Repository navigation

修复 WebArena 轨迹重放评测并发布五方法对比与独立解读 - #23

Merged

lanyue-llk merged 71 commits into

mainfrom

codex/browser-root-cause-report

Sep 22, 2026

Merged

修复 WebArena 轨迹重放评测并发布五方法对比与独立解读#23
lanyue-llk merged 71 commits into
mainfrom
codex/browser-root-cause-report

Commits on Sep 17, 2026
- Diagnose browser benchmark failures and publish repair evidence
  kevin
  committed
  55f9d40
- Expose diagnostic replay appendix through release targets
  kevin
  committed
  38bbbbf
- Tighten report source and path privacy guards
  kevin
  committed
  82405e0
- Report unmatched diagnostic browser audits explicitly
  kevin
  committed
  8ad269b
- Cover diagnosis pages in semantic release checks
  kevin
  committed
  16f0837
- Require absolute retrieval evidence and safe closure paths
  kevin
  committed
  e1737bc
- Reject symlinked report evidence paths
  kevin
  committed
  f6196b2
- Add diagnostic replay qualification candidates
  kevin
  committed
  2b4add1
- Clarify unresolved retrieval qualification in replay report
  kevin
  committed
  75e5d60
- Label limited-scope retrieval audits in replay details
  kevin
  committed
  e7aa40e
- Bind retrieval candidates to recompiled canonical prefixes
  kevin
  committed
  eab9eed
- Fail closed on unproven retrieval observation tails
  kevin
  committed
  d5b2a15
- Exercise profile-save evaluator correction end to end
  kevin
  committed
  0cb703a
- fix(replay): require full observed scope for empty retrieval
  kevin
  committed
  eea1f76
- Show latest Moli affected-task diagnostics separately
  kevin
  committed
  95c74ef
- Predeclare additional mutation completion diagnostics
  kevin
  committed
  cdbe5be
- Predeclare remaining mutation completion diagnostics
  kevin
  committed
  b67385c
- Normalize textarea HTML observation across parsers
  kevin
  committed
  f998da9
- Predeclare wishlist persisted-state observer for task 465
  kevin
  committed
  9df4d0b
- Predeclare durable order observer for task 506
  kevin
  committed
  9137a7d
- Observe GitLab comments through complete discussions API
  kevin
  committed
  d209e61
- Fix persisted-state observers for dynamic address and post body
  kevin
  committed
  147a400
- Merge remote-tracking branch 'origin/main' into codex/replay-qualification-candidates
  kevin
  committed
  82b36ad
Commits on Sep 18, 2026
- Prepare corrected 90-task three-browser replay and evidence report
  kevin
  committed
  624ec66
- Validate every task reset receipt at three-browser aggregate boundary
  kevin
  committed
  159dbfb
- Pin original cross-track report to immutable replay inputs
  kevin
  committed
  a01548d
- Bind replay dataset to corrected fork and method-level reset
  kevin
  committed
  13118ef
- Bind corrected task to paired replay qualifications
  kevin
  committed
  f3b27c8
- Clarify shared BrowserTool scope in unified replay report
  kevin
  committed
  8f3e644
- fix(replay): enforce method-scoped reset in qualification receipts
  kevin
  committed
  5ab2ee4
- docs(replay): describe current three-method cohort workflow
  kevin
  committed
  875f511
- fix(replay): name per-method site reset scope accurately
  kevin
  committed
  a537fba
- Bind replay cohort to latest Moli main and new report site
  kevin
  committed
  5b7bbc7
- fix(replay): stabilize sequential Chromium qualification and shared observers
  kevin
  committed
  da21388
- fix(replay): preserve revised input provenance and update regression gates
  kevin
  committed
  e0e3833
- fix(report): expose corrected acceptance from bound evaluator evidence
  kevin
  committed
  44c6844
- fix(report): distinguish acceptance authority from independent review scope
  kevin
  committed
  0f9b7fe
- fix(report): expose recorded page state alongside step errors
  kevin
  committed
  2918063
- Freeze released and PR Moli 1.1.8 comparison participants
  kevin
  committed
  2104da9
- Localize replay failure treatment labels in details and dialogs
  kevin
  committed
  68f1b40
- Distinguish unmatched network acceptance from absent requests
  kevin
  committed
  fc6ecfe
- Complete three-method replay report and preserve historical review
  kevin
  committed
  523bbf6
- Bind new verified three-browser replay report publication
  kevin
  committed
  5d11a3f
Commits on Sep 20, 2026
- Audit replay input symmetry and upstream task contract changes
  kevin
  committed
  bf92b05
- Add evidence-derived failure family matrix and drilldown release checks
  kevin
  committed
  5e1c4dd
- Bind read-only geometry probes and audit changed task contracts
  kevin
  committed
  7fcafae
- Restore final-save scoring and expose redirected request recording defect
  kevin
  committed
  f43acfe
- Add bilingual replay interpretation and self-contained article export
  kevin
  committed
  624a961
- Bind five additional replay failures to read-only geometry evidence
  kevin
  committed
  740f7e3
- Separate replay readiness races from browser defects with dispatch evidence
  kevin
  committed
  6c707dc
- Explain replay mechanisms with bilingual diagrams and measured specimens
  kevin
  committed
  fc1f203
- Verify cookie-gated header loss and restore cart quantity acceptance
  kevin
  committed
  c32a055
- Explain form lookup work with regression specimens and bilingual diagrams
  kevin
  committed
  512e9d6
- Make replay English generation opt-in and extend mechanism article
  kevin
  committed
  640211d
- Make Browser Use English generation and checks opt-in
  kevin
  committed
  67d4b97
- Exclude unused English diagnosis prose from Chinese replay bundles
  kevin
  committed
  f062be4
- Avoid duplicate standalone article image payloads
  kevin
  committed
  6ceb31e
- Make Web Fetch English reports opt-in
  kevin
  committed
  f0d794a
- Support staged upload for large replay reports
  kevin
  committed
  9cad952
- Allow slow staged artifact uploads
  kevin
  committed
  0d2d6de
- Bind replay report to published version
  kevin
  committed
  8da79a6
- fix(replay): restore evidence bindings and integrate main
  kevin
  committed
  e871316
- Append Moli 1.1.9 and rebuilt PR replay results
  kevin
  committed
  1864f52
Commits on Sep 21, 2026
- Fix standalone replay interpretation access and navigation
  kevin
  committed
  67c64fc
- Bind verified standalone interpretation publication
  kevin
  committed
  2e4f1e0
Commits on Sep 22, 2026
- chore(replay): sync latest main
  lanyue-llk
  committed
  170bae3
- refactor(replay): remove retired diagnostic report pipeline
  lanyue-llk
  committed
  8c57fd5
- refactor(replay): externalize report content and share execution
  lanyue-llk
  committed
  f39edab
- refactor(replay): archive retired diagnostic records
  lanyue-llk
  committed
  926de10
- chore(replay): sync main and retain both asset contracts
  lanyue-llk
  committed
  72de960
- refactor(replay): remove archived diagnostic snapshots
  lanyue-llk
  committed
  afa380f

Commits

Commits on Sep 17, 2026

Commits on Sep 18, 2026

Commits on Sep 20, 2026

Commits on Sep 21, 2026

Commits on Sep 22, 2026