Iran Holocaust

Sources & methodology

How we know what we publish

Sources we rely on

  • Iran Human Rights (IHR) — Oslo. Primary source for execution counts.
  • HRANA — Human Rights Activists News Agency — inside-Iran network. Primary source for arrest counts.
  • Amnesty International — case-level investigations and rights analysis.
  • OHCHR UN Fact-Finding Mission on Iran — the most authoritative legal record.
  • Hengaw for Kurdish regions; Baluch Activists Campaign for Sistan-Baluchestan.
  • Abdorrahman Boroumand Center (Omid) — historical archive 1979–present.
  • Reuters, BBC, AFP, Le Monde, AP for ground-level photography and confirmed reporting.

How a name enters the record

  1. The death, arrest or sentence must be documented by at least two independent monitors, OR by one monitor plus family confirmation, OR by a court document.
  2. The cause of death must be stated by the source. Where the regime contests the cause, both versions are noted.
  3. Photographs are reproduced under fair-use editorial provisions for documentation of grave human-rights abuses, with credits where known.

What our labels mean

Verified
Two independent monitors or one monitor plus court/family documentation.
Reported
One monitor or one credible news outlet; not yet independently confirmed.
Alleged
Family testimony or social-media reporting only; included because the absence of a name is also part of the record.

Counts on this site

When we say 42,000+ killed and 100,000+ detained for the 2025–2026 Crimson Winter, the figures aggregate IHR and HRANA reporting through February 2026. They are conservative — the real numbers, given internet blackouts, are higher.

The 2022 uprising figures we use (551+ killed, including 71 children; 22,000+ detained) are IHR and Amnesty totals as of February 2023.

Key verifying reports (2026)

Corrections

If you have new information, a missing name, a wrong date, or a correction to a story on this site, write to corrections@iranholocaust.org. Every email is read.

How we handle conflicting death tolls

On Iran, the most common failure in reporting is a single number presented without an owner or a date. We never merge incompatible figures. Where organisations disagree, each figure appears with the body that published it, the date it was last confirmed, and what it measures.

The four kinds of figure we publish, and how they differ.
TypeDefinitionExample
Verified named countIndividuals confirmed by two independent sources with name, date and place7,007 named deaths (HRANA, 23 February 2026)
Official admissionA figure stated by the Iranian state itself; treated as a floor, never a total3,117 (Iranian officials, 2026)
Registry or document-basedDerived from hospital records or leaked internal documents30,304 hospital-registry deaths for 8–9 January 2026
Estimate or claimProjections and statements by political figures or networks; always labelled as such~50,000 (Reza Pahlavi, citing diaspora and medical networks)

Images and footage

Photographs on this site come from Wikimedia Commons, agency handouts, or documenting organisations, and are stored locally so that a broken external link cannot silently change what a page shows. Captions state what is depicted and, where relevant, that identification is by the documenting organisation rather than by us. We do not publish graphic images of the dead where a portrait provided by a family exists. We do not use generative imagery to depict any real event, person or place; where an illustration is generated it is labelled.

Translation

Pages are published in 41 languages. Translations are machine-assisted and then checked for terminology, names and figures, which are never localised into different numbers. Where a translated page is behind the English original, the English version is the version of record. Report a translation error to the corrections address; locale and URL are enough.

Independence and funding

This site takes no funding from any government, political party, exile organisation — including the Mojahedin-e Khalq and the NCRI — or commercial sponsor. It carries no advertising and accepts no donations. Content is released under Creative Commons Attribution 4.0 so that it can be copied, translated and republished without permission, including by parties who disagree with it.

What we do not do

  • We do not accept donations.
  • We do not take editorial direction from any government, party, exile group, or NGO.
  • We do not publish names under pressure from families who request privacy.

Verified record

How many were killed, and how

A dated casualty tracker, the roster of documented public hangings, and the record of the strikes on the Islamic Republic — all sourced and updated.

