Interface ContainmentVerdict

Fired on both.

Prompts present in BOTH inputs — the only ones a comparison can speak about.

weakOnly.length === 0 — the weak model never fired where the strong did not.

Fired on neither.

Fired on the STRONG model but not the weak one. EXPECTED, and not a defect: it is the weak model under-selecting, which is the known reason for the model floor. Reported separately so the two are never conflated.

Fired on the WEAK model but not the strong one. Each is a counterexample to "the weak model is a floor" — the whole point of the comparison.