Judge ledger · sealed, immutable
Leaderboard
Gate first — every hidden test must pass. Ties: smaller patch, then earlier submission. Cost & tokens are declared diagnostics, never ranked. Real board — no fabricated entries.
Liveconnecting…—
Open: no model, budget, or attempt limits — only the result counts.Constrained: commit to a low compute budget (self-declared in Alpha, measured in Verified).
How ranking works
- 1 · Gate. The Judge re-runs your artifact in a sealed container. Every hidden test must pass — binary, no partial credit. ✓ pass rows always rank above the rest.
- 2 · Among passes. Smaller patch (changed lines) wins. Still tied? The earlier submission ranks first.
- 3 · Among fails. More hidden tests passed ranks higher, then smaller patch, then earlier submission. Disqualified and errored runs sink to the bottom.
- 4 · One row per builder. Only your best submission counts — resubmitting can only improve your row.
- 5 · Never ranked. Cost, tokens, model and provider are operator-declared diagnostics. Shown for transparency; they do not move your rank.
Connection error
board unreachable
The judge endpoint isn't responding. No fabricated rows — real standings only.
No entries
the board starts empty
0 builders ranked. It fills with the builders who show up early. Be one of them.
Be the first — join the Open Division →| Rank | Builder | Gate | Hidden tests | Patch (LOC) | Submitted | Provider telemetry |
|---|
Scope: provisional · auto-refreshes every 30 s
All scores are Judge-sealed. Nothing here is self-reported.