[MrScripty]

The caption browser used three fixed 100 ms pauses before handling confirmation dialogs. An older duplicate CI run received “No dialog is showing,” then never completed its unbounded diagnostic evaluation. The test now arms a fresh exact-confirm waiter before each click, bounds CDP requests and diagnostics, drains pending requests on close/error, and awaits owned child shutdown before removing the private browser profile. All original cancel/discard, review, reopen, export and narrow-layout assertions remain intact; product and workflow files are unchanged.

At abbae728434fa0670478e12216a1d94603a97daf, independent syntax checks and the controlled protocol regression pass. The regression reproduces the old failure with a 180 ms dialog delay and covers delayed cancel/accept, absent/wrong dialogs, trigger failures, request expiry, late replies and socket drainage. The existing caption CI entrypoint runs these checks before its real browser fixture.

The complete PostgreSQL 17.6 / Chromium 151 caption journey also passed with exact source-file hashes matching this commit: three observed decisions false/true/true, all original assertions, four-record ZIP and 390 px layout. Both registered test schemas were removed and queried absent; owned child exits preceded temporary removal. The receipt was generated before committing, so its source_commit identifies the base HEAD; the recorded modified-file hashes establish the tested source. The existing generation provider is a controlled fixture, with no model-quality or training claim. Full evidence, original cancelled-run disposition and diagnosis limits are in docs/plans/lab-workflow-reliability/reports/caption-dialog-ci-20261020.

Current composed head 0dedbc953902ac89389dca0693006aaa02a0a87c follows PR111 at exact base 29011daea37f101ee30be9bc10aab702a16c830a. A normal whole-history merge carries the independently reviewed project-startup readiness fix; product bytes exactly match PR111, and both caption and startup qualification file hashes remain equal to their original tested snapshots. The two focused regressions and syntax checks also pass on the composed tree. Hosted acceptance and serial integration remain pending. The original duplicate is cancelled, never counted as passing; its newer independently reviewed same-head primary remains the PR105 acceptance gate.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[MrScripty]

Disposition (2026-10-21): closing as already integrated by PR114, merge 360bbda950cc48ed3c3d9193539039b68e0ac8cb. Exact head 0dedbc953902ac89389dca0693006aaa02a0a87c is an ancestor of current main 133d53c0e2985a6559ba9cebd7d4a9f530051f06; all 18 changed paths match main byte-for-byte. Independent GPT-6.1 Sol High disposition review cleared closure. Fresh local CDP/dialog, access-navigation, page-load and owned-child regressions passed.

Exact-head run39370108074 passed; sibling 39370107550 remains failed. Current main's five automatic push workflows passed, including lifecycle39539151192 and UI39539151200. Neither the failed nor cancelled historical run is counted as passing.

Dependent #113 was closed first after confirming complete integration; its attempted main retarget was rejected by GitHub because no new commits remain. All original source/evidence, branch and commits are preserved. No new scientific/model acceptance is claimed.