Submissions stay hidden until the deadline. Then one batched AI inference ranks every answer against the rubric, and you finalize the winner.
Correctness 50%, clarity 30%, novelty 20%.
{ "winnerIndex": 1, "summary": "ok" }Set the rubric, deadlines, and reward. The reward is escrowed on-chain.
Participants submit only a hash of their answer. Nothing is public yet.
After the deadline, answers are revealed and verified against the hash.
One batched AI inference ranks every revealed answer against the rubric.
The owner ratifies the winner. The escrowed reward is paid out.
Commit-reveal keeps submissions private during the window, so no one can copy the leading answer.
Every answer is judged in a single on-chain LLM inference, not one call per submission.
Reward escrow, judging, and payout all happen on Ritual. No off-chain trust.
The AI recommends a winner; the bounty owner makes the final, on-chain call.
Escrow a reward, set the rubric, and let the commit-reveal flow keep it honest until the AI judges.