A global mapping of AI in transitional justice names Colombia 25 times and Iraq none
We counted the words in Impunity Watch's November 2025 mapping report. Iraq, Yazidi and Kurdish appear zero times in 141,651 characters, and being precise about what that does and does not prove is the whole point.
Impunity Watch
Global
20253 minutes
- Researcher or analyst
- Legal work and accountability
- Director or leadership
- Research
- Drafting reports and advocacy material
What happened
In November 2025 Impunity Watch published a mapping of how artificial intelligence and data-driven technologies are being used in transitional justice around the world.
We downloaded the PDF, 2,972,244 bytes, pulled the text out with pdftotext, and counted case-insensitive matches across 141,651 characters of it.
Iraq: 0. Yazidi: 0. Kurd, a substring that also catches Kurdistan and Kurdish: 0.
Colombia: 25. Syria: 20. Palestine: 19. Ukraine: 17. Tunisia: 0. Lebanon: 0.
That is the entire method. It takes about five minutes to repeat, and you should, rather than take our word for it.
What it means
A document whose stated purpose was to find AI use in transitional justice, worldwide, contains no mention of Iraq, no mention of the Yazidi genocide, and no mention of the Kurdish region at all. Syria appears twenty times and Palestine nineteen, so this is not a survey that skipped the Middle East. Tunisia and Lebanon are also at zero.
Our reading, not the report’s: a mapping can only find what someone has already written down in a form a researcher can reach, in a language they searched, published by an organisation already inside the networks they asked. Every one of those filters can produce this result without anyone making a decision about Iraq.
What it does not mean
It is not proof that nobody in Iraq is doing this work. It is proof that this survey did not find or include any of it. Those are different sentences and the difference is the reason this page exists: you may be the counter-example, and the report contains no footnote saying so.
That is the quiet problem with an absence. A gap in a survey has no marker. There is no line reading “we found nothing in Iraq”. There is just nothing, and nothing reads the same as not-applicable.
Why a word count is worth a page
Surveys are what the next person cites. Somebody writing a funding brief, a conference panel or a literature review on AI and accountability in this region next year will start from documents like this one, and what is not in them is not available to be cited. Invisibility compounds quietly: work nobody wrote down cannot be found, copied, funded, or usefully criticised.
What to do differently
If you are in Iraq or the Kurdistan Region and you have used any of this on documentation, archiving, translation, verification or case preparation, even briefly, even badly, then on the evidence of this report nobody has written it down. Writing it down is the task.
It does not have to be a report. One page is enough if it carries six things: what the material was, what you used on it, what it got wrong, how a person checked, what you would not do again, and the date. That is citable. A polished case study with no error in it is not.
Publish it in English as well as in Arabic or Kurdish. The researcher assembling the next mapping searches in English, and will not find your Sorani post.
Tools the sources name
The sources name no tool for this case. We have not guessed one, and neither should anyone repeating the story.
What we do not claim
Everything above is limited by what its sources actually prove. This is the part they do not.
This is a word count, not an assessment of the report. It does not prove that no AI work on transitional justice exists in Iraq or in the Kurdistan Region. It proves that this survey did not find or include any. Absence in one mapping is evidence of invisibility, not of non-existence, and the two are not close enough to substitute for each other.
The report gives no reason for the omission and we are not supplying one. Nothing here is a claim that Impunity Watch excluded Iraq deliberately.
The substring "kurd" also covers "Kurdistan" and "Kurdish". All three are zero.
The character total depends on the extraction tool. Ours is 141,651 characters, from pdftotext run against a 2,972,244 byte PDF. The word counts themselves are stable: repeat them and you get the same nine numbers.
We counted a PDF. We did not interview the authors, and we did not search for Iraqi work that the report might have missed. This page is one measurement and nothing more.