The harm ledger

Who this happens to

The person this happens to is usually a woman, and she is typically between forty and sixty.

That is where the chapter has to start, because everything after this is a table and she is not a table. She lives in a village. Her accusers are very often her own kin, which is to say the people who have known her longest. Something has gone wrong in that village: a child’s illness, an adult’s illness, a piece of land, a jealousy that has been running long enough for everybody to be able to recite it. None of that is unusual. Villages are full of grievance and most of it stays where it is.

Then a name is said out loud, by somebody whose word carries there.

That is the moment. Nothing new has been discovered about her. No investigation has taken place, and none is going to. What she was on Monday and what she is on Tuesday are the same thing, and the whole of the difference is that somebody with standing said it out loud in front of people who were already unhappy.

Some of the women this happens to are killed.

Those are the ones who can be counted, and even they cannot be counted well. A national count exists and it runs into thousands, and every figure in it arrives with a caveat that has to travel with it. This chapter carries the caveats beside the numbers rather than after them, because a figure about murdered women that is printed without its qualifications gets quoted by somebody else without them.

So the ledger opens on the question it would be dishonest to hold until the end.

Is this corpus why?

That is the question a book about Śābara mantras owes its reader before it prints a single number. Nineteen chapters of a book about a body of vernacular spell literature stand behind this one, and here in the same country, in the same social world, is a documented record of people destroyed by an accusation about exactly this kind of power. If there is a line running between the two, this is the chapter where it would show.

Chapter 18 went looking and found nothing: no journalism, no judgment, no official or non-governmental documentation of any kind naming this corpus as the instrument of a harm to a specific person. That chapter has the anatomy of what came back instead, and this one will not rebuild it. It needs a single line of it to begin.

No documented harm attaches to the Śābara corpus by name.

Which leaves the harder question. If the corpus has no body count, what harm is documented in the ritual world it belongs to, and to what does that harm attach?

The answer is on record, it is substantial, and it does not attach to a book.

The sentence, and where it sits in the report

In December 2021, ActionAid Association and the Odisha State Commission for Women published Witch-Hunting in Odisha, a study built on 102 case files. One sentence in that report is the sentence everybody reaches for, and it will not carry what is usually loaded onto it.

The report is in copyright and this chapter reports it in paraphrase rather than quoting it. The sentence has three moving parts and all three matter. It opens by attributing what follows to the report’s own fieldwork. It is scoped by the words in most cases. And what it then says is that it was the ojhā who identified a witch in the community concerned.

It is on page 11. Delete the fieldwork clause, or raise the scoping words to the head of the sentence, and a finding attributed to fieldwork reads as a flat statement of how the thing works. It is not reproduced here as a quotation at all.

And it does not sit in a findings chapter. It is not a statement about the 102 Odisha files. It sits in Chapter II of the report, “Review of Literature”, under the sub-heading “Definitions”, as the answer to a rhetorical question the report has just put to itself about whether there is any justified belief in the existence of witches. The fieldwork it refers to is ActionAid’s own ongoing five-state field study, in Assam, Jharkhand, Madhya Pradesh, Meghalaya and Odisha, which the report’s foreword describes separately. A remark in a literature review cannot carry the weight of a case-file finding.

Read on and the report supplies the same point in its own voice, without the borrowing. Two paragraphs further down, over the page break, the report says that the person who gets identified is, on the basis of tensions related to land and property or to jealousies, an unsupported single old woman. It then describes what follows the identification in terms this chapter will list once and not elaborate.

The same page also carries the report’s definition of the role, taken from a focus group in Mayurbhanj on 18 October 2020. The ojhās are those individuals supposed to have some magical power or healing capacity; people believe they can save others from witchcraft; and the ojhās are, in the report’s word, sometimes responsible for branding an individual as a witch. Sometimes. The report is more careful than any one sentence lifted out of it.

