Expand description
Agreement and cost aggregation for benchmark move selections.
Port of benchmarks/agreement.py. Rows are JSON objects so the bundle
schema matches the Python harness.
Functions§
- aggregate_
agreement - Aggregate exact-reference agreement by engine, config label, and phase.
- aggregate_
cost - Aggregate measured selection cost by engine and config label.
- observation_
key - The key an observation row would have (used for checkpoint resume).
- run_
agreement - Return one move-observation row per adapter, position, and seed run.
Type Aliases§
- Observation
Key - Identity of one observation run —
(position_id, engine, config_label, seed), matching Python’scheckpoint.observation_keytuple order.