# Fast-task results summary
Latest graded run per task for each agent on the official 22-task suite
(`python ops/fast_task.py tasks`). Reward is binary host grade (`1` = pass, `0` = fail).
## Sources
| **malvin** | `_logs/log-FT-*` Evaluation blocks; artifacts under `~/.malvin_home/fast_task_results` with `malvin` docker cmd | 2026-08-06 (full suite) |
| **prime** | `_logs-prime/log-FT-*` and `_logs-prime/_board.txt`; `agent_name=prime` artifacts | 2026-08-06 (full suite; FT-01/FT-24 from earlier that day) |
| **cursor** | `agent_name=cursor` artifacts (`cursor-agent --force -p`); no `_logs-cursor/` board present | mostly 2026-07-31; FT-01 also re-run 2026-08-06 |
Dry-run / skipped metadata entries were ignored. For each (agent, task), the newest non-dry graded artifact wins.
## Overall
| malvin | 19 | 3 | 0 | 86.4% |
| cursor | 17 | 5 | 0 | 77.3% |
| prime | 17 | 5 | 0 | 77.3% |
## Failures
- **malvin** (3): FT-24, FT-27, FT-28
- **cursor** (5): FT-25, FT-27, FT-28, FT-29, FT-32
- **prime** (5): FT-03, FT-08, FT-25, FT-28, FT-35
## Agreement
- All three pass: 13 tasks — FT-01, FT-05, FT-09, FT-12, FT-13, FT-15, FT-17, FT-20, FT-26, FT-30, FT-31, FT-33, FT-34
- All three fail: 1 task — FT-28
- Split outcomes: 8 tasks
| FT-03 | 1 | 1 | 0 |
| FT-08 | 1 | 1 | 0 |
| FT-24 | 0 | 1 | 1 |
| FT-25 | 1 | 0 | 0 |
| FT-27 | 0 | 0 | 1 |
| FT-29 | 1 | 0 | 1 |
| FT-32 | 1 | 0 | 1 |
| FT-35 | 1 | 1 | 0 |
## Per-task board
| FT-01 | 1 | 1 | 1 |
| FT-03 | 1 | 1 | 0 |
| FT-05 | 1 | 1 | 1 |
| FT-08 | 1 | 1 | 0 |
| FT-09 | 1 | 1 | 1 |
| FT-12 | 1 | 1 | 1 |
| FT-13 | 1 | 1 | 1 |
| FT-15 | 1 | 1 | 1 |
| FT-17 | 1 | 1 | 1 |
| FT-20 | 1 | 1 | 1 |
| FT-24 | 0 | 1 | 1 |
| FT-25 | 1 | 0 | 0 |
| FT-26 | 1 | 1 | 1 |
| FT-27 | 0 | 0 | 1 |
| FT-28 | 0 | 0 | 0 |
| FT-29 | 1 | 0 | 1 |
| FT-30 | 1 | 1 | 1 |
| FT-31 | 1 | 1 | 1 |
| FT-32 | 1 | 0 | 1 |
| FT-33 | 1 | 1 | 1 |
| FT-34 | 1 | 1 | 1 |
| FT-35 | 1 | 1 | 0 |
## Agent wall time (seconds, latest run)
| malvin | 216.6 | 97.1 | 517.0 | (none) |
| cursor | 90.0 | 22.9 | 396.4 | (none) |
| prime | 143.7 | 27.8 | 605.0 | FT-08 |
Detail (latest run seconds):
| FT-01 | 142.9 | 37.0 | 39.8 |
| FT-03 | 162.0 | 31.5 | 53.2 |
| FT-05 | 124.9 | 37.0 | 34.0 |
| FT-08 | 253.5 | 41.4 | 605.0 |
| FT-09 | 171.9 | 22.9 | 36.8 |
| FT-12 | 133.9 | 27.6 | 29.5 |
| FT-13 | 148.2 | 38.2 | 57.1 |
| FT-15 | 127.3 | 42.1 | 46.8 |
| FT-17 | 168.3 | 34.5 | 58.2 |
| FT-20 | 112.1 | 27.5 | 27.8 |
| FT-24 | 517.0 | 396.4 | 368.4 |
| FT-25 | 279.5 | 151.1 | 69.0 |
| FT-26 | 148.6 | 51.1 | 71.7 |
| FT-27 | 266.9 | 87.8 | 253.3 |
| FT-28 | 348.0 | 86.5 | 215.5 |
| FT-29 | 234.4 | 120.4 | 77.7 |
| FT-30 | 437.8 | 106.7 | 73.0 |
| FT-31 | 124.7 | 71.4 | 252.0 |
| FT-32 | 420.3 | 277.8 | 266.8 |
| FT-33 | 97.1 | 39.5 | 307.6 |
| FT-34 | 154.2 | 46.2 | 61.0 |
| FT-35 | 192.6 | 204.5 | 158.2 |
## Notes
- Official suite size: 22 tasks.
- Workspace `fast_tasks/*/reward.txt` files can be stale relative to the suite boards; prefer `_logs*` Evaluation blocks and `fast_task_results/*/metadata.json` + `reward.txt`.
- Cursor has no `_logs-cursor/` directory in the repo; the cursor column is from graded artifacts only. Most cursor latest runs are from 2026-07-31, so that column is older than the malvin/prime 2026-08-06 boards.
- Prime FT-08 latest run timed out at the 600s agent limit (`agent_seconds` ≈ 605) and graded reward 0.
- Malvin fails match the 2026-08-06 `_logs` board: FT-24, FT-27, FT-28.
- Prime fails match the 2026-08-06 `_logs-prime` board: FT-03, FT-08, FT-25, FT-28, FT-35.