[rigel-mintaka]

What does this PR do?

A recurring cron job whose run is killed by a gateway shutdown no longer pauses forever. On mattmini, nemo-dev-oversight ended paused with "Canonical cron unknown" and missed about 100 runs silently.

Related Issue

RIG-8801 (Linear).

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• cron/jobs.py: a shutdown kill stamps shutdown_interrupted_execution with the run's execution id. New skip_unknown_occurrence books that slot as skipped, advances next_run_at, keeps the job scheduled, does not spend a repeat count, and lifts the recovery pause.
• cron/scheduler_authority.py: an unknown receipt follows one ordered, retry-safe path. If already booked, or its slot already completed (for example by the feat(cron): keep a run cut off by a gateway restart as pending replay (RIG-8084) #45 replay), the receipt is retired. If it matches the shutdown stamp, the slot is skipped. Anything else pauses once and sends one notice via _deliver_crash_failure.
• cron/executions.py, cron/scheduler.py: the execution id goes through the shutdown sweep and the skip is recorded in run history.

How to Test

1. tests/cron/test_unknown_outcome_pause.py: 11 passed. The first 3 cases failed before the fix ('paused' == 'scheduled', 0 == 1 notices). Also covered: both shutdown phases, replay ordering, crashes between writes, a missing ledger row.
2. scripts/run_tests.sh tests/cron -j 2: 1547 passed. The 2 failures are test_cron_script.py cases that already fail on rigel-release.

Checklist

Code

• Conventional Commits
• Only this fix
• Tests fail before and pass after

Documentation & Housekeeping

• cron-internals.md updated

For New Skills

N/A.

Screenshots / Logs

N/A.

Refs RIG-8801

Co-authored-by: Matt Wilkinson matt@rigel.build

[linear-code]

RIG-8801

[rigel-mintaka]

Review rounds (single review lens, pre-submit), final head 7a5ccaf:

Round	Finding	Action
1	medium: completed-execution early return moved below the owner fence	Fixed: restored above the fence; test added
1	open: replay finishing before recovery could be skipped again	Fixed: a completed slot retires the receipt; real-path test
1	open: crash between the jobs, ledger and journal writes	Fixed and tested
1	low: tests hand-set the unknown row	Fixed: uses mark_running_jobs_interrupted and give_up_execution
2	open: replay cut off a second time paused instead of retiring	Fixed: already-booked execution retires; test
3	low: completed-slot lift untested; half-recorded skip	Fixed and tested
4	high: staged skip_detail could leave the job paused after a failed booking	Fixed: replaced by one ordered, retry-safe recovery path (no staged journal state)
5	low (unverified reachability): shutdown-stamped unknown whose ledger row is pruned AND whose slot already completed would book a skip over a completed slot	Accepted: needs ledger pruning before recovery and a lost replay booking; the stamp is cleared by any later booked run

All-clear on 7a5ccaf. Tests: test_unknown_outcome_pause.py 11 passed; tests/cron 1547 passed, 2 failures shared with rigel-release.