An assistant must refuse to answer when its approved corpus has insufficient evidence. Which two evaluation practices best test this behavior? Select 2 answers.

1. Assume a citation string proves the cited source supports every claim.
2. Measure supported-answer quality and inappropriate abstention alongside unsupported-answer rates.
3. Include questions whose correct behavior is abstention or escalation because the evidence is absent.
4. Remove all ambiguous and unanswerable questions from evaluation.
5. Reward every fluent answer even when no approved source supports it.