Validation-Frontier Representation Selection under Constrained Observation

Reproducibility artifact accompanying “Validation-Frontier Representation Selection under Constrained Observation.” The archive contains benchmark reproduction and validation scripts, public-dataset manifests, generated CSV outputs, verification records, and frozen evidence-lock materials supporting the reported representation-selection results under constrained observation. The benchmark evaluates state-representation choices across natural, low-resource, missingness, class-imbalance, and covariate-shift regimes using public scikit-learn bundled datasets and fixed target transformations. Internal legacy filenames are retained where necessary to preserve provenance, checksums, and the frozen evidence trail.