Skip to content

Historical field note

This record preserves prior investigation or planning evidence. It is not a current operating contract; use the numbered book and current runbooks for instructions.

Dream-repair review finding: stalled-finalizer recovery

This note records the investigation outcome for a low-confidence dream-cycle repair finding (fingerprint dreamrepair:9a65379bca2038941c22a69eca8814c2, scope project, repo area mac). It is the durable, product-tracked artifact for that review; the per-run executor/worker diagnostics live in the task workspace and are intentionally kept out of git.

Scope

The finding is a failure_pattern supported by a single memory record (mem_a52c2968410947e5baad3b3974c78845, task task_repair_e57870febced381b628b6a68) whose candidate summary is:

Repair contract prerequisites: Implement hub-side stalled-finalizer recovery in src/mac/repository_recovery.py (inspect + recover).

This review examined whether that finding maps to an actionable defect or a missing deliverable in the repository-recovery code path.

Finding

The finding is not actionable as a code change: the deliverable it points at is already implemented, wired, documented, and tested in the current tree.

  • src/mac/repository_recovery.py provides both halves of the stalled-finalizer path: inspect_stalled_finalizer_recovery returns a mutation-free recovery plan for a finalizer interrupted by timeout, cancellation, or crash (a finalizer-progress.json stuck in a non-terminal status), and recover_stalled_finalizer harvests verified work, revalidates, and performs the guarded push.
  • The path is CLI-wired via task recover-stalled-finalizer (cmd_task_recover_stalled_finalizer in src/mac/cli.py), including --approve-new-file, --evidence-id, and --execute (read-only plan by default), and is listed in docs/reference/cli.md.
  • Coverage exists in tests/test_repository_recovery.py exercising both the inspection plan and the recovery flow; the suite passes cleanly.

The single supporting record is a [failure] on the original implementation task's contract prerequisites, not evidence of a defect in the shipped recovery code. With only one low-confidence evidence record and no reproducible fault in the current code, there is no defect to repair.

Disposition

  • Classification: low-confidence dream finding, not actionable — deliverable already present.
  • Code change: none in src/mac/repository_recovery.py or its CLI wiring.
  • Evidence gap: the finding lacks any record isolating a reproducible fault in the shipped inspect_stalled_finalizer_recovery / recover_stalled_finalizer path; its lone record is the original repair task's prerequisite failure. Reopen only if a concrete reproduction or a new fingerprint recurrence surfaces an actual defect.

References

  • src/mac/repository_recovery.py — inspect_stalled_finalizer_recovery and recover_stalled_finalizer implementations.
  • src/mac/cli.py — task recover-stalled-finalizer command wiring.
  • tests/test_repository_recovery.py — inspection and recovery coverage.
  • docs/archive/field-notes/crash-incident-finding.md — companion investigation-finding note for an unactionable low-signal repair candidate.