Methods

Why and where we use AI

Draft for public review, 2026-10-25. Not yet signed off by an editor. This page is a statement of intended practice, not proof that every item has been reviewed.

We use AI to make large amounts of public information easier to explore, not to replace evidence, expertise or human accountability. AI can be wrong, incomplete or overconfident.

Where AI may be used

1. Research discovery: search terms, document leads, duplicate finding. Leads are not findings.
2. Document processing: transcription help, extracting dates and amounts, link suggestions. We keep the original file.
3. Analysis: flagging possible contradictions or data revisions. A pattern an AI finds does not prove intent, wrongdoing or cause.
4. Writing and accessibility: plain-language drafts, alt text. AI-written prose is kept separate from quoted source text.
5. Software and graphics: help building maps, timelines and tests. Illustrations made by AI are labeled and never passed off as archival records.
6. Workflow: deciding what a person should check first, and keeping the correction log.

What AI does not decide

AI does not certify truth, declare anyone guilty, decide which beliefs are acceptable, or find legal violations. The Institute decides what is published, and a person checks each source against each claim before it is called checked.

Our promises about evidence

• We keep the exact source, the date and the version where the law allows.
• We separate primary records, news reports, later scholarship, AI-made guesses and unresolved allegations.
• Material claims point to records you can open, and show whether a person has reviewed them.
• Corrections stay visible with dates and reasons. We do not quietly rewrite history.

An honest limit

This is a policy and a plan. We report what is built and checked in the Development Ledger as it goes live.