ADEPT Unifies Five Families of Deep-Learning Test Adequacy Metrics Under One YAML-Configured Workflow

arXiv 2608.12144·low signalXBlueskyLinkedInCopy link

A decade of DL test adequacy metrics — neuron coverage, surprise adequacy, input distribution coverage, boundary coverage, and source- and model-level mutation score — shipped as independent research prototypes with incompatible installs, preprocessing, and execution workflows, making them near-impossible to reproduce or compare. ADEPT integrates all five under a consistent execution workflow with a template-based metric interface and defined extension points for adding new metrics. It adds YAML-based configuration management, preprocessing-cache reuse, and structured result reporting, positioning it as infrastructure rather than a new metric.