Understand GPTNT¶

These pages explain how benchmark inputs become running experiments and recorded outcomes. Use the reference when you need an exact option, setting, state, or channel.

Follow the system from input to execution¶

1. Benchmark and player model explains the relationship among a model, configured player, running service, role, and benchmark outcome.
2. Roles, protocols, and capabilities separates role assignment, experiment rules, service matching, and recorded participant identity.
3. Manuals and rule seeds connects source pins, compiled handbook identity, protocol access, and the current rules boundary.
4. Experiment hierarchy traces a run manifest through suites, missions, specifications, attempts, sessions, records, and outcomes.
5. Runtime services assigns process, matchmaking, communication, game, model-call, liveness, and recording responsibilities.
6. Suites, revisions, and comparability explains which identities and digests determine whether results measure the same benchmark.
7. Results and provenance connects player records, terminal evidence, DuckDB, static outputs, and submission bundles.

Choose another depth¶