Skip to content

Native Qualification Evidence And Calibration

sergio velderrain edited this page Nov 12, 2026 · 1 revision

Native Qualification Evidence And Calibration

Every receipt binds the candidate/source digest, policy/grant digest, VM UUID, opaque capture UUID, model digest, frame hashes, watermarks, outcome classification, and monotonic events. It excludes credentials, raw capture bytes, host storage paths, and raw model prose. A correct task acknowledgement cannot substitute for correct state grounding; model-observed reading/limit, oracle-selected action, and server-accepted task result are scored independently. No performance threshold, production pin, or stable-release status follows from offline fixture tests or one successful trial.

Clone this wiki locally