fix: bound stale wakes for desk-parked backlog holds

[slee029]

Intent

Bound stale wakes for parked-hold rows: admit hold-kind parked into captain_call_stale_bound. Parked-hold rows get no stale bound because fm-captain-hold.sh open reads only hold-kind captain, so a bare stale wake fires on every pane-hash churn. Evidence and the smallest fix (from the referenced report): captain-held lanes (hold-kind captain) are already bounded by design - first sight alarms once per declaration (hold identity x status signature), then at most once per 4h re-surface cadence. Desk-parked backlog rows (hold-kind parked) are invisible to every wait bound, so a parked lane whose newest status line is not paused:/captain-held: (for example a resolved: close) gets a bare stale wake on every new pane-display hash, and an idle pane's footer refresh changes that hash every few minutes. The smallest fix is to admit parked holds into the existing bound family instead of inventing a new cadence: let fm-captain-hold.sh open (or a sibling read-only predicate) report hold-kind parked rows, and in captain_call_stale_bound treat them as bounded waits whose declaration scope is the parked row's identity x status signature, owed by the supervisor/desk rather than the captain, keeping first-sight alarming. The fix must not: change the away-posture captain-held silence (parked rows are not captain calls), absorb a live parked gate worker (the config-gated wedge-defer-parked-gate path keeps its own owner), or change unheld lanes, which must keep alarming on every new hash as the pinned tests require. Bounding by a closing verb such as resolved: alone was rejected because it would silence finished unheld deliveries. The pinned test should be extended to a parked fixture mirroring test_open_captain_call_bounds_stale_churn.

What Changed

• Add a park command that records a parked-hold occurrence, and let open --include-parked report parked rows without changing captain-only reads.
• Apply the existing first-sight and re-surface cadence to parked-hold stale wakes, while preserving away-posture captain-call silence and the live parked-gate path.
• Extend lifecycle and watcher tests for parked holds, re-parking, pane churn, and unrelated backlog edits; update the architecture and configuration docs.

Risk Assessment

✅ Low: The parked-hold bound is limited to open parked rows, preserves the existing captain-held and unheld paths, and the re-review found no substantiated new defect.

Testing

The hold lifecycle run printed passing parked-row admission and user-prose preservation checks before timing out, but the prior payload did not establish a live product result. The watcher run timed out before its parked scenarios; no live product evidence was captured.

• Live validation: ⚠️ inconclusive - 0 of 6 scenarios driven live against the product

Why the scenarios were not driven live: they exercise Firstmate watcher internals with no separate product surface, and driving parked-pane churn against the maintainer's live Firstmate home would disturb its supervision. Each scenario is covered instead by a regression test in tests/fm-watch-triage.test.sh or tests/fm-captain-hold-lifecycle.test.sh that fails before the fix and passes after, and upstream CI runs those suites.

Known limit: the first-sight guarantee for a re-parked row is scoped to holds made through bin/fm-captain-hold.sh park. A raw tasks-axi unhold followed by tasks-axi hold --kind parked bypasses the parked hold occurrence tracking, so that re-park's first stale sight may be absorbed within the four-hour re-surface window. Firstmate routes holds through the wrapper.

Scenario	Result	Live	Evidence
Opening a parked row requires --include-parked, and parking preserves user-written occurrence prose	⏸️ untested	no	The prior payload recorded test output but did not establish that this scenario was driven against the live product; a live product run is needed.
A parked row alarms on first stale sight, absorbs pane-hash churn, and alarms after the four-hour window	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
Releasing and re-parking through the hold wrapper alarms on the new hold’s first stale sight	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
An unrelated backlog edit does not restart a parked hold’s alarm cadence	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
Away posture bounds parked-row churn while retaining captain-held silence	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
A live parked gate and an unheld lane retain their independent stale-wake behavior	⏸️ untested	no	The prior watcher run timed out before these scenarios and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.

• Outcome: ⚠️ 2 warnings across 1 run (6m7s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 1 issue found → auto-fixed ✅

• ⚠️ bin/fm-captain-hold.sh:1885 - If an existing task body begins with Parked hold occurrence: planned work\n\n..., the first park call treats that user-written paragraph as a stamp and silently deletes it. Strip only a valid wrapper-generated occurrence marker.

🔧 Fix applied.
✅ Re-checked - no issues remain.

⚠️ **Test** - 2 warnings

• ⚠️ Live watcher behavior remains undemonstrated: the targeted watcher run timed out before reaching the parked-hold scenarios. A bounded isolated product run is still needed to verify the stale-wake cadence and its guards.
• ⚠️ live validation verdict: inconclusive (0 of 6 scenarios were driven live against the product); untested: Opening a parked row requires --include-parked, and parking preserves user-written occurrence prose, A parked row alarms on first stale sight, absorbs pane-hash churn, and alarms after the four-hour window, Releasing and re-parking through the hold wrapper alarms on the new hold’s first stale sight, An unrelated backlog edit does not restart a parked hold’s alarm cadence, Away posture bounds parked-row churn while retaining captain-held silence, A live parked gate and an unheld lane retain their independent stale-wake behavior
• Live validation: ⚠️ inconclusive - 0 of 6 scenarios driven live against the product

Scenario	Result	Live	Evidence
Opening a parked row requires --include-parked, and parking preserves user-written occurrence prose	⏸️ untested	no	The prior payload recorded test output but did not establish that this scenario was driven against the live product; a live product run is needed.
A parked row alarms on first stale sight, absorbs pane-hash churn, and alarms after the four-hour window	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
Releasing and re-parking through the hold wrapper alarms on the new hold’s first stale sight	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
An unrelated backlog edit does not restart a parked hold’s alarm cadence	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
Away posture bounds parked-row churn while retaining captain-held silence	⏸️ untested	no	The prior watcher run timed out before this scenario and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.
A live parked gate and an unheld lane retain their independent stale-wake behavior	⏸️ untested	no	The prior watcher run timed out before these scenarios and did not establish a live result; a longer isolated live test window or bounded scenario-specific runner is needed.

• timeout 110 tests/fm-captain-hold-lifecycle.test.sh
• timeout 175 bash tests/.fm-targeted-run.tmp.sh (temporary targeted watcher runner; removed afterward)

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

Pipeline status for head 55a75c3: the review, test and document steps all completed in this run (none skipped), along with lint, push and PR. The test step's live-validation finding was approved rather than skipped.

[slee029]

This PR comes from a fork, so its CI workflow run is waiting for maintainer approval (action_required, no jobs started). Could a maintainer please approve the workflow run so the checks can execute? Thanks.

[slee029]

The branch was re-validated and pushed, so there are new workflow runs waiting for maintainer approval (action_required, no jobs started). Could a maintainer please approve them so the checks can run? Thanks.