scientific-critical-thinking

Evaluate evidence quality by assessing study design against the evidence hierarchy, then applying GRADE criteria: downgrade for risk of bias, inconsistency, indirectness, imprecision, and publication bias, and upgrade for large effects or dose-response relationships. The references/evidence_hierarchy.md file provides detailed criteria.

GRADE rates evidence as high, moderate, low, or very low quality. RCTs start as high and observational studies as low, then ratings adjust based on risk of bias, consistency, directness, precision, and publication bias. It guides how much confidence to place in estimated effects.

Detect p-hacking by checking for suspicious p-values clustering just below .05, undisclosed analyses, and missing preregistration. Detect HARKing by comparing published hypotheses against preregistered protocols and checking whether exploratory findings are presented as confirmatory.

Use this skill for understanding evidence quality, identifying methodological flaws, and teaching critical analysis. For writing formal peer review reports, the skill's own guidance directs you to the dedicated peer-review skill instead.

No, the analytical guidance works entirely offline. Only the optional figure generation through the scientific-schematics skill requires an OPENROUTER_API_KEY and outbound access to the OpenRouter API.