Module score 

2026-09-26: The serve matrix’s bars and tally, as pure functions of the round outcomes.

Owner: bench, serve matrix. Invariants:

• No I/O.
• An identity, codegen or tool-call signal that is NotRun fails its bar, and a planned round with no result fails as no-result.

Structs

RoundResult

2026-09-26: One planned round’s result.

Signals

2026-09-26: What one booted round measured.

Tally

2026-09-26: The whole matrix, scored.

Enums

Outcome

2026-09-26: What became of a planned round.

Signal

2026-09-26: One probe’s answer. NotApplicable passes only the tool-call bar, where it means the server refused the tools request with a 4xx; on the identity and codegen bars it fails.

Constants

COHERENCE_MIN_PASS

2026-09-26: Coherence probes that must pass. bars caps it at the number run, so with fewer probes every one must pass.

TPS_TOLERANCE

2026-09-26: Fraction below a stored baseline tok/s that counts as a regression.

Functions

tally

2026-09-26: Score every planned round against its result. Walks the plan, not the results, so a planned round with no result fails as no-result.

verdict_text

2026-09-26: The verdict sentence: coverage, skips (up to three named), the filter count, and each failing round’s bars.