Replication package: Separating update benefit from verification value in test-time adaptation

Replication package for Separating update benefit from verification value in test-time adaptation. The package provides experiment runners, configurations, frozen decision rules, analysis and figure code, paired model outcomes, adaptation-state records and physical serving logs. The studies separate fixed-payload receiving-distribution effects, probe-channel effects on fixed outcomes, original-image-separated action assessment, and measured whole-policy utility. Three experiment-record archives retain the initial experiments, the expanded controls and serving studies, and the final registered mechanism/continuation studies. A companion README gives extraction order, reproducibility commands, data roles and the contents of each file. The retained serving evidence comprises 126 physical two-device trajectories; requested-policy aliases remain identified in the data. Public image datasets and pretrained checkpoints are obtained separately from the upstream sources listed in the documentation and retain their original terms. The archived image files are author-generated plots and PDF previews.