What problem does it solve? Research conclusions often rest on metrics that do not actually measure what the claim asserts, numbers that cannot be traced to a concrete run, or citations that were never verified. This Skill enforces a warrant check between every claim and its evidence so conclusions, metrics, and citations stay aligned. ## Core Features & Use Cases - Metric-claim warrant checks: For every metric, explicitly state what it measures, why it supports the claim, and what it does not support, catching metric substitution such as claiming robustness while reporting accuracy. - Traceable numbers and citations: Require every key number to carry a run record, log, or config path, and verify every citation for existence and retraction status before it enters the record. - Uncertainty and scope discipline: Attach variance, replication counts, and matched statistical tests to point estimates, and scope conclusions to the evidence's distribution with external validation for generalization claims. - Use Case: When writing the results section of an ML paper, use this Skill to verify that each conclusion names its metric's warrant, traces to a logged run, reports mean±std across seeds, and avoids out-of-distribution generalization. ## Quick Start Review my results section and check whether every claim is supported by evidence that actually measures what the claim asserts.
