aetherbound-guild/.agent-taskgraph/queue/failed/ABG-P9.1-FOUNDATION-REVIEW-R5/report.md

Review Report: ABG-P9.1-FOUNDATION-REVIEW-R5

ABG-P9.1-FOUNDATION-REVIEW-R5 FAIL: successful-command revision invariant (open_new_session revision delta=2, expected 1)

• Reviewer HAPI: 6fac5053-6144-4bfd-944d-b131988066cd (archived).
• Immutable candidate: bd505d6188556b5d407bebe9769d1c99e74dc9c1.
• Earliest product boundary: one successful open_new_session command advances revision twice.
• Fresh runtime observation: revision_before=0, revision_after=2, screen=first_mission.
• Cause: reset_transient() increments revision before _change_screen() -> _success() increments it again.
• Evidence: /private/tmp/abg-p91-foundation-r5-evidence.UYsX7a/terminal.txt (ae6045c523036e2aeb075303980d8145cdba5961d5b54b8cae13d38f7e510945).
• Evidence manifest: 9048e01d86f8c5bdaa8143ef0d4a6452d9f8556bf97dc416e9864c93c10a000b, all 8 listed evidence rows verify.
• Cleanup/integrity: 317/317 tracked runtime blobs match; reviewer import root absent; worktrees clean; owned Godot zero.

The successor repair must change only the smallest state/service/test boundary needed to make every successful public command increment revision exactly once, while rejected commands remain snapshot/revision no-ops.