Recovery-First Metacognitive Science: A Preregisterable Measurement-Standard and Claim-Licensing Protocol

Metacognitive learning studies frequently interpret changes in meta-d’, M-ratio, calibration error, or confidence bias as evidence of enhanced introspective sensitivity. Three failure modes can render such interpretations invalid: (1) distributional misspecification (equal-variance SDT assumed when unequal-variance structure holds), (2) hierarchical inference-geometry pathologies (e.g., Neal’s funnel under centered parameterization in low-signal regimes), and (3) cross-model non-identifiability, whereby distinct generative architectures produce statistically indistinguishable predictions under typical constraints. MIPU (Minimal Irreversible Predictive Update) and MAP* (Maximal Anticipatory Pattern) are defined as constructs whose measurement requires validated engines. A four-gate recovery-first pipeline is proposed: (G0) inference viability, (G1) parameter recovery, (G2) model recovery / confusion-space mapping, and (G3) construct validity. The near-term deliverable comprises preregistered simulations that map the computational cliff under centered versus non-centered parameterization and quantify cross-model confusion, thereby determining when mechanism claims become identifiable rather than underdetermined. This protocol offers a transdisciplinary template for rigorous latent-variable inference, with direct relevance to metascience, artificial intelligence interpretability, clinical assessment, and educational measurement.