OptionalbinsNumber of bins to fit when op is bin. The sidecar fits the edges; absent ⇒ sidecar default.
OptionalcolSingle target column for column-scoped ops (e.g. impute, onehot).
OptionalcolsMultiple target columns for multi-column ops (e.g. standardize).
OptionalfillFill value when strategy is constant.
OptionalindicatorWhen op is impute: also emit a <col>__missing column (1 where the value was missing) before filling, so a missing value stays distinct from a true zero.
The transform to apply. Examples from the plan: impute, standardize,
onehot. Left as an open string union member so new sidecar-supported ops
can be added without a breaking change to this contract.
OptionalstrategyImputation strategy when op is impute.
A single preprocessing operation in the train-time preprocessing list. These are declarations of what was fitted; the fitted parameters travel with the model as its fitted preprocessing. Mirrors the sklearn fit/transform split executed in the sidecar (plan §6.2).