This page explains exactly how pages on this site are made, checked and changed. It is written so that you can judge the work rather than trust it.
What counts as a source
We work outward from primary records. In order of preference:
- Official documents. The US Army's own investigation into the airstrike, court-martial filings and transcripts, indictments, court judgments, and agency releases.
- Statements by the organisations involved. What Reuters said about its staff, what the military said at the time, what WikiLeaks said when it published.
- Contemporary reporting by named journalists at outlets with a published corrections process.
- First-hand accounts from people who were there, clearly labelled as one person's account rather than as established fact.
What is not a source
Blog posts, forum threads, video essays and social media are not sources here. They may be the reason we go and look at something, but they never carry a claim on their own.
How a page gets written
Every page follows the same route. We collect the records that touch the subject. We write the account from those records. Then we read the page back against them line by line, and anything we cannot point at gets cut. What survives is published with the list of records at the bottom.
Why plain language
Military and legal shorthand is defined in the glossary rather than assumed. Pages are written in plain language on purpose. A 14-year-old should be able to read any page on this site and follow it. Simple wording is not a shortcut around detail — the detail is all there — it just means the reader spends their attention on the subject instead of the sentences.
How we handle disagreement
A lot of this story is contested. The central question — whether the aircrew could reasonably have believed they were looking at armed fighters — has been answered one way by the Army's investigation and another way by many of the people who watched the video. This site does not pretend that gap is not there.
Naming the disagreement
Where accounts conflict, the page names who says what and on what basis. Where the record simply does not say, the page says that too. "We do not know" is an acceptable sentence here and appears more than once.
Language we avoid
- We do not use the title Collateral Murder as a statement of fact. It is the name WikiLeaks gave the video, and pages say so.
- We do not call anyone a murderer or a terrorist in our own voice. Where those words appear, they are quoted and attributed.
- We avoid "allegedly" as a decoration. If a claim is disputed, the page says who disputes it.
- We do not use estimates as though they were counts. Where a number is an estimate, it is labelled as one and the body that produced it is named.
The people in this story
Several people in this record were killed, and several more were children at the time. We name the dead because the record names them and because being named is part of not being a statistic. We do not publish addresses, family contact details or medical detail beyond what is already in the public record and necessary to understand what happened.
Who gets named
Soldiers who have been identified in court documents or who have spoken publicly are named. Personnel who have not been publicly identified are described by role, not guessed at.
How images are made
Every illustration on this site is generated artwork, made for the archive, and captioned as such where it appears. None of it depicts a real person's face or reproduces any frame of the footage. We think it matters more that a reader is never misled about what they are looking at than that a page looks like a news report.
How these pages are actually produced
We would rather tell you this plainly than have you work it out.
AI assistance and human review
Research, drafting and editing here are AI-assisted and human-directed. A person decides what each page should cover and which records it must rest on, AI tools help assemble and draft the prose, and a person then checks the result line by line against the sources before anything is published. Nothing goes up that a human has not read against the record.
We think that combination is defensible and we also think it puts an extra duty on us, because a fluent paragraph is not the same as a true one. So the checks on this site are deliberately mechanical as well as editorial. Every page must clear a written rule set before it publishes: a minimum length, a sourced claim for anything contestable, working internal links, no invented statistics, no invented quotations, and no page that exists only to catch a search phrase. Pages that fail are fixed or not published.
What the pictures are
Every illustration on this site is AI-generated artwork, made for the archive and captioned as such wherever it appears. None of it depicts a real person, a real face, or any frame of the footage.
What we do not do is publish material nobody has read, generate pages in bulk to cover more search terms, or dress up a summary of someone else's article as our own reporting. If you ever think a page here reads as though it was produced without a person paying attention, tell us, because that is a failure by our own standard and not just yours.
Reviewing and updating
Pages carry a publication date and a review date. We revisit a page when the record changes — a court ruling, a new release of documents, a death — and when a reader tells us something is wrong. Substantive changes are logged on the corrections page rather than made silently.
What we will not do for traffic
This archive is built to be useful for years, not to chase whatever is being searched for this week. In practice that means a few hard rules: no pages written to catch a search phrase that we cannot actually answer, no bulk-generated filler, no rewriting of other people's articles, no paid links in or out, and no page that exists only to point at another page. If a subject does not have enough record behind it to make a real page, we do not make one. The about page says who is behind this and the corrections page is where mistakes get fixed in public.