writ petition: an application to a High Court or the Supreme Court asking it to restrain a public body from acting beyond its powers.
It is a protection of the first importance and this series has relied on it repeatedly — it is what struck down the privy purse order in Part Eight and what limited the Emergency amendments in Part Nine.
It is also available to anybody. When a committee recommended that cases be registered against named individuals after 1984, those individuals filed writ petitions challenging the committee’s competence. The proceedings ran for years and the committee’s work stopped.
Why it matters here: the same instrument that protects a citizen from the state protects a powerful citizen from an inquiry, and the families had no equivalent instrument because they had no case to protect.
3.2 — What the succession achieved
Count what fourteen bodies over thirty-four years actually produced.
They established the death toll. That is real and Part Ten uses it throughout.
They established the conduct of the police, the use of voter lists, the distribution of kerosene, and the failure of the administration. That is real and it is the evidentiary basis of everything in Part Ten’s Chapter Six.
They did not produce prosecutions. The prosecutions that exist came from two sources: a small number of cases that were pursued through the ordinary courts by families and lawyers over decades, and the reopening of closed files by investigation teams under Supreme Court supervision from 2015.
So the inquiry machinery worked as an establishing instrument and not as a prosecuting one — which is exactly what it is designed to be, and which is why appointing a commission remains the standard Indian governmental response to an atrocity.
A witness who was thirty in November 1984 was sixty-four when the first senior conviction was handed down in 2018.
A widow who was twenty-five was fifty-nine.
An investigating officer who was forty when he closed a file as untraced was ninety when a team reopened it.
Delay in a criminal process is not a neutral cost borne equally. It transfers the case from the living to the dead, and it does so in one direction only: the accused’s ability to be tried expires with him, and so does the survivor’s ability to give evidence, but the survivor’s need does not.
Which raises the question this chapter exists to put, and it does not have an obvious answer.
This is the central question about the aftermath and it has two serious answers.
Indian courts are enormously overloaded, cases routinely take decades, and conviction rates for serious offences are low across the board. Nothing that happened after 1984 requires a special explanation: mass-casualty cases with hundreds of accused, thousands of witnesses and no forensic evidence are the hardest thing a criminal system attempts, anywhere. The 1992–93 Bombay cases and the 2002 Gujarat cases also took decades, under different governments of different parties, which strongly suggests a structural cause rather than a conspiracy.
Grant every word of that. It does not explain the specific decisions: that about 240 cases were closed as untraced within months, that recommendations to register cases were successfully blocked by writ petitions, that fourteen bodies were appointed and none had prosecuting power, and that when a team was finally given the files and judicial supervision it obtained convictions within three years from cases marked untraced for twenty. A system that is slow for everybody does not produce convictions in thirty-six months once somebody with authority looks.
What would settle it: the case disposal records showing who authorised each untraced closure and on what material.
Why people care so much: because “systemic” spreads the responsibility across a country and “designed” places it on a party, and both answers are politically useful to somebody.
One clarification. Nothing in this chapter argues that the fourteen bodies were appointed in bad faith. Several were appointed in response to genuine public pressure, several did serious work, and the Nanavati Commission’s findings are the evidentiary foundation of Part Ten’s Chapter Six. The finding is narrower: an instrument that establishes facts was repeatedly used where an instrument that prosecutes was needed, and appointing the first has been available whenever the second was not.
“Delay” is not a mechanism. The mechanisms are: commissions that report to the body they examine, committees that can only recommend registration, writ petitions by the named, untraced closures, unmoved jurisdiction, and the appointment of a new body as the answer to demands about the last one.
Fourteen bodies established the death toll and the conduct. They did not prosecute, because that is not what they are for.
The accused could challenge a committee’s jurisdiction in court. The families had no case to protect, because no case existed.
4What Was Paid
Eight and a half lakh rupees over thirty years, in four instalments. Two of them were ordered by a court, and the families had to bring the case.
4.1 — The four tranches
Compensation for November 1984 was not paid once. It was paid four times across three decades, and the sequence is more revealing than the total.
| When | What was paid | Who caused it |
|---|---|---|
| 1984–85 | ₹20,000 per death. ₹500–2,000 for injury. ₹5,000 for a house destroyed completely; ₹1,000 for minor damage. | The government of the day, immediately after. |
| 1996 | Enhanced to ₹3.5 lakh per death, with the balance of ₹3.3 lakh to be paid within four months. | A Delhi High Court judgment in a case brought by a widow whose husband was pulled from a train. |
| 2005–06 | ₹1.23–1.25 lakh to each of roughly 2,800–3,000 injured people, who had originally received ₹2,000. | A Delhi High Court judgment. The Delhi government decided not to appeal. |
| Jan 2006 | A ₹717 crore rehabilitation package, paying the ₹3.5 lakh the court had ordered a decade earlier. | The Union government, following the Nanavati Commission report of February 2005. |
| Oct 2014 | An additional ₹5 lakh per death, for the families of all 3,325 killed nationally. About ₹166 crore. | The Union government, following representations, thirty years afterwards. |
Of the ₹717 crore announced in 2006, about ₹517 crore was spent. The remaining ₹200 crore was not distributed, on account of disputes over who the claimants were.
4.2 — Who had to ask
Read the third column again.
The increase from ₹20,000 to ₹3.5 lakh came from a judgment in a case a widow brought. The compensation for the injured came from a judgment in a case an injured man brought. The package of 2006 followed a commission report that itself followed twenty years of demands. The payment of 2014 followed representations by Sikh organisations.
Not one of the increases originated with a government deciding, on its own, that more was owed.
And the delay after each is its own item. The court ordered ₹3.5 lakh in 1996 and directed payment within four months. Most families did not receive it until the package of 2006 — ten years later. In one case that reached the Delhi High Court again in 2024, a man had been paid a sum in 2016 that had been due since 2006, and the Court ordered ten per cent interest for the intervening decade, describing the administration as insensitive and callous and imposing costs on the Union government.
Every figure and date in section 4.1 is a government order, a court judgment or a parliamentary answer.
The relief rates of 1984–85 were notified. The 1996 and 2005 enhancements are reported judgments. The January 2006 package was announced with a figure and its disbursement was reported to Parliament, including the fact that about ₹200 crore of ₹717 crore was not spent. The October 2014 enhancement was announced by the Home Minister and confirmed in a written answer in the Rajya Sabha.
That is why the third column of the table can be stated as fact rather than characterisation: the causal chain is visible in the documents, because each payment cites what preceded it.
What it cannot show: how many eligible families never claimed. A claimant must know the scheme exists, produce proof of a death certified in circumstances where certification frequently did not happen, and establish relationship. The undisbursed ₹200 crore is the visible part of that problem and there is no measure of the invisible part.
