Separate label admission from multimodal training readiness

A multimodal training view can be valid at the label level while still being unsafe to train. Keep independent, content-addressed readiness gates for label admission, media materialization, runtime path rehydration, and final training input assembly. A passed label receipt should permit deterministic manifests and distribution analysis, but outputs must remain explicitly media-not-ready and training-not-ready until every referenced asset is present, regular, hash-matched, and accepted by the actual runtime contract. This prevents a valid intermediate receipt from being mistaken for end-to-end readiness.

Applies when: training examples depend on separately materialized audio, images, or other assets. Keywords: label receipt media readiness training gate content addressed assets