What problem does it solve? Research artifacts often contain numeric claims that drift from their underlying evidence, and manually cross-checking every value, unit, and statistical context is slow and error-prone. This Skill systematically compares one numeric claim against its linked Evidence Bundle entry and reports whether the evidence actually supports it. ## Core Features & Use Cases - Claim-to-Evidence Comparison: Checks value, unit, range, sample size, population, subgroup, endpoint, and statistical context of a claim against its [evN] Evidence alias. - Structured Finding Codes: Returns concrete COMPUTATION_* codes such as COMPUTATION_EVIDENCE_VALUE_MISMATCH, COMPUTATION_EVIDENCE_SCOPE_MISMATCH, COMPUTATION_EVIDENCE_INTERPRETATION_OVERREACH, or COMPUTATION_EVIDENCE_INSUFFICIENT. - Bounded Review Scope: Reviews exactly one claim per run on a locked Artifact, returning INCONCLUSIVE when evidence lacks context rather than misreporting it as a value mismatch. - Use Case: In a Deep research pipeline, before citation source verification, run this review on a claim like "treatment reduced mortality by 23% [ev3]" to confirm the Evidence Bundle entry actually reports that figure for the same population and endpoint. ## Quick Start Review the numeric claim marked with [ev3] in the locked Artifact against its Evidence Bundle entry and return the structured computation finding.
