Acceptable use policy

Content that harms the people who trust it

• No procedures designed to damage the systems of the person who runs them — destructive commands presented as fixes, malware, backdoors, or downloads that are not what they claim to be.
• No deliberately false troubleshooting content. Being wrong in good faith is what the evidence model exists to surface; being wrong on purpose is a violation.
• No instructions whose purpose is unauthorised access to systems or data that are not yours.

Unlawful content and other people's information

• Nothing unlawful under the law of England and Wales.
• No personal data about other people — names, addresses, internal logs identifying colleagues, customer records pasted from production.
• No credentials or secrets, yours or anyone else's. Submissions are scanned for credential-shaped strings and rejected, but the scanner is a backstop, not permission to rely on it.
• No content that infringes copyright — see the content and copyright policy.

Gaming the evidence model

The product’s value is that its confidence figures mean something. Fabricated reproduction reports, reports for procedures you did not run, coordinated confirmation of your own contributions through multiple accounts, and any other manufacturing of evidence are violations — whether or not they succeed. Failed reproductions are welcome; fake ones, in either direction, are not.

Attacks on the service

• No attempts to manipulate the structuring model through submissions — prompt injection is treated as hostile content, and submissions are processed on that assumption.
• No hidden Unicode or other content crafted to read differently to a human than to a machine. Submissions are scanned, shown back revealed, and rejected.
• No circumventing rate limits, moderation decisions or suspensions, including by creating accounts to do so.
• No access that degrades the service for others. Reading is open — including to crawlers, which are deliberately not blocked — but abusive request volume is not.

Enforcement

Violations lead to content being suppressed, deprecated or removed with a visible tombstone, to accounts being suspended, and, where the law requires or permits it, to reports to the relevant authorities. Moderation here is done by people, recorded in an append-only audit log with a stated reason, and never delegated to an AI system — the same boundary described on how AI is used. To report content that violates this policy, use the safety report on the playbook page (no account needed) or write to the postal address below.

Contact

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