OptionalcarriedEdges that are not rows: they follow their endpoints and land iff both ends end up on canvas. Not shown in triage — an edge to a node the analyst did not take is not a decision of its own.
How many of them are already on canvas.
Edge-only results: both endpoints already on canvas, so they are rows of their own.
OptionalerrorSet when fetch rejected — the pane offers a retry.
How many nodes the provider returned, before dedup or suppression.
True while fetch is in flight.
The narrowing this set was fetched with.
The nodes the pivot was run on. Empty for an origin-less pivot.
OptionalpendingA re-run of this pivot, waiting rather than replacing: it landed while rows were marked, and throwing the analyst's triage away unasked is not a swap the library gets to make. The surface announces it and offers both ways out — PivotManagerLike.showPending and PivotManagerLike.dismissPending.
A re-run with nothing marked replaces outright, and never lands here.
OptionalprovisionalThe run might still land without triage, so nothing here has been offered yet.
The set exists from the moment fetch starts — that is what makes the run
watchable and cancellable — but a surface must not put a provisional set in front
of the analyst: it is about to disappear onto the canvas, and taking the screen to
say so costs more than the result it is announcing.
Cleared the moment the run turns out to need triage, which is the surface's cue to come forward. A run that was always going to stage is never provisional.
OptionalrefusedSet when the provider blew through the absolute ceiling. Nothing was staged.
The id this set's first ingest will be recorded under.
How many were dropped because they were explicitly rejected earlier this session.
One
fetch's worth of candidates, held for triage. There is at most one set per pivot id: a re-run replaces it.