4.3 — How it compares
Part Ten’s “what comes next” asked how the compensation for 1984 compares with what has been paid after other organised violence in India. The answer is that it is broadly typical, and that this is the finding rather than a defence.
The central scheme for civilians killed in communal, terrorist or Maoist violence paid ₹3 lakh until 2014, when it was raised to ₹5 lakh — which is the figure the 1984 families received in their fourth tranche, thirty years afterwards. In other words, the enhanced payment of 2014 brought them to the general rate, and no further.
State-level awards have varied and courts have repeatedly had to intervene. A High Court in 2006 directed two state governments to pay ₹2 lakh with an additional forty per cent to families of those killed in their territories in November 1984 — twenty-two years later, and again on a petition.
Two things follow.
There is no special scheme. A person killed by an organised mob in 1984 is compensated, in the end, on the same schedule as somebody killed by a bomb or caught in crossfire. India has a general rate for a civilian death caused by violence the state failed to prevent, and it applies it.
The general rate is set administratively and raised by litigation. The pattern in section 4.2 is not peculiar to 1984. It is how the whole category works, which is why the same sequence of court orders and delayed disbursement appears after other episodes.
So the comparison does not show that the families of November 1984 were treated worse than others. It shows something less satisfying and more useful: that the arrangement they encountered was the ordinary one, and that its features — payment without liability, a low administrative rate, and increases obtained by suing — are what any Indian family in that position meets.
4.4 — Tilak Vihar
Families who had lost the men of the household were resettled, and a large number of them ended up in one area of west Delhi.
It has been known for forty years as the Widows’ Colony.
That is not an official designation. It is what the place is called, by the people who live there and by everybody else, because that is what it is: a neighbourhood constituted out of households that lost their men in a single week, still identifiable four decades later, still occupied substantially by the same families and their descendants.
A state that resettles bereaved families together produces a place where the loss is the shared characteristic and does not disperse. Whether that is a kindness or the opposite is genuinely arguable. What it is not is neutral.
Compensation and reparation: two different things that the same money can be called.
Compensation is payment for a loss, regardless of who caused it. Governments pay it after floods and earthquakes.
Reparation is payment for a wrong the payer is responsible for. It carries an admission.
Everything paid for November 1984 has been framed as compensation, and specifically as ex gratia — a payment made without accepting any liability at all.
Why it matters here: the framing is why the amounts could be set at ₹20,000 in the first place, why increases required litigation, and why a payment thirty years later can be announced as generosity rather than as a debt discharged late.
Which raises a question survivor organisations have disagreed about among themselves since the first payment.
Survivor organisations have argued about this since the first payment and the disagreement is genuine.
Convictions after forty years bring nobody back and reach a handful of families. Money reaches everybody on the list, is measurable, and does the thing bereaved households most immediately need, which is to survive the loss of an earner. A state that pays and does not prosecute has done something real; a state that prosecutes twelve murders and pays nothing has done less for the other 2,721 families.
Every rupee has been paid ex gratia — expressly without admitting that anything was owed. That framing is not incidental; it is what allowed ₹20,000 in 1984, what made every increase require litigation, and what permits a payment thirty years late to be announced as generosity. A state that pays while denying liability has bought the appearance of redress at the price of the admission, and the admission is the part that would have changed anything.
What would settle it: nothing evidential. It is a question about what a state owes.
Why people care so much: because forty-one years on, the families of 1984 have received about eight and a half lakh rupees each and no Indian government has ever accepted that it owed them anything at all.
Underneath that disagreement, though, both positions share a picture of where the money came from.
Everybody who discusses relief for 1984 assumes that compensation is something a state gives.
The critical version: what was given was inadequate and late. The defensive version: substantial sums were eventually given, by governments of both parties. Both are arguments about the generosity of a giver.
Almost none of it was given. It was extracted, case by case, by people who had to hire lawyers and sue the government that had failed to protect them.
Follow what that requires of a claimant. She has lost the earning member of her household. She is living in a resettlement colony. She must find a lawyer, sustain a case for years, and prove her entitlement to a state that has classified her husband’s death as a matter on which it accepts no liability. If she wins, the ruling benefits others in her position — which is why a handful of names appear in judgments that determined the entitlements of thousands.
The design consequence is the important one. A system in which relief is set administratively at a low figure and raised only by litigation has transferred the cost of establishing the correct figure onto the people least able to bear it, and has made the passage of time work for the payer, because every year of delay is a year of the money’s use.
And it explains the shape of the sequence. Not a government reconsidering, but a court ordering; not a court ordering once, but three times across three decades; and each order arriving after the previous one had been ignored long enough to require a new case.
The general point applies well beyond 1984. Where a state’s liability is framed as grace, the burden of proof moves to the victim and the clock works for the state. That is not a failure of the arrangement. It is what the phrase ex gratia is for.
What the first figure meant to a household is worth stating before the summary.
Twenty thousand rupees in 1984 was roughly two or three years of a labourer’s earnings.
Set against what a household had actually lost — the man who earned, in most cases the only one — it was a payment that covered the gap for as long as it took the family to understand that nothing else was coming.
The court that raised it to three and a half lakh did so in 1996, twelve years later. Most families received that money in 2006, twenty-two years after the death, by which time the children who had lost a father had raised their own.
Compensation delayed by a generation is not late payment. It is payment to different people, for a different purpose, at a point when the thing it was for has already been survived without it.
One even-handed note. The 2006 package and the 2014 enhancement came from governments of opposite parties, and both were substantial. The objection in this chapter is to the mechanism and the framing, not to the fact of payment, and a reader who takes from it that nothing was paid has read it wrongly.
₹20,000 per death in 1984. Raised to ₹3.5 lakh by a court in 1996, paid to most families in 2006. A further ₹5 lakh in 2014. About ₹8.5 lakh over thirty years, in four instalments.
Not one increase originated with a government deciding more was owed. Two came from judgments in cases brought by a widow and by an injured man.
Families were resettled together in an area of west Delhi that has been called the Widows’ Colony for forty years. Where liability is framed as grace, the clock works for the state.
5The Word
It is still officially called a riot. Part One promised to name the concealing word every time. This is the one that has done the most work for the longest.
5.1 — What the word means
A riot, in ordinary usage and in Indian criminal law, is a violent disturbance by an unlawful assembly. The elements that go with the word are spontaneity, disorder, and — usually — more than one group involved.
Part Ten set out what is documented about November 1984: crowds arriving with electoral rolls and ration records to identify households by religion, in areas where the crowds did not live, supplied with kerosene, while the police did not intervene.
None of those elements is compatible with the word. A spontaneous crowd does not carry an electoral roll.
