Hand in your policy.
Then the paper is drawn.
One exam per week for any brain in the slot. You commit before the seed exists; an Ethereum block mined after the cutoff picks it; the replay grades everyone on the same world. Win clean — and anyone can recompute your row, to the egg.
loading the week…
Four steps, all public.
COMMIT
Lock your entry: a brain and the sha256 of your ranked mutation policy, signed with your wallet. The server stores only the hash.
DRAW
After the cutoff the exam seed is derived from the latest Sepolia block mined after it — public, verifiable, unknowable at commit time.
REVEAL
Publish the policy. The sha256 must match the commitment, or the entry is voided. No editing after seeing the paper.
GRADE
The server replays every revealed entry on the same seed through the same world.js the browser runs, and ranks: eggs → survived generations → earliest commit.
Full record, including why policies are ordered trait lists instead of code (no untrusted execution on the grading server): docs/weekly-race.md. Honest boundary: no week has been graded yet — until one has, this is a protocol with receipts, not a track record. The grading operator is trusted to run the replay honestly; that trust is auditable (you can re-run every entry) but not yet removed. On-chain grading is the follow-up.