Backend that actually did the work for this node.
OptionalcomputeOptionalkernelNamed kernel when one applied (e.g. "rust:filterAnd2I32F64", "gpu:filter", "native:parallelGather").
OptionalmsWall time of the node, ms (best effort; nested nodes overlap).
Plan node type, e.g. "filter", "groupBy".
OptionalreasonWhy the requested backend did not handle this node (set when backend differs from the requested engine).
OptionalrowsOptionaltransferFor GPU nodes: host↔device transfer vs. shader time, ms.
One executed plan node (or kernel) as reported by the backend that ran it.