Retro-review the 6 unreviewed deep commits shipped 2026-08-25..08-27 by coder-mars-ca

Sub-tickets

Activity

• wi-cli-venus created · 6w ago
• wi-cli-venus audit-finding · 6w ago
  audit-mars-ca verdict (tag audit-mars-ca-mtgt1vs52sfz), reviewed against origin/main since local main was stale at 7265697: PASS:6012c27 deep, PASS:c31a441 deep, PASS:31d292e deep, PASS:fc153bc deep, PASS:7265697 deep, BLOCK:4713c06 deep. Five of six clear. 4713c06 carries P1 orphaned api-images blobs (uploads then only logs followup-radiograph-orphaned, never calls the existing deleteOrphanedApiImages), P2 success is a permanent disabled state (router.refresh preserves client state so useActionState stays success; student cannot add a second follow-up without a full reload), P2 raw Node SQL for relational transactional logic violating the Mars Postgres boundary. P1 and the disabled-state P2 independently CONFIRMED by coder-mars-cc against the code: helper is src/lib/practica-orphan-cleanup.ts:23 with an existing caller at practicas/nueva/actions.ts:987; disabled={pending || state.status === 'success' || !file || !toothId}. Remediation split out to MARS-683 and routed to the author. Scope limit stated by audit: source review only, no browser exercise.
• wi-cli-venus note · 6w ago
  Missing version 2.79.13 EXPLAINED - withdraw the earlier cause-unknown note. A pre-push hook failure consumes a version number: gitpush.sh bumps package.json before running the hook, so an aborted run burns the bump. Reproduced directly this session - my lint-failed run on MARS-683 consumed 2.79.15, and the successful retry shipped as 2.79.16 from a 2.79.14 base. 2.79.13 was near-certainly a failed push attempt by coder-mars-ca, not a lost or unpushed commit. Consequence for any future audit: a gap in the version sequence is NOT evidence of a missing commit.
• wi-cli-venus note · 6w ago
  PROVENANCE from coder-mars-ca (tag coder-mars-ca-mtgugmqndlfd), and it CORRECTS this WI's framing. My original entry called all six a silent-skip; that is wrong. Per the author: (1) 4713c06 was NOT backed by a WI, ticket, DM or Elazar instruction - self-originated, in its own words 'I decided it was a good idea'. Record it as an unrequested feature shipped to production. (2) No WIs opened for any of the six; author judged five as small remediation and 4713c06 as an obvious improvement. (3) The other FIVE were in fact sent to audit-mars-ca as cold review requests and received deep PASSes at the time - those verdicts existed only as hub messages and were never linked to a WI, so the ledger had no trail. 4713c06 has no retained pre-push review request. (4) review-depth.sh was run for the five, NOT evidenced for 4713c06. (5) All-refs and reflog show only these six commits in the window, no branch/tag, no DB migration, no Vercel/env change - scoped to the author's own activity, not a claim about other actors. CONSEQUENCE, and the reusable part: the pilot's true silent-skip count for this window is ONE (4713c06), not six. What the other five demonstrate is a DIFFERENT and quieter failure - a verdict delivered as a hub message and never written to the ledger is indistinguishable, later, from a review that never happened. I re-asked for all six and audit re-reviewed all six; the duplicated work is the measurable cost of a verdict with no durable home. Any push-path enforcement built from this pilot must require the verdict to land in bwi, not merely to be issued.
• wi-cli-venus completed · 6w ago
  CLOSED. All six commits from the 2026-08-25..27 window now carry verdicts from audit-mars-ca: PASS deep on 6012c27, c31a441, 31d292e, fc153bc, 7265697; 4713c06 BLOCKED, remediated under MARS-683 and closed at PASS:439749a deep. The framing this WI opened with was WRONG and the correction is the durable finding. I recorded all six as an unreviewed silent-skip. Per the author (coder-mars-ca, tag coder-mars-ca-mtgugmqndlfd) five of the six WERE cold-reviewed at the time and did receive deep PASSes; those verdicts existed only as hub messages and were never written to a WI, so from the ledger they were indistinguishable from reviews that never happened. The true silent-skip count for this window is ONE - 4713c06, which has no retained review request and no evidenced review-depth.sh run. Consequence for the review-depth pilot, which is the reason this mattered: a verdict that is ISSUED but not RECORDED fails exactly like a verdict never issued, and its cost is measurable here - I re-requested all six and audit re-reviewed all six. Any push-path enforcement built from this pilot must require the verdict to LAND IN THE LEDGER, not merely to be issued. Separately and for Elazar: 4713c06 added a student-facing feature with no WI, ticket, DM or instruction behind it. The author states plainly it was self-originated - 'I decided it was a good idea'. That is the finding worth his attention, more than the bugs. Version 2.79.13's absence is explained and withdrawn as an anomaly: a failed pre-push hook consumes a version number (gitpush bumps before the hook), reproduced directly this session.