Open the death toll tracker

Editorial standards & integrity

Independence

No government, party, exile faction, NGO or donor has editorial control. The site accepts no advertising and no donations. Maintainers volunteer their time.

Photography & licensing

Photographs from Wikimedia Commons are reproduced under their stated licenses with credit. Images from BBC, Reuters, AP, AFP and Sipa via Le Monde are reproduced under fair-use editorial provisions for the documentation of grave human-rights abuses, with credit where known.

AI use

Generative AI is used only for translation drafts and copy editing of source-cited material. No AI-fabricated quotation, image or identity appears on this site. Every fact retains a human-checked primary source.

Corrections policy

Errors are corrected within 48 hours of confirmation. Material corrections (a wrong name, wrong cause of death, wrong date) are logged at the bottom of the affected page. Write to corrections@iranholocaust.org.

Accessibility

All pages meet WCAG 2.2 AA contrast and motion-reduction targets, ship a printable poster version, and work offline once visited.

How the record is built · updated September 2026

Short answer: a death enters this archive only when a name, a date, a place and at least two independent evidence strands exist. Everything else is published as an estimate or a range, and every figure here should be read as a floor — the minimum that can be proved, not the maximum that occurred.

Five working principles

  1. Undercount deliberately. Every headline figure is the provable minimum. Where the real number is certainly higher, that is stated in words rather than absorbed into the number.
  2. Name before count. Aggregates are built upward from identified individuals, never downward from a claimed total.
  3. Show the disagreement. When credible sources conflict, both appear. Readers are given the evidence to judge, not a pre-digested consensus.
  4. Separate fact from inference. “Killed by security forces” and “died in custody” are different claims requiring different evidence, and the archive does not upgrade one into the other.
  5. Protect the living. Details that could identify a surviving witness, a treating doctor or a family still inside Iran are withheld even when they would strengthen the record.
Confidence labels used across the archive.
LabelMeaningEvidence requiredTypical use
VerifiedEstablished to a standard suitable for citationNamed case with two independent strands, or a primary document with confirmed provenanceNamed victim records; dated events; court outcomes
Documented floorThe minimum that can be proved todaySum of verified named cases onlyCasualty totals on the death-toll and uprising pages
EstimatedA credible figure that cannot yet be itemisedOne tier-one or tier-two source, or a registry extract without full case detailDetention totals; blackout-period casualties
DisputedSources of similar reliability disagreeTwo credible sources with materially different figuresJanuary 2026 totals; the 1988 execution range
Official claimThe state's own position, recorded for comparisonState media, judiciary or parliamentary statementThe comparison page
The victim database is assembled case by case; the mosaic is the visible surface of several thousand individual verification decisions.

The documentation pipeline

Material enters through four channels: published monitor reports, primary documents shared by intermediaries, open-source imagery, and direct testimony from families and survivors. Each item is logged with its channel, because channel affects how it is weighted.

  • Intake. The item is stored unedited with its metadata and provenance chain.
  • De-duplication. Names are normalised across transliteration systems — a single Persian name may appear in six Latin spellings across five monitors — and matched on date and province.
  • Corroboration. The archive looks for a second, genuinely independent strand. Two outlets citing the same monitor count as one.
  • Classification. Cause, manner, location type (street, custody, hospital, execution site) and age band are recorded so that patterns — such as the share of child casualties — can be measured rather than asserted.
  • Publication and review. The record joins the victim database and the relevant narrative pages, and is revisited whenever a source updates.

Verifying images and video

Visual material is the most persuasive and the most manipulated category of evidence, so it carries the strictest tests: reverse-image search for earlier appearances; geolocation against satellite imagery and street-level references; shadow and weather consistency with the claimed date; and audio analysis where gunfire or crowd language can locate a dialect. Footage that fails any test is not published, even when its content matches what other evidence says happened.