The word is nevertheless the official one. The commissions are titled with it. The government schemes use it. The court records use it. Forty-one years on, the standard English phrase in Indian public life for what happened is the 1984 anti-Sikh riots.
5.2 — Why the word is there
Here is the part that matters, and it is not primarily about politics.
Indian criminal law has no offence that fits.
The Penal Code has rioting. It has unlawful assembly. It has murder, arson and grievous hurt. Each of those is an offence committed by an individual on an occasion.
What it does not have is an offence describing organised mass violence directed at a group, with the features that distinguish it: prior preparation, identification of targets by community, distribution of materials, and the acquiescence of officials whose duty was to prevent it.
Because there is no such offence, there is no charge to lay. Because there is no charge, the acts have to be disaggregated into thousands of individual murders and arsons, each requiring its own evidence, its own witnesses and its own trial — which is Chapter Three’s machinery, arriving as a consequence of a gap in a statute.
And because the law has only the smaller word, the smaller word is what the paperwork says, and what the paperwork says is what the country calls it.
Three words are proposed as alternatives to “riot” and they are not interchangeable.
Pogrom describes an organised, officially tolerated attack on a minority. It is a historical and political term, not a legal one, and it fits the documented features of November 1984 more closely than any other available word.
Massacre describes the killing of many people. It captures the scale and nothing about the organisation.
Genocide is a legal term with a defined meaning: acts committed with intent to destroy a national, ethnic, racial or religious group as such. The intent requirement is very demanding and has rarely been established anywhere.
Why it matters here: victim organisations use the third, critics object that it does not fit, and the argument consumes the energy that the second side of the box below says should go to the absence of any Indian offence at all.
5.3 — What was proposed and not done
The gap has been identified repeatedly and legislation has been attempted.
A Communal Violence Bill was drafted in 2005 and revised in successive versions over the following decade. It would have created offences for organised communal violence, addressed the liability of public servants who failed to act, and provided for relief and reparation as entitlements rather than as grace.
It was never enacted. Versions were opposed on federalism grounds — policing is a state subject — and on the ground that the definitions were too broad, and by parties who believed it was aimed at them.
Forty-one years after November 1984, and after every episode in Chapter Six, there is still no Indian offence of organised mass violence against a community.
The claim in section 5.2 — that Indian law contains no offence fitting organised mass violence against a group — is a claim about the contents of a statute, and it is checkable by reading one.
The Penal Code’s provisions on rioting and unlawful assembly describe conduct by a crowd on an occasion. Its provisions on murder, arson and grievous hurt describe acts by individuals. Nothing in it addresses prior organisation, the targeting of a group as such, or the liability of an official who could have prevented the acts and did not.
The successive Communal Violence Bills are the other half of the evidence. Their existence, their drafting history since 2005 and their contents establish that the gap has been identified repeatedly, by people whose job was to identify it, and that filling it was drafted and not enacted.
What it cannot show: whether an offence would have produced convictions. No Indian prosecution has ever been brought under such a provision because none exists, so the comparison is with other countries and their records are mixed.
The practical consequence of that absence is best seen from the desk of somebody trying to work around it.
Consider what a prosecutor in 2016 actually had to prove, reopening a case from 1984.
Not that an organised attack on a community took place — nobody disputes that and there is no offence of it. He had to prove that this named man struck this named victim at this address on this afternoon, beyond reasonable doubt, on the evidence of witnesses who saw it thirty-two years earlier.
Multiply by 2,733.
That is not a system failing to prosecute a massacre. It is a system with no concept of a massacre, being asked to process one as several thousand unrelated afternoons.
Whether that is a problem about vocabulary or about statute is genuinely disputed.
Victim organisations have argued about this for four decades. The objection to their position is serious.
Nobody was acquitted because of a noun. The convictions that exist were obtained under the ordinary law of murder, and the cases that failed did so for want of evidence, not for want of a better label. Changing the word would not have registered one additional first information report in November 1984. And the alternatives proposed carry their own problems: “genocide” has a specific legal meaning with an intent requirement that would be extremely hard to establish, and using it loosely weakens it everywhere.
The argument is not about a label but about a charge. Because there is no offence of organised communal violence, prosecutors must prove thousands of separate murders, each on its own facts, forty years later. Command responsibility — the liability of an official who could have stopped it and did not — does not exist as a charge, which is why convicting a policeman is nearly impossible. The word “riot” is not the cause of that; it is the symptom, and it is accurate about the law as it stands. The law is the problem.
What would settle it: nothing about vocabulary. Enactment would.
Why people care so much: because for survivors the word is an official description of what was done to their families, and being told it was a riot is being told it was nobody’s plan.
One last observation. The word has survived forty-one years of objection because it is accurate — about the law. Change the law and the word changes by itself, which is why Chapter Nine’s fifth rule is the one that would do the most work.
A riot is spontaneous, disorderly and usually two-sided. A spontaneous crowd does not carry an electoral roll. The word is nevertheless the official one, forty-one years on.
The reason is not primarily political. Indian criminal law has no offence of organised mass violence against a group, so the acts must be broken into thousands of separate murders, each needing its own evidence and trial.
A Communal Violence Bill was drafted in 2005 and repeatedly afterwards. It has never been enacted.
6The Comparison
Four later episodes under four governments of three different parties. The pattern is not partisan, and establishing that is not an exoneration of anybody — it is a worse finding.
6.1 — The set
A charge that a particular party protects its own is testable. Look at what happened after comparable episodes under other parties.
| Episode | Who was governing | What followed |
|---|---|---|
| Delhi, Nov 1984 | Congress at the centre | 2,733 killed. Fourteen inquiry bodies. 587 reports, ~240 untraced. 25 convictions, 12 for murder. First senior conviction 2018 — thirty-four years. |
| Hashimpura, May 1987 | Congress in Uttar Pradesh | Around forty Muslim men taken from a locality by an armed constabulary and shot. Trial began in 2006 after transfer out of the state. Acquittals in 2015. Sixteen convicted and sentenced to life by the Delhi High Court in 2018 — thirty-one years. |
| Bhagalpur, Oct 1989 | Congress in Bihar | Around a thousand killed. Convictions in 2007 — eighteen years. |
| Bombay, Dec 1992 – Jan 1993 | Congress in Maharashtra, then Shiv Sena–BJP from 1995 | A commission first refused by the Congress Chief Minister and appointed at the Prime Minister’s intervention. Disbanded in January 1996 by the successor government, revived by the Bombay High Court, reported in 1998. Recommendations largely not implemented. Around twenty convictions by 2017. |
| Gujarat, Feb–Mar 2002 | BJP in Gujarat | Over a thousand killed. Investigation eventually placed under a Supreme Court-monitored special team from 2008, with several trials transferred out of the state. Convictions in a number of major cases. A plea alleging a larger conspiracy by the state administration was rejected through the courts. |
Before drawing anything from it, the most striking column in that table is the one nobody reads across.