Further on, in the same literature review, the report summarises Govind Kelkar and Dev Nathan’s study of 110 case studies across five states, and reports their finding that in 50 out of 56 cases the accusers were close relatives. That is a third source, inside the document this chapter treats as its principal primary, saying the same thing about who does the accusing.

The ojhā is the ritual specialist. Chapter 19 introduced him under his own name, alongside the baiga and the gunia, and established the structure that governs him: the healer’s power and the witch’s power are held across this literature to be the same power, and the difference between them is which way it is pointed and who is entitled to hold it. He is the formal role. He is the named role. In the material this treatise reached he is male with great consistency, and the reader should note that the one statutory definition taking the word itself as its headword, printed further down this chapter, does not confine the role to men in what it defines, although the single pronoun the drafter reached for is masculine, and that the report’s 295 pages never gender the term either.

The person he identifies is typically a woman between forty and sixty.

Now state the argument of this chapter plainly, because it is the reason the chapter exists, and because it is an argument that is easy to state too strongly.

The harm attaches to a ritual-specialist role and to an accusation. It does not attach to a textual corpus.

A book about a corpus has to be able to say when the corpus is not the problem. That is not offered as a defence of the corpus. It is a statement about where the evidence sits. No text is required for what happens in these files. No manuscript, no printed grimoire from the Bombay and Banaras circuit of chapter 17, no vernacular incantation of any kind is necessary to the mechanism, and none is named in the documentation.

Who does the naming, and what the naming does

There are three ways to read this record wrongly, and all three run in the direction that costs most.

The first is to say that what kills people here is a man with standing in a village naming a woman, and to compress that into four words: the naming is the mechanism. That makes the ojhā the author of the accusation.

Madhu Mehra and Anuja Agrawal, writing in the Economic and Political Weekly LI.13 of 26 March 2016 on the Partners for Law in Development study, report the sequence differently. Discussing the 48 case studies collected in Bilaspur, Janjgir-Champa, Jamui and Ranchi, they find that although the ojhā appears in a majority of cases, the tensions and conflicts in many of them exist before he enters, which suggests to them that he tends to confirm an accusation rather than to initiate it. Land, property, jealousy, sexual advances and the ordinary sources of friction between social intimates were present in a large number of the cases. A very large segment of the perpetrators, on their account, are related to the victim by marriage.

Confirm rather than initiate. The second way to go wrong is to attach that finding to the wrong body of evidence. It does not come from the 85 police records, which are drawn from a different district set and are used in the article for registration patterns. It belongs to the 48 case studies, and one word of the citation decides whether it survives checking.

The third is the most flattering to whoever makes it. Set the two sources side by side, find them apparently at odds, stage the disagreement as a showpiece of scholarly honesty, and observe that neither of them says the two accounts are the same sequence seen from different ends. The report says it. It says so two paragraphs below the sentence usually taken from it, over the page break at page 12, in the terms Mehra and Agrawal use, and it says so again in its own literature review when it summarises Kelkar and Nathan on kin as accusers. The gap is made by stopping at one sentence and not reading on to the one that answers it.

What the three sources jointly support is not a contest. It is a convergence. The ojhā’s identification lands on a woman already at the centre of a conflict over land, property or jealousy, and in the majority of documented cases the accusers are her own kin. From that the narrower and defensible sentence follows: the naming is the moment at which a private conflict becomes a public licence. The specialist does not manufacture the grievance. He converts it into a verdict the village can act on, and the acting on it is what the case files record.

One complication belongs here before the picture sets. Helen Macdonald, in her 2004 SOAS doctoral work on Chhattisgarh, registers the underlying logic, crediting it to Nadel, Mair and Nitibaskara: those who protect society against illness are often assumed to possess the same power as witches. That is the structure chapter 19 built. Her own table breaks down sixteen female accusers, and Mehra and Agrawal report the same thing in their own data, finding evidence of women as instigators of witch-hunting while noting that men outnumber them in that respect. Any account of this record that presents the accuser as uniformly male is tidier than the evidence.

