ci: target a four-minute full PR quality loop

[msrivas-7]

Goal

Move normal full PR CI toward a 4-minute median / 5-minute p95 without removing tests, weakening browser/security coverage, relying on retries, or requiring paid runners.

Changes

• Balance all 439 Chromium tests by learned duration instead of test count; every test is assigned exactly once.
• Learn timing only from clean exhaustive runs, with a tracked cold-start seed and trusted cache.
• Split the complete 63-test advisory critical lane across isolated zero-retry jobs.
• Pull immutable backend, runner, and frontend images concurrently.
• Run duration learning, shadow evidence, and image cleanup in parallel after testing.
• Remove unused frontend/browser setup from the API-only security suite while preserving every security scenario and tcpdump sentinel.
• Keep the measured 16-full/2-critical topology; no tests or browser/security lanes are removed or demoted.
• Make Docker cancellation completion race-safe after the confirmation run exposed a real fast-Stop hang.

User journey change

Before: a very fast Docker kill could occasionally leave Editor stuck on “Stopping this run…”.
Now: Stop always settles to “Run stopped,” restores Run, and the learner can execute again immediately.

Measured results

Workflow	Before	Final-head clean 1	Final-head clean 2
Full E2E	6m41s	4m30s	4m42s
Security suite	3m58s	2m54s	—
Platform CI	—	3m25s	—

The two clean exact-head E2E samples have a 4m36s observed median, and both finish below five minutes. All 439 Chromium tests, the complete advisory critical lane, Firefox, WebKit, and every security scenario passed without retry dependence. Two samples do not establish a statistical p95; trusted clean history will continue measuring that target.

The first overlapping final-head E2E attempt exposed one transient bootstrap data-load failure and was not counted. Its retry passed, but the workflow remained red by design. The two complete reruns above then passed with zero flaky tests; no timeout, assertion, or retry policy was weakened.

Rejected experiments

• 20 full / 4 critical: increased aggregate database pressure and produced failed shards; the clean test-path gain was below the tracked 20-second adoption threshold.
• Per-stack database pool caps: a moderate cap reproduced the upstream postgres.js 3.4.9 transaction-reservation defect; one connection avoided the immediate error but missed the runtime target. The experiment was removed completely.
• No unsafe or unreliable experiment remains in the final PR diff.

Validation

• 439/439 Chromium selectors accepted exactly once.
• Workflow, duration-planning, image-pull, security, harness, benchmark, and release contract tests pass.
• Pinned actionlint and diff checks pass.
• The Stop fix passed 31 focused backend tests, the complete 1,424-test backend suite, 10/10 zero-retry focused E2E repetitions, and a real authenticated in-app-browser journey with two Stop cycles plus a successful run between them.

Final gate

The exact final commit passed two complete clean 16/2 E2E runs, security, platform CI, and Codex review. After merge, timings will be verified again on main.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-08-31T22:37:11.817881Z	733ba45	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5adf6bdc62

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[msrivas-7]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Delightful!

Reviewed commit: 733ba45e5b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".