Line up the intervals in that table.
Bhagalpur: eighteen years to a conviction. Bombay: about twenty-four years to a substantial number. Hashimpura: thirty-one. Delhi: thirty-four to the first senior one.
A child born on the day of any of those episodes was a middle-aged adult before a court convicted anybody of it.
And in every case the accused had aged at the same rate. Prosecutions of this kind are races between a criminal process and human mortality, run over decades, in which the process has never once been the faster.
6.2 — The sharpest single comparison
Take Bombay on its own, because it contains a controlled experiment.
The riots of December 1992 and January 1993 were followed, in March 1993, by a series of bomb blasts in the same city.
The blasts were prosecuted under special legislation, in a dedicated court, over many years, and produced a large number of convictions including death sentences.
The riots that preceded them, in the same city, in the same months, affecting far more people, produced a commission that was disbanded and revived, recommendations that were not implemented, and convictions in the low tens.
Same city. Same period. Same criminal justice system. Same police force.
The difference is not capacity. The Indian state can prosecute mass-casualty crime when it decides the crime is terrorism. What it has never built is the equivalent apparatus for organised violence by a majority against a minority, and Chapter Five explains why: there is no offence to prosecute.
Each row of the table is drawn from the public record of that episode: commission reports where they exist, court judgments, and reported conviction outcomes.
The comparison is deliberately built on cases where the essential facts are agreed by both sides of Indian politics. That Hashimpura took thirty-one years is not contested. That the Srikrishna Commission was disbanded by the Maharashtra government in January 1996 and revived by the Bombay High Court is not contested. That a Supreme Court-monitored team was constituted for the Gujarat cases is not contested.
What it cannot show: whether the episodes are equivalent in kind. They differ in scale, in who the victims were, in whether the violence was one-sided, and in what the state’s role was. A table cannot settle those differences and this chapter’s claim is deliberately narrow — it is about the processing afterwards, which is comparable, and not about the events, which are not.
6.3 — What predicts a conviction
Across the whole set, one variable tracks outcomes and it is not party.
Whether the investigation was removed from the local police and placed under judicial supervision, and whether trials were moved out of the affected jurisdiction.
Hashimpura produced convictions after the case was transferred to Delhi. Gujarat produced convictions after a Supreme Court-monitored team was constituted and trials were moved. Delhi 1984 produced its convictions after teams under Supreme Court supervision reopened closed files from 2015.
Bombay, where the commission was disbanded by the state government and the investigation stayed where it was, produced the fewest.
Which produces a conclusion that neither side of Indian politics has any use for.
Everybody arguing about impunity for communal violence in India assumes that it is a party problem.
One side says Congress protected the killers of 1984. The other says the BJP protected the killers of 2002. Each cites the other’s episode as proof of the other’s character, and both are entirely correct about the episode they name.
The table above contains five episodes under three parties across four states and thirty years, and the outcomes are broadly the same. Two decades to a conviction, or three. A commission whose recommendations were not implemented. A police force investigating itself. Compensation set low and raised by litigation.
A pattern that survives every change of party is not a fact about parties.
This is a worse finding than the partisan one and both sides should notice why. If impunity is a party’s habit, it ends when the party does, and voting differently is a remedy. If it is a property of the rules — of who investigates whom, of which court has jurisdiction, of what offence exists to charge — then it survives every election, and it has.
Note also what the partisan framing has cost. Forty-one years of Indian argument about 1984 and twenty-four about 2002 have been conducted as a competition in atrocity, in which each side’s evidence for its own grievance is the other’s proof of hypocrisy. In all that time the Communal Violence Bill of Chapter Five has been drafted repeatedly and passed never — because each version was opposed by whichever party expected to be its target, which is precisely what you would expect if both were right about each other.
The finding is not that everybody is equally guilty. Congress presided over 1984 and that is this book’s subject. The finding is that the machinery which processed it afterwards was not built by Congress and has not been dismantled by anybody, and that every party has since used it.
One thing this chapter is not doing. It is not saying that because others did the same, the government of 1984 is excused. This is a book about that government and Part Ten reached its verdict. The comparison establishes something different and more useful: that the machinery which processed the aftermath was not that government’s invention and has outlasted it by four decades, and that the argument between the parties about whose massacre was worse has been the most reliable obstacle to changing it.
Five episodes, three parties, four states, thirty years: Delhi 1984, Hashimpura 1987, Bhagalpur 1989, Bombay 1992–93, Gujarat 2002. Two to three decades to a conviction in each.
Bombay is the controlled experiment. The 1993 blasts produced death sentences; the riots months earlier in the same city produced convictions in the low tens. The state can prosecute mass-casualty crime when it calls it terrorism.
One variable predicts convictions and it is not party: whether investigation was removed from the local police and put under judicial supervision, and whether trials were moved out of the state.
7The Pattern of Four
Part Two’s rule about counting has now been applied four times across this series. Four cases is enough to say what it establishes, and the answer is not the one the rule was written to expect.
7.1 — The rule
Part Two set it out: when a state that counts everything has no count of something, somebody decided that.
The reasoning was that modern states count compulsively. They count births, deaths, harvests, rainfall, cattle, literates and voters, because counting is what administration consists of. So an absence in a state’s numbers is not usually an oversight. It is a decision, and it is worth asking who made it.
The rule has been applied in four places.
7.2 — The four
| Case | Was there a count? | What happened to it |
|---|---|---|
| Partition, 1947 (Part Two) | No. Estimates range from about 200,000 to two million and none is authoritative. | No count was ever attempted. Both successor states were three weeks old, and both had reasons not to establish which community had suffered more. |
| Hyderabad, 1948 (Part Two) | Yes. A committee appointed by the Prime Minister visited nine of sixteen districts and produced a figure. | Not published for sixty-four years. The count was made and withheld. |
| The Emergency, 1975–77 (Part Nine) | Yes, thoroughly. 110,806 detained, 25,962 employees retired, 1,774 deaths from sterilisation procedures. | Published in three volumes in 1978 — and withdrawn from circulation after 1980 when the government changed. It survived in foreign libraries. |
| November 1984 (Parts Ten and Eleven) | Yes, precisely. A committee was appointed for the specific purpose and reported 2,733 in Delhi and 3,325 nationally. | Published, accepted, repeated by every subsequent body, used in every government scheme — and followed by twelve murder convictions in forty-one years. |
7.3 — What four cases establish
Read the third column downward and something unexpected appears.
The transparency improves at every step. No count. A count suppressed. A count published and then recalled. A count published, accepted, and permanently available.
