Promote staging to main

[stateofkate]

Caution

DO NOT USE THE MERGE BUTTON ON THIS PULL REQUEST.
THE BUTTON CREATES A NEW COMMIT AND BREAKS THE RELEASE MODEL.
COMMENT /promote TO COMPLETE THE PROMOTION.

Summary

Ships the quota lock removal, the QA accept/reject verdict rework, the network-shape e2e de-race, and /promote <sha> support in the promotion workflow.

Commits in this promotion

e43f64f8 ci(promote): support /promote <sha> and share the promotion checks (#1133)
8b7b2367 test(e2e): de-race the network-shape request counting (#1128)
fa608980 QA verdict: accept/reject labels, plain prompt, keep verdict until replaced, scoped rerun (#1120)
9f26989e fix(quotas): remove the org-wide advisory lock from trial submission (#1118)

Validation

Staging Deploy is green on the staging tip e43f64f8a1228767055eb76930fc9b0362cbe93a (run completed 2026-08-08T23:39:32Z). main is an ancestor of the tip, so the promotion fast-forwards cleanly.

---

Note

High Risk
Quota admission without locks changes overshoot behavior under concurrency; verdict prompt and persistence logic affect production QA outcomes; promotion script changes gate what ships to main.

Overview
This promotion bundles release tooling, quota admission, a QA verdict overhaul, and Playwright network-shape hardening.

Promotion workflow centralizes preflight in verify_promotion_target.sh. Bare /promote and Promotion Preflight default to the promotion-target SHA in the PR body (unfilled placeholder fails; no marker → staging tip). /promote <sha> and the target_sha input override the pin. Success comments note when staging still has unpromoted commits after a partial promote.

Quotas: trial submission and sweeps no longer take blocking org/user advisory locks; admission can briefly overshoot caps and the enforcement sweep cancels overage using non-blocking try_acquire_quota_locks only.

QA / verdict: the model outputs accept / reject (stored payloads still include is_good). The verdict prompt is rewritten around immediate rejection rules (leaks, weak tests, baseline failures). Successful verdicts persist through new trials, retries, and QA cancel until a new pass replaces them; Rerun verdict calls qa/backfill with force: false. The task overview lists all version trials (foreign trials marked “elsewhere”). cc_chat is removed end-to-end per the changelog.

E2E: shared network-log.ts attributes requests by issue time and holds responses so StrictMode abort races do not flake counts. Trial retry locks the task row before reserving the next trial index.

^{Reviewed by Cursor Bugbot for commit 33a8e4e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
oddish-app	Ready	Preview	Aug 9, 2026 12:38am
oddish-app (staging)	Ready	Preview	Aug 9, 2026 12:38am

[cursor]

Quota cancel ignores kept verdict

High Severity

_reconcile_cancelled_tasks still marks exhausted tasks FAILED and hard-sets an in-flight verdict to FAILED, without cancel_verdict or the SUCCESS→COMPLETED promotion used by user cancel. After append keeps a successful verdict, a quota overage cancel can leave FAILED with an accept payload (or wipe a restorable verdict while the payload remains). Lock-free admission makes this cancel path more common.

Additional Locations (2)

• oddish/src/oddish/queue.py#L287-L294
• oddish/src/oddish/core/verdict_sync.py#L117-L131

 

^{Reviewed by Cursor Bugbot for commit fa60898. Configure here.}

[cursor]

Risk: high. Left a non-blocking comment: Cursor Bugbot finished as skipped and left an unresolved high-severity finding on quota-cancel verdict handling, so this is not approved. Assigned reviewers for human review of this billing/quota promotion.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

There are 2 total unresolved issues (including 1 from previous review).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Want reviews to match your repository better? Bugbot Learning can learn team-specific rules from PR activity. A team admin can enable Learning in the Cursor dashboard.

^{Reviewed by Cursor Bugbot for commit 33a8e4e. Configure here.}

[cursor]

Empty QA path clears kept verdict

Medium Severity

When a QA job finds no eligible trials, this path sets verdict to None and records SUCCESS. With verdicts now kept across append until a replacement write, that race no longer completes a blank slate — it deletes the kept accept/reject payload the append path preserved.

 

^{Reviewed by Cursor Bugbot for commit 33a8e4e. Configure here.}