fix(review): regate_repair_exhausted on gittensory#4976 (5 attempts, 78 Sentry events)

[JSONbored]

Summary

PR gittensory#4976 has exhausted all 5 automatic repair attempts and remains in a presumably-stuck state.

Evidence (Sentry)

GITTENSORY-1E — regate_repair_exhausted: (JSONbored/gittensory#4976) headSha=cb0517aa7c03acac534ce400ec5ac40c56fa4588, attempts=5. 78 events, first seen 2026-07-07, last seen 1 hour before filing (ongoing).

Requirements

1. Check the live current state of gittensory#4976 on GitHub — is it still open, what's actually blocking it (CI, review, a genuine stuck-review condition)?
2. Determine whether this is a one-off (a specific PR that needs manual intervention) or a symptom of a systemic issue in the repair/regate mechanism itself (e.g. the same underlying cause as GITTENSORY-1C's orb-relay race, or GITTENSORY-1F's stuck-review pattern on a different repo).
3. If manual intervention resolves it: no code change needed, just document the finding. If it reveals a systemic gap in the repair-exhaustion path (e.g. no escalation/alert when repairs exhaust, which is arguably itself a bug — 78 silent exhaustion events with nobody notified), consider whether that's worth its own fix.

Deliverables

• Investigation finding on PR fix(review): guard contributor screenshot fetches #4976's actual current state.
• A decision, documented in the issue/PR: manual fix, or does the repair-exhaustion path itself need a change (e.g. escalating after N exhaustions instead of repeating silently).

[linear-code]

JSO-3618

[JSONbored]

Finding

PR #4976's actual state (requirement #1): it was blocked by a genuine merge conflict against main (mergeStateStatus: DIRTY, mergeable: CONFLICTING), not a stuck-review condition — the repair loop was correctly repeatedly trying and correctly repeatedly failing, because a merge conflict isn't something an automated repair pass can resolve. Rebased it (worker-configuration.d.ts's generated header only), which surfaced a second, real, pre-existing regression in the PR itself (its screenshot-table-vision-wiring.test.ts coverage had gone stale against its own new Content-Type validation — 7 failing tests) — fixed both; the PR is now mergeable: MERGEABLE with green tests. See the PR for detail.

Systemic vs one-off (requirement #2): pulled the raw event list behind GITTENSORY-1E rather than trusting the issue's own summary — the same "one Sentry issue aggregates every occurrence of a generically-templated event" pattern found in #4998. The 78 (in fact, far more in the full history) events span dozens of distinct PRs across all three tracked repos (gittensory, awesome-claude, metagraphed), not one stuck PR. Checked several of the other repeat offenders (gittensory#4204, #4215, #4371, metagraphed#4283) directly on GitHub: all of them are already MERGED. They weren't stuck — they were high-churn PRs (several are bot/automation PRs: chore(release), chore(registry): sync operational-surfaces.json) whose head SHA moved faster than the ~2-minute priority-repair sweep could keep up with, correctly exhausted their 5 attempts, correctly fell back to ordinary staleness cadence exactly as surfaceRepairPriorityPullNumbers's own code comment describes, and merged fine afterward via normal review passes.

Decision (requirement #3)

No change to the repair-exhaustion path itself. The mechanism is working as designed: it's the correct, intentional backoff for a PR whose head keeps moving (or that's structurally blocked, like a merge conflict) faster than the aggressive priority-repair cadence can complete — and the regate_repair_exhausted log is the escalation (it reaches Sentry via forwardStructuredLogToSentry, which is exactly how this got triaged). Building a further escalation tier (e.g. auto-comment on the PR, or a dedicated alert) isn't warranted by what the evidence shows: the volume is dominated by benign, self-resolving churn, and PR #4976's specific case needed exactly the kind of one-time manual intervention (rebase) this issue's own text anticipated as a valid resolution.

PR #4976: #4976 (fixed, rebased, tests green)