[pr-attention-queue] Guard merge recommendations with current review evidence

[PureWeen]

Summary

Prevent PR Attention Queue and Pulse from recommending a pull request for merge when current reviewer feedback or discussion still needs action, and refresh the dashboard daily.

• Track feedback and renewed review requests per reviewer, so one reviewer’s later approval cannot hide another reviewer’s outstanding concern.
• Distinguish Waiting on author from Verify discussion before merge instead of inventing author ownership for uncertain evidence.
• Require complete, clear bounded discussion evidence before displaying a prospective ReadyToMerge candidate.
• Carry the same additive eligibility contract through standalone queue output and both Pulse scopes.
• Schedule Pulse daily at 12:50 UTC, including weekends, while retaining manual dispatch.

Why

The previous aggregate review state could classify a PR as ready after a later approval even when another reviewer still had actionable feedback. Retained examples also showed unresolved or ambiguous discussion being omitted from merge recommendations.

This change keeps existing ranking, budgets, display caps, and review-now behavior. It adds conservative merge qualification rather than changing the underlying prospective classification. It does not merge PRs.

Important behavior

• A same-reviewer approval can settle feedback when it covers that feedback’s commit or the current head. An already settled feedback/approval pair does not become unsettled solely because of a later push; an unrelated older approval does not clear current feedback.
• Missing, truncated, unknown, or current unresolved context routes to verification. Empty review text alone is not clearance, but complete resolved/outdated context can legitimately be clear.
• Legacy payloads without the additive extension remain supported conservatively; partial or inconsistent extensions are rejected.

Review follow-up

Commit da49acb428 addresses the review round:

• Per-reviewer chronology: fixed and covered with faithful red/green cases.
• Comment count versus excerpt coverage: the reported defect was not confirmed; assessment completeness is distinct from the ten-excerpt display cap. Added producer/consumer controls.
• Empty review bodies: clarified and covered the existing bounded-context policy instead of adding a blanket empty-body rule.

All three threads have evidence-backed replies and are resolved.

Daily schedule

Commit 007c7b8e18 adds schedule: daily, compiled to 50 12 * * *, alongside workflow_dispatch.

Scheduled execution starts after the workflow change is merged to the default branch. It updates the same dashboard, #69328, using the existing fixed-target checks, publication validation, model settings, and serialized workflow. The compiler also adds its standard scheduled-agent concurrency group. Queue and reporting scripts are unchanged by this scheduling commit.

Validation

• Queue: identical 60-case harness had 13 expected failures and 47 passing controls on 0e7570ec; all 60 pass with the review fix. Full queue entry point passes.
• Pulse comment coverage: 45/45 controls pass on baseline and fixed code through the real producer, sanitizer, and private validator.
• Two-scope pipeline: 131/131 assertions pass through producer → sanitizer → renderer → private validation. The six synthetic variants are test cases, not claims about similarly numbered live PRs.
• Scheduling: 9/9 focused controls pass, including two missing-trigger failures on the manual-only baseline and seven passing activation controls. The generated activation script and pinned helper are exercised with mocked GitHub permission responses.
• Official gh-aw v0.88.7 strict compilation and explicit validation pass; repeated compilation produces identical lock output. The complete Pulse suite passes on the scheduling commit.

Hosted verification

Run 35137102234 successfully executed review-fix commit da49acb428 and published #69328. Both scopes collected 379/379 open PRs; the stored body exactly matched the protected canonical artifact. Exactly one safe-output mutation occurred, changing only the dashboard body and timestamp; the fork dashboard was unchanged. Both inference stages used only gpt-5.6-sol, with token steering disabled and clean threat detection.

A prior attempt timed out in threat detection and failed closed without publishing. No new hosted run was dispatched for the subsequent scheduling commit. Its compilation and activation controls are verified locally; actual GitHub scheduled execution remains to be observed after merge.

Scope

This remains a PR-attention correctness change plus daily scheduling—not a broader repository-activity dashboard or deterministic-publisher conversion. No script relocations, new scopes, budgets, dependency updates, or permission changes are included. Maintainer approval is required; the user owns the merge.

[Copilot]

Copilot review overview

🟡 Changes recommended

Two critical and one moderate unresolved findings affect merge-gate correctness.

Get a fresh assessment by requesting another Copilot review.

Review tier: Lite
Findings: 2 · 1

Open (3)

• Earlier actionable review feedback is lost after another approval · New
• Reject false non-truncated counts for eligible candidates · New
• Empty review bodies are treated as clear merge evidence · New

What changed in this PR

This PR adds bounded current-review and discussion evidence to merge recommendations and carries conservative eligibility through the PR Attention Pulse pipeline.

Changes:

• Adds merge eligibility, verification views, counters, and caps.
• Extends Pulse sanitization, validation, rendering, and legacy compatibility.
• Adds queue/Pulse tests, fixtures, documentation, and presentation updates.

Outstanding findings: Critical (3 votes): later approval by another reviewer can incorrectly clear actionable feedback. Moderate (3 votes): empty COMMENTED bodies do not require verification. Critical (1 vote): inconsistent truncation metadata can allow incomplete evidence through the merge gate.

File	Summary
.github/​workflows/​pr-attention-pulse/​Sanitize-PRAttentionPulse.ps1	Sanitizes merge evidence and views.
.github/​workflows/​pr-attention-pulse/​PRAttentionPulseContract.psm1	Validates and renders merge extensions.
.github/​workflows/​pr-attention-pulse/​Combine-PRAttentionPulse.ps1	Normalizes combined areas.
.github/​workflows/​pr-attention-pulse-tests/​Test-PulseReviewRequirements.ps1	Updates Pulse controls and reference validation.
.github/​workflows/​pr-attention-pulse-tests/​Test-PulseMergeRequirements.ps1	Adds Pulse merge contract tests.
.github/​workflows/​pr-attention-pulse-tests/​Test-PRAttentionPulse.ps1	Updates presentation and integration tests.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​published-34643961191.expected.txt	Updates the published presentation golden.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​provenance.json	Records fixture provenance.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​normal-legacy.expected.txt	Updates the legacy presentation golden.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​incomplete-query.expected.txt	Updates the unavailable-state golden.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​complete-zero.expected.txt	Updates the empty-inventory golden.
.github/​workflows/​pr-attention-pulse-tests/​fixtures/​presentation/​bounded-empty-verification.expected.txt	Updates the verification golden.
.github/​skills/​pr-attention-queue/​tests/​Test-PRAttentionQueue.ps1	Runs merge eligibility tests.
.github/​skills/​pr-attention-queue/​tests/​Test-MergeEligibility.ps1	Adds merge behavior coverage.
.github/​skills/​pr-attention-queue/​tests/​MergeEligibilityTestHelpers.ps1	Provides queue test transport.
.github/​skills/​pr-attention-queue/​tests/​fixtures/​merge-eligibility.json	Adds live-derived merge fixtures.
.github/​skills/​pr-attention-queue/​SKILL.md	Documents merge evidence semantics.
.github/​skills/​pr-attention-queue/​scripts/​PRAttentionQueue.psm1	Collects and applies merge evidence.