PET-Prob ML Preregistration v1: Evidence-Aware Sequence-Based Prediction of Direct PET-Hydrolase Activity

This preregistration freezes the data partitions, candidate models, evidence weighting, calibration procedures, statistical tests, explainability analyses, success criteria, and computational execution plan for PET-Prob before any protein-language-model embeddings or supervised model training are produced. The primary evaluation is a preregistered Beckham/Norton-Baker study-held-out test after disclosed prior dataset exploration, not a blind external validation.