And the victims are not uniformly female either, which this chapter’s own principal source establishes. Of the 88 victims named in the first information reports, 75 were female and 13 were male. Of the 73 victims in the appellate judgments, 48 were female and 25 were male, which is more than a third. Taking the police records and the reported judgments together, the authors put the primary targets of witch-hunting at eighty-six per cent women. Men, they add, form a considerable bulk of the secondary victims, the collateral of an attack aimed at a woman. Eighty-six per cent is a strong gendering and it is not a near-absolute.

So the two agreed points in this literature are these. The primary victims are overwhelmingly women. Most of them are between forty and sixty. Two points, agreed across the disputing studies, and the first is the more important of them.

The number, and why there is no single number

The scale is large and the sources will not give a clean figure. Reporting it cleanly would be the easiest thing in this chapter and it would also be false.

One figure can be sourced to a named table through a named report, and it is the one this chapter carries: 2,468 murders committed between 2001 and 2016 in which witch-persecution was recorded as the motive, National Crime Records Bureau data as stated by ActionAid and the Odisha State Commission for Women. Sixteen years.

The same report gives two further national figures of its own, on its own front pages, for two further periods, and reconciles none of the three. They are set out together below rather than averaged, or folded into a range, because they do not count the same thing across the same years.

Three national figures, three periods, one report, no reconciliation. The first is the one this chapter carries in its own sentences, because it is the one with a stated basis and a stated period. The other two are recorded here with theirs, and the second of the two carries an annual figure as well, which the report puts above 150 killed a year.

The Bureau’s own witchcraft-motive entries are not a series, and this chapter does not present them as one. Three things stop them being one, and all three have to sit with the numbers rather than arrive after them.

The first is a contradiction the treatise has to report rather than resolve. ActionAid states that from 2017 onwards the Bureau does not have specific data on killings due to witch-persecution. The Bureau’s own published tables say otherwise, and carry a witchcraft column in Crime in India Table 2A.2, “Motives of Murder”, in every one of those years. The motive continued to be published. What changed after 2019 was the layout of the table and the ordering of its columns. The report and the Bureau cannot both be right, and this treatise has opened both and does not know why they differ.

The second is that the motive tables record completed killings only. The documented outcome types in this literature are branding, torture, forced ingestion, parading, ostracism and expulsion. Those are the modal outcomes. The report’s own statement of the problem is that the counts of witch-killings do not show these other forms of persecution, and that it is only after a woman branded as a witch is killed that she finds a place in police records. The dead are the tail of the distribution, and the table counts the tail.

The third is that killings are re-coded. Where a land dispute is present in the file, and the studies find land present in a substantial minority of files, the case can be registered and counted as a land dispute. The witchcraft motive moves into another column. That is the report’s own caution as well.

With those three carried, and as annual entries rather than as a series, this is the year-by-year record of people killed with witchcraft recorded as the motive.

The 2018 state rows put Jharkhand at 18 of those killings and Odisha at 5. In the 2017 table casteism stands as the recorded motive in 56 murders in the same year.

Nothing in this research explains the movement across those entries as a decline in violence or as a rise in one, and nothing about the way they were collected would license either reading. They are seven annual counts of completed killings with a recorded motive, taken from tables whose layout changed inside the period. Anybody who draws a line through them is drawing it themselves.

What would close this: a source reconciling the ActionAid statement with the Bureau’s post-2016 tables, and a national count built on documented outcomes rather than on completed killings alone. Neither exists in the material this research reached.

Three details in this section rest on a source that can no longer be reached. The Bureau’s 2017 and 2018 volumes are no longer served from ncrb.gov.in, they were not found on archive.org, and the Wayback index interface was blocked from this environment. So the column positions given in the table for 2017 and 2018, the 2018 pending-clarification footnote naming the five states, and the 2018 state rows for Jharkhand and Odisha are not re-verifiable from the Bureau’s current web publication, and they should be read as that. The 2017 and 2018 totals themselves are corroborated independently by IndiaSpend, which reports the same tables and gives Jharkhand’s share of both years, and the 2017 casteism figure by Deccan Herald. The 2019, 2020, 2021 and 2022 figures were opened in the Bureau’s own volumes.

