Module score 

2026-09-26: Scoring for the contamination benchmark: classify every prompt at every rung and in the post-check against its own solo reference. The module reads no handle, network or clock; score_tests.rs tests it with in-memory outcomes.

Why more than “one solo reference, then diff each rung against it”:

1. A single reference cannot tell a concurrency defect from a prompt that is not reproducible alone. The reference runs twice; a prompt whose two solo runs disagree is AloneUnstable, is excluded from the per-rung comparison, and still fails the verdict under its own name.
2. State that survives the concurrent episode shows only in later solo work, so a solo post-check leg after the rungs classifies a divergence there as Persistent.
3. The driver primes every prompt once before any measured leg, so the measured legs start cache-warm unless a prime request failed (the driver then logs a warning). cached_prompt_tokens is not part of the comparison.

Owner: bench (contamination). Invariants:

• Each cell inserted into Score::cells increments exactly one of the class counters.
• verdict passes only when at least one cell was compared and every compared cell is Identical.

Structs

Legs

2026-09-26: Inputs for one scoring pass. Every field is recorded data.

Score

Enums

Class

2026-09-26: What happened to one prompt in one leg: a rung, the post-check, or (as "ref") its reference.

Functions

score

verdict

2026-09-26: Zero tolerance: any unmeasured, alone-unstable, contaminated, persistent or diverged cell fails, as does a score with nothing compared.