Observed from the tribnest/vilajo integration (2026-09-10, confirmed by the reviewer):
A new edition was pushed for case 1abe0ad1b4bd with 37 new blobs — the window title and two steps' rendering genuinely changed. Yet GET /cases/{id}/captures shows every capture still accepted (case state accepted), nothing fell back to to-review.
This defeats the review loop's core guarantee: a capture whose content hash changed since the accepted baseline must return to to-review, otherwise regressions ship invisibly under a stale acceptance.
Expected: on edition publish, any capture whose blob hash differs from the last reviewed one flips to to-review (and the case aggregates accordingly). Steps added/removed should behave likewise.
Observed from the tribnest/vilajo integration (2026-09-10, confirmed by the reviewer):
A new edition was pushed for case
1abe0ad1b4bdwith 37 new blobs — the window title and two steps' rendering genuinely changed. YetGET /cases/{id}/capturesshows every capture stillaccepted(case stateaccepted), nothing fell back toto-review.This defeats the review loop's core guarantee: a capture whose content hash changed since the accepted baseline must return to
to-review, otherwise regressions ship invisibly under a stale acceptance.Expected: on edition publish, any capture whose blob hash differs from the last reviewed one flips to
to-review(and the case aggregates accordingly). Steps added/removed should behave likewise.