- Outcome
- FAIL19m
- Reward
- 0.0%
- Checks passed
- 19 / 20
- Runtime
- 19m 10s
- Harness
- Cursor CLI
- Task version
- v3.3.1
Checks
| Check | Result | Credit | Weight |
|---|---|---|---|
| baseline: a healthy tick collects every response, completes the run and advances the schedule | PASS | 100% | 1% |
| collection: a 429 or 503 streak that outlasts the collector's own retries still completes the run on the same tick | FAIL | 0% | 14% |
| collection: an empty model answer is asked again within the tick | PASS | 100% | 10% |
| collection: a rejected key (401) and an unknown model (404) are not requested again within the tick while a transient failure in the same run is | PASS | 100% | 12% |
| resume: a run left partial by a missing key has no edition and is resumed next tick, collecting only the missing responses | PASS | 100% | 7% |
| give-up: a partial scheduled run whose slot is 13 hours old gets its edition and the schedule moves on | PASS | 100% | 8% |
| give-up: a partial run whose pipeline was switched off gets its edition from the sweep inside the same window | PASS | 100% | 4% |
| liveness: the next tick never re-claims a run whose worker is still retrying, and still recovers one whose worker died | PASS | 100% | 18% |
| generation: an insight call that needs about 2.5 minutes is not abandoned | PASS | 100% | 12% |
| generation: a topic analysis that fails once is retried and the edition lands as a draft on the same tick | PASS | 100% | 7% |
| generation: an overall insight that fails once is retried and the edition lands on the same tick | PASS | 100% | 7% |
| preserve: the hourly job's scheduled runs (claiming, collection, schedule and edition) | PASS | 100% | 0% |
| preserve: one-off provider trouble the base already retried within the call (a timeout, a dropped connection, a server error) | PASS | 100% | 0% |
| preserve: one prompt's unreadable answer leaves its provider's other prompts collected | PASS | 100% | 0% |
| preserve: an edition whose overall insight keeps failing is not drafted without it | PASS | 100% | 0% |
| preserve: the hourly job's report sweep (editions owed, retried, reclaimed or left alone) | PASS | 100% | 0% |
| preserve: manual runs (POST /geo/workspaces/{workspace_id}/runs) | PASS | 100% | 0% |
| preserve: the edition API (/geo/workspaces/{workspace_id}/reports, /geo/reports/..., /geo/public/reports/{slug}) | PASS | 100% | 0% |
| preserve: the web app's run history (GET /api/geo/workspaces/{workspaceId}/runs) | PASS | 100% | 0% |
| preserve: the base's schedule and report-period tests (backend/tests/test_schedule_cadence.py) | PASS | 100% | 0% |