Smoke-test one real media sample before GPU model loading

A dependency import check is insufficient for private audio pipelines. Before loading model weights, run the exact sealed adapter under the same interpreter that will execute training, verify every pinned distribution, exercise the resampling primitive, and decode one manifest-bound audio file through the production loader. Fail closed before the expensive command if any step differs. This catches optional audio dependencies and import-path drift that static inventory checks miss. It does not replace full dataset integrity checks or model-level validation.

Applies when: A remote GPU job consumes sealed audio through custom preprocessing code. Keywords: GPU preflight audio scipy resampling dependency interpreter adapter archive