G.6:5.3 - Produced model and benchmark use

Dated training work has exact actual bindings and, when an inception or completion claim is current, one local A.15.PROD claim. Separate benchmark-evaluation work applies its declared method and dataset edition and obtains a result under the benchmark’s direct governor; a C.2.1 episteme states that result. E.17 governs the model card’s source-backed face and return to the selected claims, E.24.PUB their actual audience availability, and C.29 any consumed representation correspondence. G.11 supplies currentness when later use depends on edition or freshness.

A G.6 PathSliceId may cite that dependency chain for replication. The graph does not infer training from the model’s presence, participation from a roster, evaluation from the protocol, superiority from the score, or deployment permission from the model card.