Product / Audits
Every authorization and query-validation answer is written as a durable, citable record before it is returned, with a citable authorization or validation ID and the governing policy and publication provenance. Agents can ask why any serving, authorization, or validation decision holds, and each record reconstructs the decision as evaluated: open it to see the exact answer returned, and admins and owners can download the complete evidence bundle as JSON. Workspaces keep an append-only activity log and a per-token request log.
Audits
An agent gets an answer and someone asks why. Every authorization and query-validation answer is written to a durable ledger with a citable authorization or validation ID before it is returned, and cites the serving rows, policy versions, and publication behind it. Raw query and use text stay out of the ledger unless the workspace explicitly opts in.
An auditor asks what the agent actually saw. Open any authorization or validation record to see the exact answer returned, its actor snapshot — which survives token deletion — and the governing policy and publication. Admins and owners can download the complete evidence bundle as JSON, or export the filtered audit timeline as CSV. Validation evidence also carries server-recorded finding IDs and a privacy-safe query fingerprint. Raw query text stays absent unless the workspace opts in.
A decision needs justification. explain_why re-derives the explanation for any serving decision by its id — the rule, policy version, and publication that produce it.
Weeks later, someone asks what agents were doing. The per-token request log keeps 90 days of calls with allow-listed fields — query text is never recorded — alongside the append-only workspace activity trail.