Your build starts here
Replace bracketed details with your own goal, context, and checks. Keep the first build small enough to inspect.
Review this answer-from-documents behavior for [task]: [current policy and sample outputs]. Use only the supplied reference material: [documents with stable identifiers].
Create an outcome policy for a supported answer, missing evidence, contradictory passages, and a question outside the supported scope. For each outcome, specify a useful short response and the next input or check that could resolve a gap. Do not treat every uncertain case as requiring the same blanket refusal.
Evaluate these samples against the policy. Identify unsupported claims and unnecessary abstentions separately, pointing to the relevant supplied passage or explaining the missing evidence. A confidence percentage alone is not evidence.
Return the revised policy, a small test set, and the concrete checks needed to compare behavior. Give concise rationales tied to the references; do not invent results for evaluations that have not run.Make this prompt yours
Use when incorrect confident answers and excessive non-answers both matter. Review real supporting passages rather than only citation formatting.
You might also hear
- abstention audit
- unknown policy
- source-supported answer
The docs behind it
These references explain the underlying concepts and APIs. The prompt templates and project plans are original examples to adapt to your own setup.
Research sources reviewed .
Editorial review · October 28, 2026
Supported: Original, unexecuted evidence audit. Allowing missing-evidence answers and verifying support agree with provider grounding guidance; refusal quality still needs task-specific evaluation.