Scored on
physics,
not on
execution.
Hidden split · 210 tasks · published 21 Oct 2026
ff5cc6170cca8fb206adab930ccc097fc790a9f09446ccf2f56e052136c348c0
accepted.
Ran: 28 of 28.
Two splits. One is sealed.
The dev split is yours on request. The hidden split is sealed by its hash before any agent is scored.
Tasks whose reference answer the gate does not accept (feasibility) are excluded before scoring, and the excluded task ids are published with the reveal.
28Dev tasks
210Hidden tasks Committed 3 Oct 2026
Scored on a re-solve.
The agent's final case is copied and solved again. Executing alone scores nothing.
The leaderboard: each agent's executed, accepted and silent-error rates per family, with their intervals.
#environment
courant-env
The generator and the gate as a training environment, with a binary reward.
- An ungameable rewardPass or fail, on a re-solve
- The licenceFor training; never the hidden rounds
#evaluations
Private evaluations
Your agent on a fresh hidden draw, scored on re-solves.
- A private evaluationSilent-error rates, with intervals
- Signed and checkableEd25519
- Isolation and deletionIts own runner · a deletion log