The true figure is unknown and it is higher than any of these. We do not know how much higher.

Seven districts, and a loop this treatise will not close

The ActionAid report states three times, in three separate places, that eighty-three per cent of the witch-hunting and witch-branding cases in Odisha are concentrated in seven districts, and it names them each time in the same order: Mayurbhanj, Sundargarh, Rayagada, Malkangiri, Keonjhar, Gajapati and Ganjam. The 102 case studies themselves were collected across twelve districts.

Two things have to be said about that figure before anything is built on it.

The first is that the report never sources it. There is no National Crime Records Bureau citation behind it, no state police series, no methodology. It is the report’s own assertion, repeated three times without an underlying dataset.

The second is that it is the report’s sampling criterion. At the first of those three appearances, the sentence that follows the eighty-three per cent is that the study therefore collected its data from those districts. The district distribution of the 102 files cannot then be offered as independent evidence for the eighty-three per cent, because the files were drawn to match it. The confounds the Saora inference below is tested against apply with the same force to the figure that inference runs on.

A third thing has to be said because a reader who goes to check will hit it. The launch-day newspaper report of this study, by Priya Ranjan Sahu in Down To Earth on 20 December 2021, gives the cluster as six districts and omits Rayagada from the list. The report’s own text, in all three places, gives seven and includes it. This chapter follows the report. Anyone checking the newspaper will find a different list, and the difference is Rayagada, which is the district on which the paragraph below turns.

Three of the seven are Rayagada, Gajapati and Ganjam. Those are the core Lanjia Saora districts, and chapter 14 spent its length in exactly that country. The treatise is built on a word that is a people’s name, and the harm concentrates in the districts where that people lives. A reader can see where that goes.

It does not go there, and this treatise is going to say so on the page rather than let the omission do the work.

No source in this research disaggregates the victims by tribe. Not the ActionAid study, not the Bureau, not the state commission, not the journalism. The district-level co-location is what the report asserts, on a figure it does not source and drew its sample by. The inference from district to ethnonym is documented by nobody, and there are ordinary reasons it might fail: these districts are ethnically mixed, poverty and health-system absence track district lines as well as community lines, and reporting density varies with where organisations keep field offices.

The Saora link here is geographic. It is not evidenced at the level of the ethnonym.

What would close it: a study disaggregating witch-hunting victims by community. This research found none. Anybody who converts the district data into a claim about the Saora is repeating, in a modern register, the operation chapter 2 attributes to Cunningham in 1871.

Which is why the loop stays open here, and the paragraph that would close it is not on this page.

The laws, and the six first information reports

There is no shortage of law.

Bihar legislated against witch-hunting in 1999, with the Prevention of Witch (Daain) Practices Act. Jharkhand followed in 2001 with the Prevention of Witch-hunting (Dayan Pratha) Act, Chhattisgarh in 2005 with the Tonahi Pratadna Nivaran Act, Odisha in 2013, Maharashtra in 2013, Rajasthan in 2015, Assam in 2015, and Karnataka in 2017, notified in 2020. Eight states with dedicated statutes, built over two decades, several of them drafted after sustained campaigning by people who knew the problem intimately.

Three qualifications belong with that list rather than after it. The Odisha Act was published in the state gazette on 25 February 2014, so it was enforceable from 2014 rather than from the year it was passed. The Assam Act of 2015 received presidential assent only in 2018. And the Karnataka statute of 2017 is an anti-black-magic law rather than a witch-hunting law specifically, which makes it the loosest member of the list.

Now the use, and here the striking number is the one that misleads.

Mehra and Agrawal report that of 85 first information reports drawn from Jamui in Bihar, Bilaspur in Chhattisgarh, and Gumla and Ranchi in Jharkhand for 2010, 2011 and 2012, six were registered solely on the basis of the special law: two from Jharkhand, four from Chhattisgarh, none from Bihar.

