Interface NodeTiming

False when the value was produced this call.

Still meaningful when pulled is false: a node left clean by an earlier request genuinely holds a cached value, this request just had no reason to read it.

Upstream node ids, in the op's declared input order. Raw source columns are named by column, so an entry here is either an id in RunResult.nodes or a column of the bound series.

This is what makes the response a graph rather than a list. A caller cannot derive it: the edges live in the specs, and turning a spec into an id means reimplementing specId's canonicalization — which is exactly why a selector takes an inline spec rather than an id string. Added for M4's pipeline view.

Milliseconds attributable to this node, to 3 decimal places. Zero when not pulled.

Whether this request actually read the node's value.

A plan may resolve specs nothing selects. They are compiled and they are part of the pipeline, but no value was pulled through them, so ms is zero and says nothing. Reporting only the pulled subset made nodes a half-truth — the M4 pipeline view drew a plan with whole branches missing, because the request had not asked for them.

The caller's name for this node's position, when the request was written with slots. Stable across a param edit, unlike id — which is the whole point of having both ([PND-PROCSLOT]).