Note The agent test: fifteen tasks, every exchange of 645 runs, and a runner for your own agent.

Note Filed demonstration receipts. Proposals cite this set.

Note The last request that stands and the nearest one past it, for every model the verifier serves.

Note Beside each open dataset a model was tried on: its predictions on the held-back runs, each with its cryptographic proof, crediting the people who made the measurements.

Note Tasks about real things with a check a program runs and no judge: the agent's own request, sent to the verifier again.