Optional ReadonlyassembleWhether a columns selector also assembles a widened TimeSeries.
Defaults to true, which is right for an in-process renderer:
the chart layers take a series plus a column name, and assembling
once is cheaper than making every consumer do it.
Pass false when the consumer is across a wire. Assembly there is
pure waste — the series cannot be serialized, and the columns can,
so the receiving side rebuilds one with TimeSeries.fromColumns,
which adopts a Float64Array zero-copy. Measured at 1M rows,
appendColumn costs 7.6 ms for a gapless column and 22.4 ms for a
gapped one — and every rolling study is gapped ([PND-PROCCOL]).
ReadonlynodesOptional ReadonlyonOptional ReadonlyoutputsWhich slots to surface, under the caller's own names.
Purely a projection of nodes: every result here is produced by a node the request declared, so nothing is computed that the plan does not already describe.
A request written as slots — [PND-PROCSLOT].
nodesis keyed by caller-assigned names, andoutputsis keyed by the caller's name for each surfaced result. Both names ride back on the response, so a consumer keys its UI on a slot that survives a param edit and its cache reasoning on the derived id.Every slot becomes a plan entry, so a slot nothing selects is still reported in RunResult.nodes — a pipeline view draws the graph that was described, not the subset one selector reached.