Operators

Deterministic and reshape operations

Use ordinary code for exact logic and semantic models only for ambiguous meaning.

Operation	Purpose
`Transform`	Map each row with a trusted Python callable.
`Filter`	Keep rows for which a callable returns true.
`Cluster`	Group values using a trusted pairwise predicate.
`Join`	Join two Spaces using a trusted predicate.
`Assign`	Create labeled relationships between Spaces.
`Matching`	Select a one-to-one matching.
`Relate`	Create labeled relationships within one Space.
`Sort`	Order values using a trusted comparison.
`Taxonomy`	Create broader/narrower links using a trusted predicate.
`Hierarchy`	Create a single-parent forest using a trusted predicate.
`Select`	Keep selected columns.
`Drop`	Remove selected columns.
`Flatten`	Explode a list-valued field into rows.
`Transpose`	Swap nested keyed levels.
`Union`	Combine Spaces and collapse equal keys.
`Distribute`	Route each row to one or more output Spaces.
`Apply`	Run a custom rows-to-rows reshape.
`Scatter`	Declare independent branches together.
`Barrier`	Wait for all preceding branches.
`Unfold`	Apply a transformation recursively under explicit bounds.

Worked reshape

Union(
    name="combine-and-deduplicate",
    input_spaces=("web-products", "store-products"),
    output_space="products",
    key="sku",
    merge_policy={"inventory": "sum", "tags": "union"},
)

Flatten(
    name="one-interest-per-row",
    input_spaces=("user-profiles",),
    output_space="user-interests",
    field="event_interests",
    key="user_id",
)

Union is the correct zero-cost exact deduplication operation. Use SemCluster only when equality itself requires semantic judgment.