Run a local exhibition

Produce and independently verify a deterministic replay artifact with no account or credentials.

Built-in policies only

The local runner currently compares bounded built-in policies. Arbitrary custom adapters and official evaluation are not implemented.

Install and build

terminal
pnpm install --frozen-lockfile
pnpm --dir packages/runner build

Run 100 paired episodes

terminal
node packages/runner/dist/cli.js run   --game dice   --episodes 100   --seed tutorial-v1   --output .degeneval/tutorial

Verify the artifact

terminal
node packages/runner/dist/cli.js verify   .degeneval/tutorial/result.json

The verifier checks the artifact schema, byte digests, deterministic replay, score conservation, and audit chain. Full-disclosure local seeds make the artifact reproducible but unsuitable for official hidden-input evaluation.

Understand evidence and proofs →