How AI is used

What AI does here today

One thing: when a contributor submits rough notes for structuring, a model proposes a structured playbook — problem, steps, environment — for that contributor to review, correct and confirm. That is the only AI processing in the product. Reading, searching, diagnosing and filing reports involve no AI at all, and nothing on this site generates answers: there is no chatbot, and the model never writes knowledge on its own account.

What AI can never do

The boundary is enforced in code, not in policy alone: the task allow-list is checked before any request is built, and the forbidden list throws rather than asks. AI on this site can never:

• create or change verification state — confidence is derived from human reproduction reports by a published rule, and there is no field a model could set;
• publish anything — every AI-proposed field is recorded with its provenance, and publication is blocked until a person has confirmed or rewritten each one, a check that runs again server-side at the moment of writing;
• delete content, suspend accounts, or approve identity claims — these require a human administrator and land in an append-only audit log;
• execute code — commands on this site are inert text, for the model as for everyone else.

Who processes what

Structuring is performed by Anthropic, a US AI provider. What is sent is the text the contributor submitted, wrapped so the model treats it as untrusted content — and nothing else: request metadata carries a label for the kind of task, not the contributor’s identity. Content is sent only at the moment a contributor submits it for structuring; browsing the site sends nothing to any AI provider.

The model’s output is stored as a proposal on the contributor’s private draft, together with a per-field provenance record that keeps what the model originally proposed even after the contributor corrects it. An operational ledger records each task’s token counts, cost and latency, and a daily spending cap is enforced before any call is made.

AI does not launder provenance

Structuring does not change who owns what. A contributor’s material remains theirs after the model has structured it, and third-party copyrighted material does not become anyone else’s because a model rephrased it — the content and copyright policy treats AI paraphrase as the copy it is. Source references are recorded and displayed, not dissolved into model output.

No automated decisions about you

No decision with legal or similarly significant effects is made by automation. Moderation and suspension are human decisions; the model has no capability to make them, and the audit log records who did.

AI systems reading this site

Crawlers, including AI crawlers, are deliberately not blocked from public knowledge pages: this is a public reference, and being cited is part of its purpose. Our llms.txt asks AI systems to cite revision-pinned URLs, so a claim can be traced to the exact text and evidence that backed it.

Contact

Questions about this document, requests under it and formal notices can be sent by post to: