How a Name Enters the Archive

Thousands of names now appear in The Iran Massacre.

But where do they come from?

How do we decide whether two similar names refer to the same person?

What happens when a family gives us information that does not appear in a published list?

What if an AI tool finds a possible victim?

And what happens when two sources disagree?

The answer is a process designed around one basic principle:

show what we know, show how we know it, and be clear about what we still do not know.

Step 1: A source records a name

Every record begins somewhere.

The archive brings together information from human-rights organisations, news organisations, families, witnesses, memorial projects and other sources.

Among the sources currently represented in the archive are Iran Human Rights, Iran International's Javidnaman project, Hengaw, HRANA, Human Rights Watch, the Center for Human Rights in Iran, IranWire and others.

But appearing in a source does not mean every detail is automatically accepted as fact.

A source might have a name but no age.

Another might have a town.

A third may use a different spelling.

A family may have a photograph or information that has never appeared in a published report.

The first job is therefore not simply to copy a list.

It is to work out who the record is actually about.

Step 2: Is this the same person?

Persian names do not always have one obvious English spelling.

محمد can appear as Mohammad, Mohamad, Muhammad or in other forms.

Compound surnames may be joined, separated or written with a hyphen.

Persian and English versions of the same name may look very different to a computer.

And names alone are not enough.

Two people can have exactly the same name.

That is why the archive also compares information such as town, age and date.

If two records clearly describe the same person, their sources can be brought together under one record.

If the match is uncertain, it is not simply guessed.

It waits for review.

Step 3: A person checks

Automation can help organise thousands of records.

It can flag similar names.

It can identify possible duplicates.

It can help standardise different spellings.

It can point an editor towards records that deserve another look.

But it does not make the final decision.

Nothing is published or merged without human review.

That rule matters particularly with duplicates.

Merging two different people would erase somebody from the archive.

Failing to recognise a duplicate could count one person twice.

When there is doubt, the archive is designed to preserve the uncertainty until somebody can examine it.

And a merge can be undone if later evidence shows that it was wrong.

Step 4: The evidence level is assigned

Not every name has the same amount of evidence behind it.

The archive therefore uses four evidence levels: A, B, C and D.

They describe the evidence currently attached to the record.

They do not describe the importance or worth of the person.

A — Two or more independent lists

A record reaches level A when two or more independent sources have named the person.

Multiple reports from the same organisation do not artificially create multiple independent confirmations.

B — One trusted list

Level B means a trusted human-rights organisation or news source has recorded the person.

There may be only one independent organisation behind the record, but it is a recognised source.

C — Family and witnesses

Sometimes the information comes from people closest to the victim.

A family may contact the archive directly.

Or an editor may be able to check reports from different people who knew what happened.

Those records can be classified as C.

Family testimony is therefore visible in the evidence system rather than being silently mixed with institutional reporting.

D — Not yet verified

D is deliberately different.

A name might come from one social-media post.

A community list.

An AI-generated lead.

Or another record that has not yet been independently verified.

Instead of pretending that evidence is stronger than it is, the archive labels the record clearly.

D means: we have a reason to look, but we do not yet have enough evidence to call this confirmed.

Why publish an unconfirmed name at all?

This is one of the hardest questions in building an archive like this.

Why not simply hide every D record until it is confirmed?

Because sometimes making the name visible is how confirmation becomes possible.

The January crackdown took place during an extraordinary information blackout.

Some deaths were not reported by major organisations at the time.

Videos disappeared.

Families were under pressure.

Local information did not always reach international human-rights organisations.

A person searching the archive may recognise a name.

They may have a photograph.

They may know the town.

They may know the family.

They may be able to point to another report.

So D records are visible — but visibly marked as needing verification.

If evidence arrives, the record can move.

If the information turns out to be wrong, or the person is alive, the record can be removed.

Uncertainty is not hidden.

It is part of the record.

How a record can move from D to A

Evidence can strengthen over time.

A name may first appear as an unverified lead.

A family then confirms the person or an editor verifies reports from different witnesses.

That can move the record to C.

Later, a recognised rights organisation or news organisation may publish the name.

The record can then move to B.

If another independent organisation also documents the person, it can reach A.

This is why the evidence level is not a permanent label.

It describes what the archive can support now.

When the evidence changes, the level can change with it.

AI can find a lead. It cannot create evidence.

AI tools are useful in an archive containing thousands of names in two scripts.

They can help find spelling variations.

They can flag records that may describe the same person.

They can help identify patterns in large lists.

They can also surface names that deserve investigation.

But an AI-generated list is not evidence.

If a name comes only from a list assembled using AI tools, it remains a lead.

It does not become confirmed because an algorithm found it.

It cannot move up the evidence levels until recognised evidence supports it.

And AI does not publish changes to the archive.

An editor makes that decision.

The government's list is not confirmation either

The Iranian government has also published its own list.

The archive compares that list against its records.

But the government's list is treated only as a cross-check.

If the same name appears there, that fact can be recorded.

It does not independently confirm the person.

It does not raise the evidence level.

This distinction matters because a source connected to the state accused of carrying out the crackdown cannot simply be treated as independent verification of the events being documented.

Why we don't add different organisations' totals together

Different organisations' lists overlap heavily.

One victim may appear in several archives and reports.

If we simply added the headline totals from every organisation together, many people would be counted repeatedly.

So The Iran Massacre works at the level of the individual person.

Name by name.

Record by record.

Source by source.

When several organisations document the same person, those sources strengthen one record.

They do not create several victims.

This is also why the archive does not average different organisations' death tolls.

Their figures remain attributed to the organisations that published them.

The number on The Iran Massacre means something different:

the number of individual names currently recorded in this archive.

What happens when sources disagree?

Real-world evidence is messy.

One source may say somebody was 22.

Another says 23.

One may name one town and another a nearby town.

Dates can differ.

Names can be transliterated differently.

The archive does not resolve every disagreement by choosing whichever answer seems most convenient.

Where conflicting information cannot yet be resolved, the disagreement can remain visible.

That is preferable to inventing certainty.

Corrections are part of the archive

An archive of this size will contain errors.

The important question is what happens when they are found.

Names may need spelling corrections.

Locations may change.

New sources may arrive.

Two records may eventually be identified as one person.

Families may correct information.

The archive keeps a public corrections log showing how many merges and corrections have been made.

That is not an admission that evidence should never change.

It is recognition that responsible documentation includes correcting the record when better information arrives.

Then the record becomes public

At the end of this process is the page a visitor sees.

A name.

Persian and English spellings where available.

A photograph, if one can be safely and reliably sourced.

Age and location where known.

Evidence level.

And the sources behind the record.

The final part is essential.

Readers should not have to take our word for it.

Where possible, they should be able to follow the evidence themselves.

This archive is designed to be questioned

The purpose of a sourced archive is not to say:

Trust us.

It is to say:

Here is what we recorded. Here is where it came from. Here is how strong the evidence currently is. Check it.

Sometimes that means showing uncertainty.

Sometimes it means admitting we do not know.

Sometimes it means changing something we previously published.

And sometimes it means leaving a name marked D until another piece of evidence finally arrives.

That is not a weakness in the archive.

It is the point of building one.

Because behind every record is a person.

And getting that person's identity right matters more than making the database look certain.