# Benchmark Authoring

Benchmarks should assert more than recommendation IDs.

They can also describe expected repo maps, instruction scopes, observability
signals, and safe parallelization outcomes for representative fixtures.