Audit Evidence
Build a claim-to-evidence map from the supplied target. Plausibility is not
evidence, and absence of evidence is unverified, not confirmation.
Checks
- Verify every decision-relevant number and every quantitative or completion
claim PI intends to deliver against a specific file, line, log, or structured
output. For homogeneous supporting numbers or artifacts, inspect a
representative set covering the main types and boundary cases, expanding it
only when a discrepancy or shared dependency makes the remainder material.
- Verify that cited sources exist, support the exact proposition, and are not
overstated beyond population, method, or result scope.
- Verify that decision-relevant evidence existed before the decision or
revision it is claimed to support. A post-hoc citation does not retroactively
ground an earlier choice. Use artifact versions, task records, checkpoints,
diffs, and run metadata; if they cannot establish order, identify the exact
missing chronology evidence and mark the link
unverified.
- Reconcile conflicting Expert reports and distinguish direct evidence,
interpretation, assumption, and unresolved uncertainty.
- Check that report text matches the latest artifact revision and does not turn
limitations, missing checks, or qualified findings into unconditional claims.
- Record exact missing evidence and the likely owner; never fill gaps by
inference or by recomputing results.
For a bounded parallel review, use evidence-extractor with the target report
and its cited evidence packet. Ask for a compact claim map and candidate findings,
not a verdict.
1---2name: audit-evidence3description: Audit numeric, artifact, log, citation, and cross-report claims against inspectable evidence. Use for reports, syntheses, benchmark claims, external citations, or conflicting Expert outputs.4---56# Audit Evidence78Build a claim-to-evidence map from the supplied target. Plausibility is not9evidence, and absence of evidence is `unverified`, not confirmation.1011## Checks12131. Verify every decision-relevant number and every quantitative or completion14 claim PI intends to deliver against a specific file, line, log, or structured15 output. For homogeneous supporting numbers or artifacts, inspect a16 representative set covering the main types and boundary cases, expanding it17 only when a discrepancy or shared dependency makes the remainder material.182. Verify that cited sources exist, support the exact proposition, and are not19 overstated beyond population, method, or result scope.203. Verify that decision-relevant evidence existed before the decision or21 revision it is claimed to support. A post-hoc citation does not retroactively22 ground an earlier choice. Use artifact versions, task records, checkpoints,23 diffs, and run metadata; if they cannot establish order, identify the exact24 missing chronology evidence and mark the link `unverified`.254. Reconcile conflicting Expert reports and distinguish direct evidence,26 interpretation, assumption, and unresolved uncertainty.275. Check that report text matches the latest artifact revision and does not turn28 limitations, missing checks, or qualified findings into unconditional claims.296. Record exact missing evidence and the likely owner; never fill gaps by30 inference or by recomputing results.3132For a bounded parallel review, use `evidence-extractor` with the target report33and its cited evidence packet. Ask for a compact claim map and candidate findings,34not a verdict.
neuroaihub