workflow-fix: treat in-flight Batch/long-analysis phases as healthy-slow
kind: infra#wf-fix#wf-fix-fp:4beff16cbb9b
track:Highlight text on any card (body, plan, or an event) to anchor a comment, or leave a whole-task comment here. Mention @claude to summon a reply.
No comments yet.
Overview / Motivation
Auto-filed from the predictor-line (#742/#761/#763) infra-friction retrospective (subagent, 2026-06-30). Related in-flight: #770 (RunPod no-port wedge billing), #681 (data-disk bind cutover), #667 (GCP frozen-phase), #769 (CUDA-frag OOM).
Goal
Treat an in-flight Anthropic Batch / long off-pod-analysis phase as healthy-slow: let the hourly [poll-tick] PROGRESSING markers (or a Batch-in-flight heartbeat) reset the staleness clock so a legitimately-slow phase is not respawn-eligible.
Workflow gap
- Bug observed: The staleness heuristic respawned healthy #761 twice and flirted with respawning #742's healthy 8h Anthropic Batch phase (0 errored, steady throughput) because long phases emit few markers between poll-ticks.
Scope / surfaces
- Target: scripts/autonomous_session_watch.py
Constraints / invariants
- Workflow-surface only; ruff + workflow_lint pass; add a test where practical.
Provenance
- workflow_fix_target: scripts/autonomous_session_watch.py
- fingerprint: 4beff16cbb9b
Raised by the predictor-line infra retrospective (2026-06-30).
Activity