What happened
The fullsend bot force-pushed 10 times over 27 days on PR #29. Three of those (Jun 10, Jun 23 x2) triggered review runs, each producing identical CHANGES_REQUESTED verdicts with the same protected-path finding. The PR diff against main never changed — all force-pushes were rebase-only. After Jun 23, seven more force-pushes occurred without triggering reviews, suggesting partial suppression is in place.
What could go better
Review dispatch should detect that the PR diff is unchanged after a rebase-only force push and skip re-review. The partial suppression prevented reviews on 7 of 10 force-pushes but not the initial 3 redundant runs. Token cost was 3x what was needed for this 5-line diff.
Proposed change
Additional evidence for #1287 (skip re-review on rebase-only changes) and #1355 (skip re-review for bot rebases). All three reviews produced identical findings on an unchanged 5-line diff across the same file.
Validation criteria
After #1287 is implemented, rebase-only force pushes that don't change the PR diff should not trigger new review runs.
Generated by retro agent from fullsend-ai/experiments#29
What happened
The fullsend bot force-pushed 10 times over 27 days on PR #29. Three of those (Jun 10, Jun 23 x2) triggered review runs, each producing identical CHANGES_REQUESTED verdicts with the same protected-path finding. The PR diff against main never changed — all force-pushes were rebase-only. After Jun 23, seven more force-pushes occurred without triggering reviews, suggesting partial suppression is in place.
What could go better
Review dispatch should detect that the PR diff is unchanged after a rebase-only force push and skip re-review. The partial suppression prevented reviews on 7 of 10 force-pushes but not the initial 3 redundant runs. Token cost was 3x what was needed for this 5-line diff.
Proposed change
Additional evidence for #1287 (skip re-review on rebase-only changes) and #1355 (skip re-review for bot rebases). All three reviews produced identical findings on an unchanged 5-line diff across the same file.
Validation criteria
After #1287 is implemented, rebase-only force pushes that don't change the PR diff should not trigger new review runs.
Generated by retro agent from fullsend-ai/experiments#29