Six.

Six invites a structural finding: that the special statutes go unused at the police station where the complaint is first written down, because the officers taking the complaint do not know the special law exists. Nothing in the sources supports the second half of that. It would be a causal finding about the conduct of a named institution, resting on no researcher and no officer who can be named.

The first half is refuted by the same article. What the authors report is that almost all cases are registered under provisions of the Indian Penal Code together with one or more provisions of the special law, the special law being invoked to establish the motive of the crime. The figure six is the count of registrations on the special law alone. The authors’ stated explanation for the Penal Code’s dominance is that it is the better instrument, because unlike the special laws it responds to threats, slurs, defamation, trespass, intimidation and harassment. Where they locate the failure is elsewhere: in the institutional apathy of the law enforcement machinery rather than in the absence of adequate provisions.

Odisha is the harder case, and it goes the other way. The ActionAid report states, under the heading “Gaps in the Laws,” that the Odisha Prevention of Witch Hunting Act was not at all invoked in any of the cases, and that Penal Code sections were used instead. Zero invocations across 102 case files. In some of them the victims did not receive a copy of the first information report.

Those two findings are what the record supports, and they are different findings. In Odisha’s 102 files the special Act was never reached for at all. In the three-state police data the special law is almost always invoked alongside the Penal Code and almost never alone. Both come from documents that were opened.

Maharashtra’s own usage figures were not closed. The available reporting, in ThePrint of 10 January 2024, carries two counts of registrations under the Maharashtra Act that differ from each other by a factor of four, one from an unnamed departmental official and one from an activist, in the same article, unreconciled. This chapter prints neither. What would close it: a return under the Right to Information Act, or the state’s own published figures.

One further thing has to be said about the article this section leans on hardest. Mehra and Agrawal’s piece exists to argue that a special law is the wrong instrument. Their concluding section reasons that confining these offences within the framework of a special law would be counterproductive, because prosecution would then be contingent on establishing the motive of witch-hunting and would exclude similar victimisation perpetrated for other reasons, and it argues for naming the offences inside the general penal code instead. It also argues that the police response should make us sceptical of greater reliance on criminalisation at all. Borrowing their data to support the opposite position, that the laws are adequate and enforcement fails, would be using a study against its own conclusion.

This treatise takes no position on what any legislature or campaign should do next. That is not its competence. The authors’ conclusion is stated in their own terms above and left standing as theirs.

The word in the statute book

Several of these statutes reach past the hunt and into the specialist’s practice, and the Chhattisgarh Act does something a book about this vocabulary has to stop at.

Section 2(3) of the Tonahi Pratadna Nivaran Act, 2005, defines the word ojhā in Indian statute law:

“‘Ojha’ may he be known by any other name whatsoever, means person who claims to possess power to control, cure, treat Tonahi or any person or animal or living things alleged to be affected by Tonahi and make him powerless, by jharphook, totka, tantra-mantra or by any means.”

The clause in the middle of that is doing more work than it looks. Drop may he be known by any other name whatsoever and the definition opens straight from the defined word into “means person who claims”, and two things go with it. Those words tell you the legislature is defining a function rather than a title, so that a baiga, a gunia, a disari or anyone else doing the work under a local name is inside the section. And they carry the only word in the definition that genders the ojhā. The section’s other pronoun, in make him powerless, attaches to the person or animal alleged to be affected. Which is why this chapter says the definition does not confine the role to men in what it defines, and does not say flatly that it is gender-neutral: the Act does not restrict the role to men, and it also does not write it as though a woman were expected. That is the second time in this chapter that a subordinate clause has turned out to be carrying the argument, and the repetition is itself the finding.

A book about a vernacular mantra vocabulary has found jhār-phūṅk, ṭoṭkā and tantra-mantra written into a legislative definition, alongside a defined tonahi and a defined identifier.

