🐛 - Bugbot review projection remains partial after historical review update failures #886

Is there an existing issue for this?

• I have searched the existing issues.

Which actions are affected?

Pull Request

Which platforms are affected?

Linux

Description

Bugbot completes all analysis partitions with complete context and zero new findings, but fails while projecting the current state onto historical GitHub reviews. The aggregate remains correctly fail-closed (outcome="partial", coverage="complete") and identifies two historical review IDs. The execution check reports two identical generic provider.unavailable failures with the same diagnostic reference, without an HTTP status or per-review actionable cause.

Please keep the partial result fail-closed. This report requests reliable, safely diagnosable projection/recovery: preserve completed analysis and persisted state, identify the failed review/provider operation with bounded diagnostics, and let a verified subsequent execution reconcile only valid current state. Missing writes must never be treated as completed approval. A provider outage itself may be expected; the observed consumer limitation is the incomplete historical projection and generic recovery evidence after otherwise complete analysis.

This run uses the original immutable v3 revision below. It does not establish a regression in a newer upstream revision.

Observed consumer and exact identities:

• Repository/issue/PR: vypdev/ha-ui #7, PR #10.
• Exact source head: 4c43b233f10d8b7cccd81a393f8e03afab27fb09; comparison base d885bccf2c3152c3151471d3efaabd901429d6bd; original Action-owned feature ref.
• Native PR workflow: run 41810111066, terminal failure.
• Generated failing review check: 126508813987.
• Aggregate card: comment 6622300120, observed at this head with outcome="partial", coverage="complete", open/reopened/dismissed/verification/unknown all zero, four historical findings fixed.
• Failed historical review IDs in recovery details: 5764556861, 5765662391.
• Error reference: dbbe6e75-f4eb-43e0-b425-6b16be4103bf, repeated for both published provider failures.

Previous exact-head review succeeded at 8cdf397aed36c6b1ae77a07e7e4238e0a51833ae: native run 41802704905, generated check 126483830766, complete context/outcome and zero active states. That earlier success does not certify the current head.

Reproducing the issue

1. Inspect ha-ui PR Test Automated Version #10 at the exact 4c43b23 source and existing four historical resolved findings/reviews.
2. Run the ordinary ready-PR synchronize path with the installed immutable v3 Action. This occurrence was an ordinary new source push, not an unchanged manual workflow retry.
3. Observe all 10/10 analysis partitions complete at 2026-11-04T09:54:14.1569483Z.
4. Observe the current aggregate update at 09:54:29.3517015Z, followed by projection outcome failed in telemetry, and the failed Copilot / Review check at 09:54:36.5690170Z.
5. Inspect aggregate recovery details: Unable to update Bugbot review 5764556861 and Unable to update Bugbot review 5765662391.
6. Inspect the job/check diagnostics: provider temporarily unavailable, retryable yes, persisted state/external effects retained, two entries sharing reference dbbe6e75-f4eb-43e0-b425-6b16be4103bf.

Actual telemetry establishes complete context and analysis, not successful projection:

• contextCoverageStatus: complete; all context groups complete, no omitted/truncated items.
• analysisPartitions: 10; completedAnalysisPartitions: 10.
• changedFiles: 136; changedLines: 11355; analysisDiffFragments: 148.
• candidateFindings: 0; publishedFindings: 0; finding states fixed=4, every other state=0.
• analysis: 265319 ms; projection: 13958 ms; total 282820 ms; outcome failed.

The visible logs do not expose the provider HTTP failure or failed request payload; those causes remain unverified. A GitHub push had separately returned Internal Server Error at 09:46:57Z (request FBCF:A18E6:1DEB1C:50088C:6AEAFF10); Git/GitHub proved the ref unchanged and the same normal fast-forward push later succeeded. That is a separate observation suggesting possible provider instability, not proof of a common cause.

copilot Version

vypdev/copilot@v3, downloaded SHA 3e7ceea08e6131360e026937d1c0cc1b120472e0; codex-cli 0.166.0, configured model gpt-6-luna.

Relevant Log Output

2026-11-04T09:54:14.1569483Z Bugbot reviewer completed partition 10/10.
2026-11-04T09:54:14.1574507Z Bugbot reviewer completed in 265319ms.
2026-11-04T09:54:29.3517015Z Comment 6622300120 updated in Issue 10.
2026-11-04T09:54:34.4476069Z Publishing result: 3 result(s), 4 step(s), 2 error(s).
2026-11-04T09:54:36.5690170Z Published Copilot / Review Check Run for 4c43b233f10d8b7cccd81a393f8e03afab27fb09.
Impact: The provider was temporarily unavailable.
Error code: provider.unavailable
Action: Retry when the provider is available.
Retained state: Existing persisted state and completed external effects were preserved.
Retryable: Yes
Reference: dbbe6e75-f4eb-43e0-b425-6b16be4103bf

Additional context and comments

Independent CI Check 41810111300 / job 126506963878 succeeded: 488 source/Chromium tests, all four supported runtime coverage metrics at100%, strict checks, reproducible five-package cohort/four actual compiled consumers, and the48-combination static catalog campaign. Those product results do not override this partial review or complete the retained64-case feature/six delivery refinements.

The separate push-progress assessment-text-limit occurrence (run41810104777, check126509175210, reference d0766c9d-46d5-40a1-a6db-a92846b39525) belongs to existing issue #829 and is not the projection failure reported here. No acceptance relaxation, manual review dismissal, remote branch replacement, Copilot source change or installed-resource change was used. Source, original logs, actual product inputs and verification evidence are retained locally. Product work continues independently on a newly reproduced native-form corner case; a real corrective push will receive a fresh native review.