By any ordinary measure of how open a state is about its own failures, the Indian state got substantially better between 1947 and 1987. The 1984 count is a public document that nobody has ever tried to withdraw, and it is used by the government itself in every relief scheme.
And the accountability does not improve at all.
Partition produced no prosecutions and nobody expected any. Hyderabad produced none. The Emergency produced a commission whose findings were withdrawn and no convictions of consequence. November 1984 produced the most complete count in the series and twelve murder convictions.
So the state learned to count and did not learn to prosecute. Those are different capacities, they were assumed to be connected, and four cases now suggest they are not.
Between 1948 and 1987 the Indian state’s ability to establish what it had done to its own people improved out of recognition.
Hyderabad in 1948 required a committee to travel to nine districts and produce an estimate that was then locked away for two generations. November 1984 produced a precise figure within three years, published, accepted and used in government schemes ever since.
Over the same period the number of murder convictions arising from either episode went from zero to twelve.
One capacity was built and the other was not, and it is worth asking why anybody assumed they were the same capacity.
7.4 — The case that breaks the pattern
And then there is Punjab between 1987 and 1993, from Part Ten’s Chapter Eight.
No count. No commission. No committee. No official figure for the dead, the detained or the disappeared. One private inquiry into one district, compiled from cremation-ground firewood receipts by a man who was abducted and killed for it.
That is not a slightly worse case than the four above. It is the 1947 position, arriving forty years later, in a state with a functioning administration that was counting everything else.
So the progression in section 7.3 is not chronological and not a story of national maturation. Something else determines whether a count exists, and the five cases together point at it clearly.
Where the killing was done by a party’s supporters, there was a count. Delhi 1984 was investigated fourteen times, substantially by successor governments with an interest in investigating it.
Where the killing was done by the police and the army, there was nothing. Punjab 1987–93 has never been examined by anybody, and no government of any party has proposed it.
That distinction is the finding of this chapter and it holds across every case in this series.
Which unsettles the belief that organises almost all campaigning on these subjects, including a good deal of this series.
Everybody who campaigns about these episodes assumes that the problem is that India does not know what happened.
It is the organising belief of every commemoration, every documentary and every demand for a fresh inquiry. If the facts were established, if the truth were officially acknowledged, if the country were made to see — then something would follow. Both sides accept it: campaigners demand the truth and governments respond by appointing bodies to establish it.
India knows. That is what the table above shows.
The death toll of November 1984 is a government figure, produced by a government committee, published, accepted by every subsequent inquiry and used in government schemes. The conduct of the police is a finding of commissions appointed by the government. The use of voter lists is a finding. Political patronage and indifferent policing are the words of the Delhi High Court in a judgment.
There is no meaningful sense in which the Indian state does not know what happened in Delhi in November 1984. It has said so, repeatedly, in its own documents, for forty years.
Which means the demand for truth has been satisfied and it changed almost nothing — and that the machinery which produced fourteen inquiries was answering a question that was not the operative one.
The operative question is not what happened. It is what a state does with knowledge it already has, and the answer requires a rule about who investigates, who prosecutes, in which jurisdiction, under whose supervision, and for what offence. Not one of those was addressed by any of the fourteen bodies, because none of them had the power to.
This is uncomfortable for everybody working on these subjects, including me. Establishing the record is the thing writers and campaigners can do, and it is worth doing, and Part Ten’s Chapter Seven does it. But four cases now indicate that establishing a record is not a step towards accountability. It is a substitute that has been available whenever accountability was not.
A claim of this kind should carry a statement of how far it can be pushed.
The five cases in this chapter are drawn from earlier parts of this series and each rests on the evidence set out there: Part Two for Partition and Hyderabad, Part Nine for the Emergency, Parts Ten and Eleven for 1984, and Part Ten’s Chapter Eight for Punjab.
The pattern claim is therefore only as strong as those parts, and it is stated as a pattern across five cases rather than as a law. Five is enough to notice something and not enough to prove it.
What it cannot show: whether the distinction in section 7.4 — that a count exists where a party’s supporters killed and does not where the police and army did — holds outside this series. Testing it would require examining every Indian episode of state or communal killing since 1947 against whether an official count was produced. Nobody has done that, and it is the single most useful piece of work anybody could do on this subject.
Which is the most useful thing in this part for anybody deciding what to do next, and the least useful for anybody deciding what to say.
Four applications of Part Two’s rule: Partition (no count), Hyderabad (counted and withheld sixty-four years), the Emergency (counted, published, withdrawn after 1980), 1984 (counted precisely and permanently available).
Transparency improved at every step and accountability did not. The state learned to count and did not learn to prosecute, and those turn out not to be connected.
Punjab 1987–93 breaks the sequence entirely — no count at all. Where the killing was done by a party’s supporters there was a count. Where it was done by the police and army there was nothing.
8What Actually Worked
One device produced convictions after three decades of failure. It was invented for a different massacre, under a different party, and took thirty-one years to reach 1984.
8.1 — The device
It has three components and they only work together.
A special investigation team. Investigation removed from the force whose jurisdiction it would ordinarily be, and from the force implicated in the events.
Supervision by the Supreme Court. Not merely appointment by it, but continuing supervision — the team reports to the Court, and the Court can direct.
Power to reopen closed cases, and where necessary to move trials out of the affected jurisdiction.
Take away any one and the arrangement fails. A team appointed by a government reports to a government. A team with no power over closed files can only work on what was already built. A trial held where the accused and the witnesses are neighbours produces the outcomes it has produced for forty years.
Continuing mandamus: the practice by which a court, instead of ruling once and closing a case, keeps the matter before it and issues directions over months or years as the investigation proceeds.
It is the crucial component and it is not a statute. It is a technique Indian courts developed, and it works because the investigating agency must return and report, which converts a one-off order into a supervised relationship.
Its weakness is exactly its strength. Because it depends on a court choosing to retain a matter, it is available where litigants can sustain proceedings for years and unavailable everywhere else.
Why it matters here: this is the difference between an investigation team appointed by a government and one that answers to a judge, and it is why the 2015 and 2018 teams produced what thirty years of committees did not.
8.2 — Where it came from
Not from 1984.
The components were assembled in the litigation following Gujarat in 2002 — a retrial moved out of the state in 2004, and a Supreme Court-monitored investigation team constituted in 2008 to examine major cases.
That was done under a government of a different party, about an episode in which a different party’s state administration stood accused, and it was done because the Supreme Court took control of the litigation.
It was applied to November 1984 from 2015, when a team was constituted to re-examine cases, and again in 2018, when a further team was set up under Supreme Court supervision to look at 186 closed files.
