Raw Data

raw_data_step(
    name: str = "datagen",
    *,
    states: ndarray | None = None,
    observables: ndarray | None = None,
    n_exog: int = -1,
    raw: Mapping[str, ndarray] | None = None,
    observable_names: Sequence[str] = (),
) -> MCStep

raw_data_step wraps raw_data_datagen(...). It does not require a DGP model.

Accepted Shapes:

Field	Accepted Shapes	Description
states	`(T, n_state)` or `(n_rep, T, n_state)`	A 2D array is reused for every replication. A 3D array selects `arr[rep_idx]`.
observables	`(T,)`, `(T, n_obs)`, or `(n_rep, T, n_obs)`	A 1D observable path is reshaped to `(T, 1)`. A 2D array is reused for every replication.
raw values	`(T,)`, `(T, n)`, or `(n_rep, T, n)`	Extra raw arrays are selected with the same convention as observables.

State-Only and Observable-Only Runs

raw_data_step accepts state-only and observable-only payloads. A reference filter step requires observables, while tests using source="states" require states.