Metric 4 — right action name(s). null when the case expects no action.
Metric 3 — did the agent decide the right kind of thing?
Forbidden names the agent invoked anyway. Non-empty is always a failure.
Human-readable reasons, in the order they were detected.
Metric 5 — share of parameter matchers that passed. null when the case declares none.
Individual matcher outcomes, for reporting which parameter failed and why.
Overall pass — every applicable component clean.
Mean of the APPLICABLE components. See the note in evaluateDecision.
Per-component outcome, so §6.1's metrics fall out of the stored details.