// request access
See FAST World run before you license it.
Tell us who you are and what you want to evaluate. We reply with the World Card and a time for a live walkthrough.
certified image 0.1.9 · 120 active tasks + 2 public samples · 7 sets + hard tier + 2 missions · frontier challenge tier
// a real run, not a mock-up
What the grader gate looks like.
Before a task ships, it has to score 0 when nothing is changed, 1 with the reference fix, and partial credit when part of the fix is missing. These are the recorded results for our first long-horizon mission, from its certification on image 0.1.6; on image 0.1.8 it re-passed the sealed gate (no-op 0, reference fix 1 three times). The walkthrough replays this live.
- task
- long-horizon mission (private evaluation)
- image
- channelforge-world:0.1.6 · linux/amd64
- harness
- Harbor 0.22.0 on our dedicated Linux run server
- recorded
- 2026-09-28 · about 11 minutes for the 4 gate trials
| TRIAL | WHAT IT DOES | M1 M2 M3 M4 · FINAL | SCORE |
|---|---|---|---|
| no-op | changes nothing | ····|· | 0.00 |
| oracle · run 1 | reference fix | ✓✓✓✓|✓ | 1.00 |
| oracle · run 2 | reference fix | ✓✓✓✓|✓ | 1.00 |
| oracle · run 3 | reference fix | ✓✓✓✓|✓ | 1.00 |
| oracle − fix A | one fix withheld | ✓✓·✓|· | 0.75 |
| oracle − fix B | one fix withheld | ✓✓✓·|· | 0.75 |
| oracle − fix C | one fix withheld | ··✓✓|✓ | 0.50 |
0 harness errors. Milestones are shown as m1–m4 because this is a private-evaluation task: what each one checks stays with us.