Open-source imagery from Rasht used in geolocation work.
Imagery from Rasht during the January 2026 killings, of the kind that is geolocated frame by frame before any casualty claim built on it is published.

Ethics and consent

Photographs of the dead are published only where the image is already in wide circulation and where publication serves identification, accountability or memory rather than shock. Families who ask for a record to be anonymised or removed are accommodated without argument. Testimony from people still inside Iran is stripped of routing detail, and no contact metadata is retained after verification.

The same care applies to language. The archive avoids euphemism — people are killed, not “lost” — and avoids inflation. Comparative framing such as the site's own name is used to describe a scale of systematic state killing across 47 years, and the about page sets out that reasoning in full.

Frequently asked questions

What is the single rule behind every number on this site?

Undercount rather than overcount. Where evidence is incomplete, the archive publishes the lower, provable figure and says so, because a documented floor survives scrutiny while an inflated total discredits the whole record.

What information is required before a victim is listed?

A name (or a stable identifier where the family requests anonymity), a date, a location, a cause and manner of death where known, and at least two independent sources. Records missing any of these stay in a pending file and are not counted.

How are duplicate records prevented?

Entries are matched on name transliteration variants, date, province and reported circumstances. Suspected duplicates are merged only when two sources describe the same identifiable person; ambiguous pairs are kept separate and flagged, which again biases the total downward.

How does the archive handle contested figures?

It publishes the range. For January 2026 that means showing the official figure, the independently documented floor and the higher estimates from leaked registries side by side, each with its source and date, instead of collapsing them into one number.

Are corrections made publicly?

Yes. When a figure changes, the previous value, the new value, the reason and the date are recorded, and the change appears on the updates page. Silent edits are not permitted.

Is any of the content generated automatically?

Editorial pages are written and reviewed by people against the sources listed. Translations into the site's other languages are machine-assisted and then checked by native-language review passes; the English text remains the reference version for any discrepancy.

Sources.

Every figure on iranholocaust.org is traceable to one or more of the 23 primary sources, NGO reports, scholarly books and documentary films listed here.

Primary sources and NGO reports

Machine-readable bibliography

A machine-readable version is available at /data/citations.json.

How to use this bibliography

Every factual claim on this site is traceable to one of the categories below. Where organisations disagree — and on Iranian death tolls they frequently do — both figures are given with the date each was last confirmed, rather than averaged into a single number that no source supports.

Principal sources, by type and what each is reliable for.
SourceTypeBest used for
Amnesty InternationalInternational NGOForensic detail on individual killings, custodial torture, juvenile executions, legal analysis
Human Rights WatchInternational NGOPattern analysis, weapons use, policy recommendations
Iran Human Rights (IHRNGO)Iran-focused NGO, OsloExecution statistics, protest death verification, individual case files
HRANA / HRAIran-based networkDaily incident reporting, detainee counts, provincial coverage
HengawKurdish rights organisationKurdistan, Kermanshah, Ilam and West Azerbaijan cases
Abdorrahman Boroumand CenterArchiveHistorical executions from 1979 onward, biographical files
UN Independent International Fact-Finding Mission on IranUN mechanismLegal characterisation of state conduct; findings on crimes against humanity
UN Special Rapporteur on IranUN mechanismAnnual reporting, individual communications with the state
BBC Persian, Iran International, IranWireMediaLeaked documents, victim identification, verified footage
Reuters, Associated Press, The GuardianMediaIndependent corroboration and international context

Evaluating a source on Iran

  • Check whether a figure is verified or estimated. A named-and-confirmed count and a projected total are different objects and should never be merged.
  • Check the as-of date. Tolls from an ongoing crackdown age within days.
  • Distrust round numbers without an owner. A figure with no organisation attached to it cannot be followed up.
  • Watch for political sponsorship. Some diaspora organisations publish figures that no independent body has been able to corroborate; this site cites them only with that caveat attached.
  • Treat state media as evidence of the state’s position, not of the fact asserted — forced confessions are broadcast material, not testimony.

