Blind Prediction and Falsification in the OFT-Derived VGM Program: Immutable Pre-Test Contracts, Withheld Data, Decision Logic, and Failure-Closed Empirical Verdicts

Abstract

This preprint develops the natural methodological continuation of VGM XV, moving from conditional identifiability and inference to prospectively specified empirical testing. It defines how a declared VGM prediction, measurement response, nuisance treatment, and inference procedure should be fixed before access to withheld outcomes. The framework introduces timestamped analysis contracts, information-leakage controls, calibration and provenance requirements, predeclared decision rules, and claim-specific power assessments. It distinguishes parameter constraints from model-level evidence and separates compatibility, discrimination, rejection, and invalid test conditions. Particular attention is given to the logical scope of rejection: an adverse outcome can challenge the tested realization or parameter region without automatically refuting the entire OFT/VGM program. Sequential testing, multiplicity, robustness branches, and subsequent model revisions must remain traceable to their original contracts. The contribution is a methodological architecture for a future empirical implementation; no withheld dataset, experimental validation, detection, exclusion, or falsification result is reported.