Section 6 then provides:

“Whoever does any act of jhar-phook, totka, use of tantra-mantra or anything as Ojha upon any person indicated as a Tonahi or any other person or animal or living thing alleged to be affected by such Tonahi, under any claim of treatment or control shall be punished with rigorous imprisonment for a term which may extend to 5 years and also with fine.”

Delete the object clause from section 6 and the truncated version reads as though these statutes criminalise the practice itself. They do not, and the point is worth pressing, because in a book about mantra vocabulary the false reading becomes an assertion that Indian criminal law prohibits the use of that vocabulary. Section 6 bites where the act is done on a person already named as a tonahi, or on a person, animal or living thing alleged to be witch-affected. Ordinary jhār-phūṅk is untouched by it.

Odisha’s 2013 Act is drawn on the same principle by a different route. Per the report’s summary, it punishes whoever performs any practice as a witch doctor knowing that such practices will cause harm or injury to the person on whom they are performed, with imprisonment of not less than one year and up to three.

So the true statement is narrower than the false one, and it is more interesting. In the two statutes this chapter has read closely, Chhattisgarh’s section 6 and Odisha’s provision as the report summarises it, the specialist’s act is criminalised only where it is performed on someone already named as a witch, or already alleged to be witch-afflicted, or with knowledge that it will cause harm. Both of those legislatures drew the line at the point where the practice enters an accusation, which is a legislative judgement about exactly the thing this chapter has been arguing about, and the subordinate clause is where it sits.

That finding covers two statutes. Two statutes are not eight, and one of the other six refutes the generalisation. The Rajasthan Prevention of Witch-Hunting Act, 2015, is the counter-case. Its section 5 punishes the practice of witchcraft with intent to cause harm, which the line above covers. Its section 6(i) does not: it punishes whoever claims that he has supernatural or magical power to control or cure a witch, with rigorous imprisonment of one to three years and fine. No named person. No prior accusation. No knowledge of harm. In Rajasthan the bare claim is the offence.

The eight states have not drawn one line. Two of them drew it at the accusation. At least one drew it earlier, at the claim itself. There is no tidier version of this paragraph that the statutes will support.

Behind all of them sits an older and stranger statute. The Drugs and Magic Remedies (Objectionable Advertisements) Act of 1954, at section 2(c), defines a magic remedy so as to expressly include “a talisman, mantra, kavacha, and any other charm of any kind which is alleged to possess miraculous powers for or in the diagnosis, cure, mitigation, treatment or prevention of any disease…”

Read that again with this treatise’s subject in mind. The Indian statute book names mantra. It names the kavacha, the armour-charm. It has named them since 1954, which is more than seventy years.

It is barely used. In Janhit Manch and others v. Union of India and others, a public interest litigation brought by that organisation’s convener Bhagwanji Raiyani, a division bench of the Bombay High Court dismissed the petition on 3 February 2011, accepting the Union Government’s position that the Act does not cover astrology and allied sciences such as palmistry and vāstu śāstra. The government’s affidavit described astrology as a time-tested science more than four thousand years old, and the court took that view.

The effect is that a very large part of the trade that actually operates sits outside the Act’s reach.

Six days in August

One date belongs in this chapter because the record turns on it.

Narendra Dabholkar was shot dead in Pune on the morning of 20 August 2013. He had spent decades campaigning against what he called superstition, and the anti-superstition bill he had pushed for had been stalled in Maharashtra for years.

The ordinance was promulgated on 26 August 2013.

Six days.

On 10 May 2024 the trial court delivered its verdict. Sachin Andure and Sharad Kalaskar were convicted of the killing and sentenced to life imprisonment. Virendra Tawde, Sanjeev Punalekar and Vikram Bhave were acquitted.

The matter did not end there. The appeal position can be brought up to date and it cannot be closed. Dabholkar’s daughter Mukta Dabholkar and the family have appealed the three acquittals in the Bombay High Court. Andure and Kalaskar have filed cross-appeals against their convictions. On 29 April 2026 the Bombay High Court granted bail to Sharad Kalaskar, a life convict, pending the hearing of those appeals. On 18 August 2026, the day this chapter was published, the same court granted bail to Sachin Andure and suspended his life sentence.

