result-reliability-checker

Provide the paper, abstract, or methods and results text and request a reliability audit. The Skill reconstructs the evidence chain behind each claim, checks design, bias, statistics, and validation, then assigns a per-claim reliability judgment from high to low.

The audit checks training and test separation, cross-validation discipline, tuning isolation, sample size versus model complexity, and temporal or information leakage. Suspiciously high metrics relative to sample size and design trigger explicit caution.

Internal validation uses splits or resampling within one cohort and only supports internal consistency. External validation tests the finding in a distinct cohort, dataset, or center, providing stronger but still context-dependent evidence of generalizability.

No. Patient-specific clinical decision support is explicitly out of scope, and the Skill responds with a redirect. It only audits whether reported research results are reliable enough to treat as evidence.

No. The Skill's hard rules state that statistical significance never equals reliability and high AUROC or accuracy does not equal robustness. Calibration, validation strength, bias control, and claim discipline must all support the conclusion.

Missing methods, statistics, or validation steps are labeled as unresolved rather than filled in. The Skill never invents study features, references, PMIDs, or DOIs, and states clearly when judgments rest only on user-provided text.