Scores an observation against a case's expectation, resolving anyOf.
anyOf takes the best-scoring branch rather than the first passing one. Both give the same
pass/fail, but the score differs when every branch fails, and reporting the closest miss is what
makes a failure diagnosable: "expected chat OR action, got unparseable" is less useful than the
branch that came nearest.
Scores an observation against a case's expectation, resolving
anyOf.anyOftakes the best-scoring branch rather than the first passing one. Both give the same pass/fail, but the score differs when every branch fails, and reporting the closest miss is what makes a failure diagnosable: "expected chat OR action, got unparseable" is less useful than the branch that came nearest.