## Abandonment Watchdog — Automated Repair Task
Task **[Senate] Rigor score card — 8-dim evaluation per hypothesis/analysis (Alpha1 rubric)** (`0ce71340-e3e`) has been abandoned 24 consecutive times.
### Error Analysis
- **Dominant pattern:** rate_limit
- **Error breakdown:** {"rate_limit": 22, "exit_code_0": 2}
- **Recent errors:**
- `rate_limit_retries_exhausted:glm`
- `rate_limit_retries_exhausted:glm`
- `rate_limit_retries_exhausted:glm`
- `rate_limit_retries_exhausted:glm`
- `rate_limit_retries_exhausted:glm`
### Your Job
1. **Root-cause** the failure — read the task spec, check the code/DB/filesystem for what's broken
2. **Fix it** — apply the actual fix (code change, DB repair, stale worktree cleanup, whatever is needed)
3. **Reset the original task** — run `orchestra reset 0ce71340-e3e` so it retries with the fix
4. **Verify** — if possible, confirm the original task will succeed on next run
If the original task is **already done** (the fix is on main, the feature works), mark it done: `orchestra update --id 0ce71340-e3e6-4f7d-90e8-79dfdc684872 --status done`
If the task is **impossible** (bad spec, removed API, etc.), archive it: `orchestra update --id 0ce71340-e3e6-4f7d-90e8-79dfdc684872 --status archived`
### CRITICAL: You MUST complete this task when done
When finished (fix applied, original reset, or triaged), run:
```
orchestra task complete --id $ORCHESTRA_TASK_ID --no-sha-check --summary 'Fixed: '
```
Without this call, your work is invisible and the task loops.
$ORCHESTRA_TASK_ID is set in your environment automatically.