This passage records a live matter and its status changes. Everything above is the position as it stood on 18 August 2026. Two appeals against acquittal, two cross-appeals against conviction and two bail orders are all open before the Bombay High Court, and a reader arriving here later should check the current position rather than take it from this page.

That belongs in a section whose argument is about what the law does after it is passed. The ordinance took six days. The trial took eleven years. The appeals are unheard, and both of the men convicted of the killing are out on bail.

The sequence is what the ledger records. A law that could not pass in years passed in a week, after a killing. Draw from that what you will; this book will not draw it for you.

What the record disagrees about

Three things are genuinely contested in this literature, and flattening them into a single profile of the victim would be the ordinary way to write this section and the wrong one.

Marital status. The Partners for Law in Development report, Contemporary Practices of Witch Hunting: A Report on Social Trends and the Interface with Law, New Delhi, 2014, works from 48 case studies collected in Bihar, Jharkhand and Chhattisgarh, sixteen from each. Its table of marital status covers all 48: married 36, widowed 10, separated 2, never married 0. Its table of ages puts 35 of those 48 women between forty and sixty. Both are in the report’s own tables, at the pages given in the sources note.

Against that, the ActionAid Odisha foreword states that single women who were widowed or separated were found to be the most vulnerable group, and Kelkar and Nathan’s book is summarised in the same report as putting the attempt to capture women’s land and property near the centre of the phenomenon.

They do not agree, and this treatise does not choose between them. What has to be printed alongside the disagreement is the warning that the Partners for Law in Development authors put in their own article: contrary to the received wisdom, widows and single women were not the only ones vulnerable, and their sample is not large enough to establish that married women are always more vulnerable either. A chapter that quotes their table and drops that sentence would be publishing against its own evidence.

Land. For Kelkar and Nathan the accusation operates in part as a mechanism for dispossessing a woman with a claim to property, and the ActionAid literature review summarises them that way with cases attached. The counter-evidence is easy to state backwards, and the direction of it changes the shape of the disagreement. Partners for Law in Development did not find their victims near-landless. What Mehra and Agrawal report is the absence of a finding: the economic condition of the victims does not appear to have a significant bearing on vulnerability, the households ranged from adequate to marginal, some with land and some without, with varying access to animals, vehicles and built houses. The full report’s own landholding tables point the same way and harder: fifteen of sixteen Bihar families owned land, all sixteen Jharkhand families owned land, and five of sixteen Chhattisgarh families had none. Landholding varied by state and did not sort victims from anybody else.

The contrast with Kelkar and Nathan survives, and it is a different contrast: one body of work puts land near the centre, the other finds it does not distinguish the victims. The Odisha data satisfies neither. In the ActionAid figures, land grabbing does not appear on its own at all: 24.5 per cent of cases are attributed to a combined category of misfortune or land grabbing, which cannot be split, and 5 per cent to crop failure. Twenty-seven per cent of cases followed health issues in children and 43.5 per cent health issues in an adult family member, so illness of one kind or another stands behind roughly seventy per cent of the Odisha cases. That breakdown is on page xvi, and it specifies whose illness is at issue in each case and splits the two.

Caste. Mehra and Agrawal report that no significant inter-caste or inter-community dynamics could be discerned in their case studies, with the perpetrators proximate to the victim and belonging to comparable social and economic strata, so that position in the social hierarchy does not appear to shape vulnerability directly. Stop the sentence there and the truncated version presses against a caste-atrocity reading of witch-hunting. The clause that goes missing is the authors’ own qualifier: although this may be so in regions where such dynamics play a significant part in everyday relations. They leave the caste frame open for other regions expressly. And in their concluding section they go further, observing that the violence associated with caste atrocities is often similar to the cruelty perpetrated in witch-hunting, and that the two forms may coincide in some instances, even though few such cases appeared in their own material.

