Goal
Eliminate recurring-task failures where tasks remain assigned to a worker or scheduler even though no live process exists. The fix should make phantom-run detection and cleanup reliable for recurring engines so tasks do not accumulate false failures or get stranded between scheduler and worker state.
Acceptance Criteria
☐ Root cause identified for phantom running task failures on recurring Senate engines
☐ Cleanup or reconciliation logic updated so the affected tasks return to a consistent open/running state automatically
☐ Verified against at least Agent activity heartbeat, Orphan coverage check, and World-model improvement detector
Approach
Review recent run history for the phantom-running recurring tasks and compare task state vs task_runs state.
Inspect Orchestra supervisor/reaper reconciliation paths that clear active_run_id, assignment state, and scheduler-owned runs.
Implement and verify a fix with bounded reproduction or targeted unit coverage.Dependencies
- Orchestra recurring-task lifecycle and task-run reconciliation logic
Dependents
- Senate recurring drivers that currently report phantom workers instead of doing work
Work Log
2026-04-26 06:46:33Z
- Spec created during engine-failure triage task
9e45545a-0eeb-4698-9d79-ffda3f456b45 for follow-up implementation.