Track Model Capabilities by Evidence

Build a task-specific capability evidence dossier tied to an exact model and version. Use representative tests, a fair baseline, uncertainty, failure slices, total cost, and change control to set limited use boundaries and hand reproducible evidence to later decision stages.