Visible Rigor Before Archive: A Technical Specification for Claim-Level Human–AI Review

This paper specifies the Falsifiability Sheet as a pre-archive, claim-level human–AI review protocol for evidence-limited scholarship. It addresses a specific integration gap: established fields possess research methods and peer-review traditions, but they do not always preserve one inspectable record linking an exact claim to its declared method, fixed evidence, reviewer fit, independent assessments, disagreement, decision boundary, and later revision. The protocol treats one bounded claim as the unit of review. A human reviewer, R1, is selected through a visible qualification gate covering expertise, method competence, independence, conflicts, identity verification, competence limits, and version control. An AI reviewer, R2, independently assesses the same fixed evidence packet under closed-evidence rules that prohibit outside searching, model-memory gap filling, self-generated evidence, and unilateral assignment of the final public outcome. After both reviews are locked, a comparison layer assigns Green, Yellow, Red, STOP, or INVALID/RERUN. Green permits provisional advancement as a reviewed claim record rather than declaring permanent truth. Yellow activates three fixed clarification questions identifying the unresolved claim component, the evidence or expertise required, and the narrowest defensible revision. Red preserves why the current claim version failed. STOP records that essential evidence is absent. A second human reviewer is triggered by documented disagreement, cross-domain complexity, competence limits, formal challenge, or new contradictory evidence. The paper defines the framework’s scientific-method foundation, reviewer and AI standards, decision rules, threat model, controlled EA1006 workflow demonstration, and future validation design. It establishes an implementation-independent technical blueprint for the Falsifiability Sheet. Broad effectiveness, error rates, reviewer-burden reduction, and cross-field performance remain empirical questions.