Context
During the PR #1090 review (post-install live-probe runbook row for the v4.4.51 merge-guard batch), two reviewers independently had to re-drive the full 29-probe harness to verify the recorded tally, because the probe scratchpad holds only harness.py — no persisted run output. Both re-drives reproduced 29/29, but the recorded tally itself was not artifact-reproducible without re-execution.
Proposal
For future post-install (and pre-merge) live-probes, persist the harness stdout alongside the harness, e.g.:
python3 harness.py 2>&1 | tee run.log
and reference the log's existence in the runbook row's harness-independence cell. This makes the recorded tally verifiable from artifacts alone (read run.log) while keeping re-drive available as the stronger check.
Provenance
Surfaced as the sole Future finding (devops-reviewer) in the PR #1090 three-reviewer review; review synthesis notes this gap is what forced both independent re-drives in that cycle.
Context
During the PR #1090 review (post-install live-probe runbook row for the v4.4.51 merge-guard batch), two reviewers independently had to re-drive the full 29-probe harness to verify the recorded tally, because the probe scratchpad holds only
harness.py— no persisted run output. Both re-drives reproduced 29/29, but the recorded tally itself was not artifact-reproducible without re-execution.Proposal
For future post-install (and pre-merge) live-probes, persist the harness stdout alongside the harness, e.g.:
and reference the log's existence in the runbook row's harness-independence cell. This makes the recorded tally verifiable from artifacts alone (read
run.log) while keeping re-drive available as the stronger check.Provenance
Surfaced as the sole Future finding (devops-reviewer) in the PR #1090 three-reviewer review; review synthesis notes this gap is what forced both independent re-drives in that cycle.