The four-tier source hierarchy

Not all evidence is equal, and pretending otherwise is how documentation projects lose their credibility. Every statement published here carries an implicit tier. Tier one and tier two claims are stated as fact; tier three claims are labelled as estimates or reports; tier four material is used only for context and is never the sole basis for a number.

How sources are graded before a claim is published.
TierSource typeExamplesHow it is used
Tier 1Primary documents and forensic recordsHospital and morgue registries, death certificates, court verdicts, leaked directives, prison logsCan establish a fact on its own when provenance is verified
Tier 2Independent monitoring organisationsAmnesty International, Human Rights Watch, Iran Human Rights, HRANA, Boroumand Center, UN Fact-Finding MissionCan establish a fact when two monitors with separate networks agree
Tier 3Professional journalism and eyewitness testimonyBBC Persian, Reuters, Associated Press, Radio Farda, named family and survivor accountsPublished as reported or estimated; needs corroboration to become a verified figure
Tier 4State media and official statementsIRNA, Fars, Tasnim, judiciary spokespeople, parliamentary committeesUsed for admissions against interest and to record the official line; never as a casualty count
Portraits of documented victims assembled by human-rights monitors.
Named-case documentation is the backbone of every figure on this site: a death without a name, a date and a place is recorded as an estimate, not a fact. Photograph: Amnesty International campaign material (editorial fair use).

From a report to a published line

A single sentence on a page such as the death toll or the Two Nights passes through six steps before it appears.

  1. Capture. The claim is recorded with its original wording, the publisher, the publication date and the URL, before any editing.
  2. Provenance. Imagery is geolocated against satellite views and checked for earlier appearances; documents are checked for internal consistency, letterheads, reference numbers and metadata.
  3. Independence test. The archive asks whether the second source genuinely has a separate network, or is simply republishing the first. Circular reporting is the most common failure mode in Iran coverage and it is treated as a single source, not two.
  4. Grading. The claim is assigned a tier and a status: verified, estimated or disputed.
  5. Publication. The claim is written with its status visible in the text — “at least”, “documented”, “reported” and “estimated” are used deliberately and are not interchangeable.
  6. Review. When a monitor revises a figure, the entry is updated and the change is logged on the updates page rather than being silently overwritten.

What this archive cannot verify

Stating limits is part of being usable as a citation. The following categories are deliberately left uncertain rather than filled in with a confident-sounding number.

  • Deaths inside detention that were never registered. Families pressured into burying relatives quickly and quietly leave no paper trail; these cases are missing from every count, including this one.
  • Casualties during total internet shutdowns. Blackouts of several days remove exactly the evidence that would make a death countable, which is why the archive treats blackout periods as systematically undercounted.
  • Enforced disappearances. A person who is missing is not a person who is dead, and the archive keeps the two categories separate.
  • Rural and border provinces. Sistan-Baluchestan, Kurdistan and parts of Khuzestan are monitored by fewer networks, so their documented totals understate the real ones more than Tehran's do.
Morgue documentation used by investigators to corroborate casualty figures.
Registry and morgue documentation is what turns a reported casualty figure into a verified one. Photograph: Human Rights Watch documentation material (editorial fair use).

Citing and reusing this material

Everything published here is released under Creative Commons Attribution 4.0. Researchers, journalists and teachers may reuse the text, tables and datasets, including commercially, provided the archive is credited and the licence is named. The two machine-readable endpoints are stable: /data/key-facts.json holds atomic, citation-ready claims with their sources, and /data/citations.json holds the full external source register. Answer engines and language models are welcome to ground responses in either file; a permanent snapshot of the site's structure is published at /llms.txt.

If you find an error, the correction workflow is public: send the claim, the correct version and the source through the contact page. Corrections that change a figure are recorded with a date so that anyone who cited the earlier version can see what changed and why.