8.3 — What it produced
In November 2018, two men were convicted of murders committed in Mahipalpur in 1984. The case had been registered in 1993 on a survivor’s affidavit to a commission, and closed by the police in 1994 for want of evidence. The team reopened it in 2016 after scrutinising 293 cases previously marked untraced or cancelled, re-examined eighteen witnesses, and secured convictions.
Read that sequence once more, because it settles Chapter Three’s argument.
The evidence was there in 1993. It was there in 1994 when the case was closed. It was there in 2016 when the case was reopened. Eighteen witnesses, still alive, still able to give evidence, twenty-two years after the file had been marked untraced.
The cases were available and had not been built.
8.4 — What it does not fix
Four things, and they are why this chapter is not an ending.
It depends on the Supreme Court. The device requires a court to take and keep control of a matter, which no rule obliges it to do and which happens where litigants with resources sustain proceedings for decades.
It is retrospective. Everything it can do depends on a file existing. Where no first information report was ever recorded — which is most of November 1984 — there is nothing to reopen.
It is case by case. It is not a rule about how communal violence is investigated in India. It is a series of directions in particular litigations, available to whoever gets to the Court.
It does nothing about the first week. Chapter Two’s finding stands entirely untouched: the local police still take the complaints, still investigate, and still classify. Every element of the sequence that made 1984 unprosecutable is still in place for the next occasion.
Thirty-one years is the gap between the killing and the arrival of the method that could investigate it.
Put in personal terms: a man who was thirty when he saw his neighbour killed was sixty-one when an investigator with the necessary powers finally came to ask him what he had seen. He was still there. He remembered. He gave evidence and a conviction followed.
Nothing had changed about him in those thirty-one years except his age. What changed was who was allowed to ask.
Unusually for this series, that claim rests on events recent enough to be checked directly.
The reopening sequence is documented in unusual detail because it happened recently and in public: the constitution of the teams, the number of files scrutinised, the number of witnesses re-examined, and the resulting judgments are all on the record and were reported at the time.
That is why this chapter can make a claim the rest of the series usually cannot: it is not arguing that the cases might have been buildable in 1984. It is pointing at cases that were built in 2016 to 2018 from material that existed throughout, and to a court’s finding that the earlier closures were unjustified.
What it cannot show: how many of the roughly 240 untraced files would have produced convictions if reopened. The 2018 team was given 186 of them and the work is not complete. Every number here is a floor.
Which raises the obvious question, on which people who agree about everything else in this chapter divide.
It works. Whether it should be converted from a judicial practice into a statutory entitlement is genuinely contested among people who agree it works.
A remedy available only to litigants who can sustain proceedings before the Supreme Court for years is a remedy for the organised and the well-advised. Most victims of communal violence in India are neither. Writing the device into statute — automatic transfer of investigation on a defined trigger, mandatory judicial supervision, a duty to review closures — makes it available to everybody without requiring anybody to reach Delhi.
The device works because a court chose to keep a matter and could shape its directions to the case. A statutory version would be administered by the executive, with defined triggers that would be litigated, thresholds that would be argued down, and supervision by whoever the statute names — which in practice means officials of the kind who did not act in the first place. Judicial improvisation succeeded here precisely because it was not a rule anybody could game in advance.
What would settle it: whether any of the six has ever been enacted anywhere in India, even at state level, and with what result. No state has.
Why people care so much: because the alternative to codifying it is depending permanently on the Supreme Court’s appetite, and that appetite has varied a great deal across the period this series covers.
Which is why this chapter is called what it is rather than “How It Was Fixed”. One device worked, three decades late, in a fraction of the cases, because a court chose to take an interest. Nothing about the arrangement makes that repeatable.
The device has three parts and needs all three: investigation removed from the implicated force, continuing Supreme Court supervision, and power to reopen closed files and move trials.
It was assembled for Gujarat after 2002, under a different party, and reached 1984 in 2015 — thirty-one years late. A 1993 case closed in 1994 was reopened in 2016, eighteen witnesses re-examined, and convictions followed in 2018.
The cases were available and had not been built. And the device fixes nothing about the first week, which is where the failure happens.
9What Would Have Prevented It
Six rules. Each could have been enacted at any point in forty-one years, by any government. Not one of them has been.
9.1 — Four arguments for the defence
These are the strongest arguments available. The fourth is genuinely strong and this part concedes it.
One. India did not legislate an amnesty. That is the ordinary response of a state whose own supporters have committed a massacre, and it is what a large number of countries have done in comparable circumstances. India did not. Cases remained open, commissions kept being appointed, and forty-one years later a former Member of Parliament of the governing party of the time is serving a life sentence. That is not nothing and it is more than many states manage.
Two. The convictions are real. Obtained in court, after trial, with appeals available and exhausted. They are not gestures.
Three. Compensation was paid, and by both major parties. The 2006 package and the 2014 enhancement came from governments of opposite parties, and the total is substantial. Chapter Four’s objection is to the mechanism and the timing, not to the fact.
Four, and this is the strong one. Mass-casualty prosecution is the hardest thing a criminal system attempts, and nobody does it well. Thousands of accused, thousands of witnesses, no forensic evidence, offences committed in crowds, and a requirement to prove individual acts beyond reasonable doubt. Every country that has tried has taken decades and convicted a fraction. Chapter Six’s table shows the same outcome under three different Indian parties, and the international record is not better.
9.2 — Six rules
Part One requires an alternative that was available. Here are six, all procedural, all enactable by ordinary legislation, none requiring anybody to admit anything.
A duty to register, with a sanction. Refusal to record a first information report is the origin of everything in this part. Making refusal an offence with a named officer responsible would have made 587 reports for 2,733 deaths difficult to produce.
Mandatory transfer of investigation where the police are implicated. Not discretionary, not on application, not requiring a Supreme Court to intervene. Automatic, on a finding by any court or commission that a force’s conduct is in question.
Judicial review of untraced closure. A case cannot be closed for want of information the investigating force declined to gather, unless a magistrate agrees on the material.
Presumptive transfer of trial out of the affected jurisdiction in mass-casualty communal cases, which is what worked at Hashimpura and in Gujarat and was not available in Delhi for thirty years.
A statutory offence of organised mass violence against a group, with command responsibility for officials who could have prevented it and did not. Chapter Five: this does not exist, and the Bill that would create it has been drafted repeatedly since 2005.
A duty to act on a commission’s findings or to explain publicly why not, within a stated period. At present a government may commission an inquiry, receive it, and do nothing, without ever being required to say so.
Each of those is a paragraph of a statute. Not one requires a constitutional amendment. Not one has been enacted.
The six rules in section 9.2 are unusual for this series because none of them is a counterfactual about what would have happened. Each is a claim that a rule does not exist, which is checkable by looking.