So the honest statement is that their data did not find a caste dynamic and their text declines to generalise the absence. That is not difficult to reconcile with a caste-atrocity frame. The authors reconcile it themselves, in the clause set out above.

The book behind the land argument has to be named exactly, because it is a book that gets cited loosely. It is Govind Kelkar and Dev Nathan, Witch Hunts: Culture, Patriarchy and Structural Transformation, Cambridge University Press, cited in the ActionAid report’s text as 2020 and in its bibliography as 2021, and summarised in its literature review. There is no separate Kelkar of 2022. This treatise reaches the book only through that summary, and no reading of it is claimed beyond what the summary carries.

Three live disagreements among serious researchers working on overlapping material.

No study reconciling them exists in this research, and none of these figures comes from a critically compared national dataset. What would close it: a comparative reanalysis across the Partners for Law in Development, Kelkar and Nathan, and Odisha case files on common definitions. Note also an inconsistency inside the Odisha report itself, which gives the case total as 100 where it breaks the cases down by cause and as 102 elsewhere. This chapter uses 102, which is the figure the report uses when describing its own method, and records the discrepancy rather than resolving it.

Two Indian studies that would sit directly in this section were not opened: Soma Chaudhuri’s Witches, Tea Plantations, and Lives of Migrant Laborers in India, Lexington Books, 2013, and Shashank Shekhar Sinha’s Economic and Political Weekly commentary of 12 May 2007 on witch-hunts, Adivasis and the 1857 uprising in Chhotanagpur. The second marks a structural gap rather than a citation gap: this chapter treats witch-hunting as a phenomenon documented from 1999 onwards, and there is a colonial and pre-colonial record it has not touched.

Where the record diverges, this treatise says so, and it adjudicates nothing.

A resonance, and what it is not

Chapter 10 established, by counting rather than by assertion, that the surviving corpus works in two paired moves: duhāī, the outcry made in the name of a higher authority, and ān, the binding invocation of that same authority, which the corpus itself names alongside it. Chapter 10 has the corrected counts. What they show is a body of speech in the vocabulary of people with no institutional standing, calling on names that are not theirs.

Set that beside the ledger.

The women in these 102 files, and in the wider count nobody can close, were named in a proceeding they had no standing in, and the naming was ratified by a man who did. The eight statutes written to protect them were reached for rarely in three states and not at all in Odisha’s files. Those are findings from separate sections above, and they are the whole of what this chapter supports.

Now be exact about the pairing, because this chapter exists to test a connection between a linguistic history and a social one, and that connection can be drawn wrongly in a single sentence.

It is a resonance. It is not a causal chain. Nothing in this research shows that the corpus produced these accusations, or that the ojhā who named a witch was working from a text, or that any printed page of the kind chapter 17 describes was ever spoken over any of these women. What is true is smaller and duller than the sentence a reader is braced for: a book about a word imposed from outside has followed its subject into the present, and found in the same country a documented record of people destroyed by a word imposed from outside. The two things sit next to each other. This chapter reports that they do and asserts nothing whatever about why.

That is a resonance. This book says so, and stops there.

And the women in this record are not an illustration of a thesis. They were branded, tortured, force-fed, paraded, expelled. Those are categories from the documentation and this treatise prints them as categories, without elaboration, because elaborating them would be writing about the deaths rather than recording them.

What the ledger hands on

Chapter 22 asks whether a tradition can be honoured under a slur, and who has standing to decide.

The word standing is the ledger’s contribution to that question, and it is not a metaphor here. The ojhā had standing, and what his standing did was turn a quarrel other people had already started into a verdict. The statutes had standing and the files show them going largely unreached. The accused had none, and the record of what followed runs to 2,468 recorded murders in the sixteen years to 2016, with the real number unknown and higher.

What this ledger records is a role, an accusation, and eight statutes whose use the record cannot account for.

It records nothing at all about a book.