Problem
The public verification path (docs/verify-this-review, /v1/public/eval-scores, /v1/public/stats, the fairness report) is only checked when a human walks it. Individual breaks get fixed when found (#9636, #9639, #9719), but nothing prevents the next regression: a route rename, an auth change, or an empty export can silently invalidate the published walkthrough until someone outside the project notices.
Goal
Every public verifiability claim is exercised automatically, as an anonymous outsider, on a schedule — and a break fails loudly before anyone external can find it.
Sub-issues
Acceptance
Nightly job green means: an anonymous user can complete every step of the published walkthrough end-to-end, and every published record's digest recomputes. A red run pings the maintainer before it pings anyone else.
Problem
The public verification path (docs/verify-this-review, /v1/public/eval-scores, /v1/public/stats, the fairness report) is only checked when a human walks it. Individual breaks get fixed when found (#9636, #9639, #9719), but nothing prevents the next regression: a route rename, an auth change, or an empty export can silently invalidate the published walkthrough until someone outside the project notices.
Goal
Every public verifiability claim is exercised automatically, as an anonymous outsider, on a schedule — and a break fails loudly before anyone external can find it.
Sub-issues
Acceptance
Nightly job green means: an anonymous user can complete every step of the published walkthrough end-to-end, and every published record's digest recomputes. A red run pings the maintainer before it pings anyone else.