There is no statutory sanction on an officer who refuses to register a report. There is no automatic transfer of investigation where a force is implicated. Closure as untraced is not subject to mandatory judicial review. Transfer of trial requires an application to a superior court. There is no offence of organised mass violence against a group. And the Commissions of Inquiry Act imposes no duty on a government to act on a report or to explain why it has not.
Each of those is an absence in a statute, and absences in statutes are among the few things that can be established conclusively.
What this cannot show: that enacting them would have worked. That is the argument in the box below, and the honest position is that the evidence for it is Chapter Eight’s single case — where changing who investigated produced convictions in three years — and that one case is suggestive rather than conclusive.
9.3 — What I would have to retract
Three things would break this part.
If the Delhi Police case disposal records showed that the untraced closures were made on genuine material — that the investigations were attempted and produced nothing — Chapter Two’s argument weakens substantially and Chapter Three’s mechanism fails.
If the 2018 team’s work on the 186 reopened files produces almost no further convictions, then the 2016 result was exceptional rather than indicative, and Chapter Eight’s central claim is much weaker than stated.
If a systematic comparison showed that countries with a statutory offence of organised communal violence do no better on conviction rates and timescales, the fifth rule in section 9.2 loses most of its force.
The first depends on records the Indian state holds. The second will be answered by events. The third is ordinary academic work that anybody could do.
Price the six rules in section 9.2.
Each is a paragraph of ordinary legislation. None requires a constitutional amendment, a two-thirds majority, or the consent of any state government except where policing powers are touched — and two of the six touch nothing but central law.
The Communal Violence Bill has been drafted, redrafted and circulated since 2005. Parliamentary time has been found in those twenty years for several hundred other Acts.
What the six would have cost is not money and not constitutional difficulty. It is that each of them makes somebody explain a decision they can currently take in silence.
Whether writing them down would have made any difference is the objection this chapter has to meet.
Proposing rules is easy and the objection to it is serious.
A duty to register a report is already implicit in police law and is routinely disregarded. Mandatory transfer requires somebody to make the finding that triggers it. A statutory offence requires a prosecutor to charge it. Every one of these six depends on officials choosing to apply it, and the officials in question are the ones who did not act in November 1984. Writing more rules for people who ignored the existing ones is the standard reflex of a legalistic mind and it changes nothing.
None of the six guarantees an outcome and that is not the test. What a rule does is shift the default and create a record. A magistrate reviewing an untraced closure produces a document with a name on it. Presumptive transfer means somebody must argue for keeping a case where it is. A duty to respond to a commission means silence becomes a decision that has to be signed. The evidence that this works is in Chapter Eight: the single device that produced convictions did so by changing who was permitted to investigate, not by improving anybody’s character.
What would settle it: enacting one of them and observing the next episode. India has had five in forty-one years and has enacted none.
Why people care so much: because the first argument is the one every government makes when a Communal Violence Bill is introduced, and it has succeeded every time since 2005.
That is the case against this part, and the fourth defence I think simply wins on its own terms — mass-casualty prosecution defeats every country that attempts it. What it does not reach is section 9.2, because none of those six rules is about prosecuting a massacre. They are about not destroying the evidence of one in the first fortnight.
Four defences: India did not amnesty; the convictions are real; both parties paid compensation; and mass-casualty prosecution is the hardest thing any criminal system attempts. The fourth is conceded.
Six rules, all ordinary legislation: a duty to register with a sanction; mandatory transfer where police are implicated; judicial review of untraced closure; presumptive transfer of trial; a statutory offence of organised mass violence with command responsibility; and a duty to act on a commission’s findings or say why not.
Not one has been enacted, by any government, in forty-one years.
10An Honest List of What We Do Not Know
This part’s gaps are unusual: almost everything unknown here is a record that exists, is held by the Indian state, and has been asked for.
10.1 — Genuinely unknown
Five things in this part cannot be established, and four of them are documents somebody could produce tomorrow.
Who authorised each untraced closure, and on what material. About 240 cases were closed by the Delhi Police. Chapters Two and Three turn on whether those closures reflected genuine investigative failure or the absence of investigation.
Why it is unknown: the case disposal records have been sought repeatedly, including before commissions. They have not been produced in a form that answers the question.
The Delhi Police deployment and control room records for 31 October to 3 November 1984. Part Ten’s Chapter Six identified their non-production as the specific gap in the question of whether the police inaction was directed.
Why it is unknown: the same. This is the single document set that would most change both parts.
How many of the 186 reopened files will produce convictions. Chapter Eight’s claim that the cases were available rests on a small number of results so far.
Why it is unknown: the work is not finished. This one will be answered by events rather than by disclosure, which makes it the only falsifiable claim in the part with a date attached.
How many officers faced any consequence. Committees recommended action against named officers. What departmental proceedings followed, against whom, and with what outcome, is not systematically published.
Why it is unknown: departmental proceedings are internal. This is an unmade count of the kind this series has now met five times, and the material sits in service records.
How much of the ₹200 crore was never paid, and to whom it was owed. Of the 2006 package, about ₹517 crore of ₹717 crore was disbursed. The balance was not, on account of disputes over claimants.
Why it is unknown: the claimant disputes were resolved case by case over years and no consolidated account has been published.
10.2 — Solid
The count. 2,733 in Delhi and about 3,325 nationally, established by a committee appointed for the purpose, accepted by every subsequent body and used in every government scheme.
The disposal figures. 587 reports, about 240 untraced, about 250 acquittals, 25 convictions of which 12 for murder, about 440 individuals.
The inquiry sequence. Four commissions, nine committees and two investigation teams, each with a date and terms of reference.
The findings about the police. From commissions appointed by the Government of India, including by governments of the party whose local functionaries were being named, and from the Delhi High Court in 2018.
Every compensation figure and date. The ₹20,000 of 1984–85, the 1996 judgment raising it to ₹3.5 lakh, the 2005 judgments on the injured, the ₹717 crore package of January 2006, the additional ₹5 lakh of October 2014, and the 2024 order awarding interest for a decade’s delay.
The comparison set. Hashimpura 1987, Bhagalpur 1989, Bombay 1992–93 and Gujarat 2002, with their governments, their inquiries and their outcomes.
That no offence of organised communal violence exists in Indian law, and that a Bill to create one has been drafted since 2005 and never enacted.
10.3 — What kind of gap this is
Something about the shape of section 10.1 is worth naming.
Part Five ended by naming a calculation nobody has done. Part Six ended by naming a report classified for sixty-three years. Part Seven ended by naming two counts anybody could make. Part Eight and Part Nine ended by naming closed cabinet files. Part Ten ended by naming an investigation that was never held.
This part ends by naming police station records from 1984.
