Policies define quality, security, and performance rules applied during test judging.

Policies are attached at the bot level (default policies) and/or per test case. Both are merged during evaluation.

Rule types: Threshold (min score per dimension), Metric (automated checks like latency and PII detection), Evaluation (natural-language criteria evaluated by the AI judge).

When multiple policies apply, their rules are combined. Stricter thresholds take precedence.