Verifying an Agent-Generated Page
Verification metadata is the agent's confidence signal. A page marked
`authoritative` was vetted recently by a trusted author; `provisional`
means AI-generated or unverified; `stale` means verified once but more
than 90 days ago.
When to use this runbook
After spot-checking a page, when the act of marking it verified is the
next step.
Context
`mark_page_verified` writes `verified_at = NOW()` and
`verified_by = <exchange author>` into the page's frontmatter. The
structural-index post-save event syncs the values to the
`page_verification` table, where `ConfidenceComputer` reads them along
with the `trusted_authors` registry to compute the effective
`confidence`. Trusted-author + recent verification = `authoritative`.
Either condition missing = `provisional`. > 90 days old = `stale`.
Walkthrough
The frontmatter `steps` are the canonical sequence. The trusted-author
check (step 2) is the most-overlooked: a verification call from a
non-trusted author still lands the timestamp but confidence stays
provisional.
Pitfalls
The frontmatter `pitfalls` capture the failure modes. Mass-verifying
without reading is the most damaging — once the signal is poisoned,
agents stop trusting it.