Not cabinet papers, not intelligence assessments, not anything touching national security or the deliberations of ministers. Case diaries, closure reports and duty rosters, of the kind every Indian police station produces every week and retains as a matter of course.
They have been sought by commissions the government itself appointed, by courts, and by litigants, across four decades.
I do not think there is a single explanation and I am not going to invent one. Part Six’s Chapter Eight argued that withholding requires nobody to decide anything while releasing requires somebody to sign, and that inertia explains more than conspiracy. That argument applies here and it is less comfortable than it looks, because it means the records are unavailable for no reason at all — which is a worse thing to say about a state than that it is hiding them.
Genuinely unknown: who authorised each untraced closure, the police deployment records for 31 October to 3 November, how many of the 186 reopened files will convict, how many officers faced consequences, and where the undisbursed ₹200 crore went.
Solid: the count, the disposal figures, the inquiry sequence, the findings about the police, every compensation figure and date, the comparison set, and the absence of any offence of organised communal violence.
Six parts have now ended by naming a document the state holds. This one ends by naming police station case diaries from 1984 — not cabinet papers, not security files. Duty rosters.
Sources & further reading — Part 11
Timeline of the Aftermath, 1984–2025
Dates only. The arguments are in the chapters.
| Date | What happened |
|---|---|
| 31 Oct – 7 Nov 1984 | 2,733 killed in Delhi, about 3,325 nationally. 587 first information reports are eventually registered. |
| 1984–85 | ₹20,000 paid per death; ₹500–2,000 for injury; ₹5,000 for a house destroyed. The Marwah Commission is appointed to examine police conduct and is wound up before completing its work. |
| 1985–86 | The Misra Commission. It recommends the further committees that follow. |
| 1987 | The Ahuja Committee establishes the death toll. The Jain–Banerjee Committee recommends registration of cases; those named challenge its competence in court and its work is stalled. |
| 1987–1990 | The Kapoor–Mittal Committee on police conduct. Its two members cannot agree on method. |
| 1990–1993 | The Poti–Rosha and Jain–Aggarwal Committees succeed Jain–Banerjee on the same question. |
| 1993–94 | The Narula Committee recommends registration against named politicians. Many cases from the period are closed by police as untraced. |
| 1996 | The Delhi High Court, in a case brought by a widow, raises compensation from ₹20,000 to ₹3.5 lakh and directs payment of the balance within four months. Most families do not receive it. |
| 2000–2005 | The Nanavati Commission. It reports in February 2005, naming individuals and finding a colossal failure in the maintenance of law and order. |
| 2005 | The Delhi High Court orders enhanced compensation for roughly 2,800 injured people. The Delhi government decides not to appeal. |
| 16 Jan 2006 | A ₹717 crore rehabilitation package is announced, paying the ₹3.5 lakh the court had ordered a decade earlier. About ₹517 crore is eventually spent. |
| 2008 | In the Gujarat litigation, the Supreme Court constitutes a monitored special investigation team — the device Chapter Eight describes. |
| Oct–Dec 2014 | An additional ₹5 lakh per death is announced for the families of all 3,325 killed, and distributed from December. A committee under a retired Supreme Court judge is set up to oversee it. |
| 2015 | A special investigation team is constituted to re-examine 1984 cases — thirty-one years afterwards. |
| 2016 | The team scrutinises 293 cases previously marked untraced or cancelled and reopens one from Mahipalpur, registered in 1993 and closed in 1994. |
| Jan 2018 | A further team is constituted under Supreme Court supervision to re-investigate 186 closed cases. |
| Nov 2018 | Two men are convicted of the Mahipalpur murders, on the evidence of eighteen re-examined witnesses. |
| Dec 2018 | The Delhi High Court reverses an acquittal and sentences a former Member of Parliament to life imprisonment, finding political patronage and an indifferent law enforcement agency. |
| 2024 | The Delhi High Court orders the Union government to pay ten per cent interest for a decade’s delay in a compensation payment, describing the administration as insensitive and callous. |
| 2025 | A second life sentence is handed down in a separate case. Twenty-five cases in total have produced convictions; twelve are murder cases. |
Part Twelve starts here. It returns to the chronology and takes the government of Rajiv Gandhi, 1984 to 1989 — the largest parliamentary majority in Indian history, and what was done with it.
Glossary
Every hard word used in this part, in plain English.
| Word | What it means |
|---|---|
| Command responsibility | The liability of a person in authority for acts they could have prevented and did not. It exists in international criminal law. There is no Indian equivalent offence, which is why convicting an officer for inaction is nearly impossible. |
| Commission of inquiry | A body appointed under the Commissions of Inquiry Act. It can summon witnesses and take evidence on oath. It cannot prosecute, its findings bind nobody, and a government need not publish or act on them. |
| Ex gratia | “Out of grace”. A payment made without admitting any legal liability. Everything paid for November 1984 has been on this basis. |
| First information report | The document a police station records when a cognisable offence is reported. It begins the criminal process and everything a court can later do is limited by what it contains. |
| Locus standi | The right to bring a case at all. Those named by inquiry committees had it and used it. The families had no case to protect, because none existed. |
| Reparation | Payment for a wrong the payer is responsible for. Distinguished from compensation, which is payment for a loss regardless of cause. |
| Special investigation team | An investigation removed from the force with ordinary jurisdiction. Effective only where it also has continuing judicial supervision and power over closed files. |
| Transfer of trial | Moving a case out of the jurisdiction where the offence occurred, usually because a fair trial is not possible there. It produced convictions at Hashimpura and in Gujarat and was not used for Delhi. |
| Untraced | A police classification closing an investigation because the offenders cannot be identified. About 240 of the 587 reports from November 1984 were closed this way; some were later reopened and produced convictions. |
What Comes Next
Part Twelve — Rajiv Gandhi, 1984 to 1989
Part Twelve returns to the chronology. A government elected weeks after November 1984 with the largest majority any Indian party has ever held. It asks:
- Four hundred and fourteen seats is more than Nehru ever had. What was that majority actually used for, measured against what was promised with it?
- The Shah Bano judgment was reversed by legislation in 1986, and the Ayodhya shrine was unlocked in the same period. What is the documentary record connecting those two decisions, and what is inference?
- The anti-defection law was passed in 1985. What problem did it solve and what did it do to the independence of every Member of Parliament since?
- Bofors is the best-known corruption allegation in Indian history. What has actually been established by a court, and what has not?
- Indian troops were sent to Sri Lanka in 1987. Who decided, on what assessment, and what was the professional military advice?
- Part Eight found the party made into an instrument of one office. Did anything change when the office changed hands, and what does the answer say about the finding?
Questions, not topics. If any of them is answered badly, you will be able to tell.