Assess research conclusions, evidence quality and methodological limits.
A reasoning toolkit for weighing scientific claims: how to read the evidence hierarchy from case reports up to systematic reviews; which biases each study design invites; where correlation and causation part company; how p-values and confidence intervals commonly get misread; and where conclusions typically overreach past their data. It teaches the methods too, so the critical eye becomes yours; on request it grades evidence quality with explicit reasons for any downgrade.
It appraises what you bring in — papers, news coverage, fetched pages, documents — and writes the appraisal as a workspace file or Note. A claim without support is recorded as 'evidence not supplied', an uncertain bias call stays flagged: the output is an inspectable assessment, never a one-word verdict.
Examples describe the request and the shape of the expected deliverable; they are not completed run results.
A headline claims 'study proves coffee extends lifespan' with the abstract attached: grade it against the evidence hierarchy and bias checklist — what do the paper and the coverage each say, and where is the overreach?
A graded appraisal: design, sample and controls, the claim gap between coverage and paper, uncertainties marked as such.
Pre-review this controlled trial protocol: likely bias channels, endpoint-choice problems and how far the conclusions can travel, point by point in the quality framework.
A point-by-point list, each item tied to a named bias or design issue with a suggested direction.
iOS · Android
iOS
Android
MIT
MIT (root LICENSE.md)
The pinned skill subtree and shipped references were audited; full applicable license and per-file adaptation/asset notes are preserved under references/licenses/ and references/PROVENANCE.md.
Publication commit 25e07dbeeb, audited source 84c524b6fb; catalog digest iOS 46196ba57ed7… · Android 16de89d019e2….
Version records ship with the signed catalog; the target and released versions are shown below.
