Serialized fitted-transform parameters (means/stds, one-hot vocabularies, bin
edges, imputation fill values) produced by the sidecar at /train and
replayed at /predict. The concrete shape is sidecar-defined and opaque to
the orchestrator — it is stored verbatim on the trained model
(MLModel.FittedPreprocessing, plan §4.3) and round-tripped unchanged.
Serialized fitted-transform parameters (means/stds, one-hot vocabularies, bin edges, imputation fill values) produced by the sidecar at
/trainand replayed at/predict. The concrete shape is sidecar-defined and opaque to the orchestrator — it is stored verbatim on the trained model (MLModel.FittedPreprocessing, plan §4.3) and round-tripped unchanged.