Step Catalog

The Monte Carlo step catalog is the single core-side registry for built-in operation metadata. It drives spec compilation, graph validation, and the GUI step palette without depending on the [ui] extra.

FieldSpec

@dataclass(frozen=True)
class FieldSpec()

One configurable field exposed by a step.

Fields:

Name	Type	Description
key	str	Runtime kwarg name.
label	str	Human-readable label.
type	str	UI/editor field type.
default	Any	Default value.
required	bool	Whether the field is required.
options	tuple[str, ...]	Allowed options for select-like fields.
minimum	float \| None	Optional numeric lower bound.
when	tuple[str, ...]	Conditional display metadata.

Methods:

FieldSpec.to_dict() -> dict[str, Any]

StepDefinition

@dataclass(frozen=True)
class StepDefinition()

Everything the library knows about one built-in Monte Carlo step kind.

Fields:

Name	Type	Description
step_type	str	Serializable step kind.
title	str	Display title.
default_name	str	Default runtime step name.
description	str	Short user-facing description.
op_role	Literal["datagen", "filter", "transform", "terminal"]	Graph role.
factory	Callable[..., MCStep]	Step factory used by build(...).
fields	tuple[FieldSpec, ...]	Declared configurable fields.
compile_params	Callable \| None	Optional parameter normalization hook.

Properties:

Name	Type	Description
is_terminal	bool	True for test and regression terminal steps.
is_transform	bool	True for transform steps.
category	str	GUI grouping: "core", "tests", "regressions", or "transforms".

Methods:

StepDefinition.catalog_entry() -> dict[str, Any]
StepDefinition.build(name: str, params: dict[str, Any], dgp: SolvedModel | None) -> MCStep

Registry Objects

Name	Type	Description
STEP_CATALOG	dict[str, StepDefinition]	Built-in step definitions keyed by step_type.
DATAGEN_STEP_TYPES	frozenset[str]	Datagen kinds: simulation and raw data.
TRANSFORM_STEP_TYPES	frozenset[str]	Built-in transform kinds.
TERMINAL_STEP_TYPES	frozenset[str]	Test and regression terminal kinds.
INPUT_SOURCES	list[str]	Source channels diagnostic/regression steps may read.
FILTER_SOURCES	set[str]	Source channels produced by the filter step.

catalog_payload() -> dict[str, Any]

Return a JSON-like payload of catalog entries for UI consumers.

Custom steps

STEP_CATALOG only describes built-in operation kinds. Bundle-safe custom operations use the reserved custom step kind and are restored through bundle resources rather than through catalog dispatch.