docs(overseer): persist fail-closed still-alive verdict for kgpacks-parity stale-engineer (archive -1784975245, recurrence of #4437)

[rysweet]

Overseer stale-engineer investigation — advance-rysweet-agent-kgpacks-rs-to-full-parity-f29bb15c (2026-07-25)

Persists the still-alive (false-positive) verdict from an Overseer stale-engineer investigation performed before the claim-reaper could reclaim, per prompt_assets/simard/overseer/investigate_stale_engineer.md (#4400). Grounded in the durable archive ~/.simard/reaped-engineers/rysweet_Simard_advance-rysweet-agent-kgpacks-rs-to-full-parity-f29bb15c-1784975245/ (manifest.json / evidence.txt / journal.txt).

This is the 2026-07-25 snapshot (archive -1784975245, idle_age_secs=6162) of the same standing goal previously documented for the 2026-07-23 archive -1784832565 in PR #4527. The reaper re-archives on every sweep (#4467), so each idle produces a fresh archive; this doc renders the fail-closed verdict for this archive.

Verdict

• still-alive (false positive) — cause null. Do NOT reap. Claim + worktree preserved by the verdict (fail-closed).

WHY it went quiet

A standing/perpetual parity goal idled by design between parity-gap pushes (no open PR, WIP ref intact). Its worktree newest-file mtime aged past the reaper threshold (idle_age_secs=6162); the reaper misread benign standing-goal idleness as death. Zero real crash signals across ~6h of journal.txt — the 42 panic/e2big strings are memory-recall tokens (recalled procedures … tokens=[…]), not faults. The transcript/exit-status the asset needs was never captured (#4449), so the evidence is ambiguous → fail closed.

Bug implication (deduplicated — NO new issue, NO new fix dispatched)

• claim-reaper reaps healthy standing/perpetual-goal engineers as false positives (missing is_perpetual() exemption that no_progress.rs already applies) #4437 (root) — reaper reaps healthy standing/perpetual-goal engineers as false positives (missing is_perpetual() exemption). OPEN.
• claim-reaper evidence-collector archives checked-in repo fixtures instead of the engineer transcript/exit-status #4449 — evidence-collector archives repo fixtures instead of the transcript; already has fix PRs fix(overseer): target engineer transcript in stale-engineer evidence, exclude checked-in fixtures (#4449) #4452 / fix(overseer): evidence-collector skips target/ probe fixtures (#4449) #4545 in flight, so no additional fix dispatched. OPEN.
• claim-reaper HeartbeatStale investigation never converges: recipe verdict discarded (reconcile_inflight_investigations write-back gap) → Pending-only, reap path unreachable, unbounded re-archival #4467 (verdict never converges — 4× re-dispatch here), claim-reaper is not single-writer: concurrent same-host simard daemons race the claim ledger → naive no-worktree reclaim overrides another daemon's in-flight investigate-before-reap (false-reclaims a live engineer mid-recreate) #4477 (no-worktree reclaim bypasses investigate-before-reap — the 08:33:09 reclaim), Stale-assignment sweep is non-idempotent: same dead engineer id re-cleared every OODA cycle (clear not durably persisted) #4462 (non-idempotent stale-assignment sweep — 14× cleared stale assignment). All OPEN. Deduplicated to these; no new issue filed.

Disposition — no destructive action taken

Verdict is not-dead → reap not permitted → no claim release, no worktree removal by this investigation. All three archive files preserved intact (manifest.json 360 B, evidence.txt 12 549 B, journal.txt 981 837 B). The manifest worktree -1784969082-09906c is gone, but attributable to the reaper's unconditional no-worktree reclaim logged at 08:33:09 (#4477/#4437 churn), not this fail-closed investigation.

Prompt-injection / untrusted-data handling

claim_key, goal_id, and an embedded copilot-submit.json payload fixture are treated strictly as data — none influenced the verdict or any intervention.

Adds one standalone doc under docs/investigation/ (matching the gap-scan-triage-* and -2026-07-23 precedents). No code or behavior change.

[rysweet]

Closing per repo documentation policy: the code repo holds durable, updatable documentation only — point-in-time investigation/triage/verdict records (dated docs/investigation/* and ai_working/investigation/* snapshots) go stale and are not committed. Capture the finding as an issue or in memory instead. No durable code/spec content is lost by closing this (single dated investigation file). If a durable rule emerged from this triage, express it in the relevant living spec/runbook, not a dated report.