Member Junction
    Preparing search index...
    • Advances the frontier: given the node that just finished, returns the nodes to start next.

      sequential returns at most one successor — the old single-program-counter behavior, kept as the default so existing design-time flows traverse exactly as they did. parallel returns every successor whose join is satisfied, which is what graphs built from a TaskGraphSpec always use: an agent that expressed independent work as independent nodes meant them to run at once, and serializing them would discard the only information the decomposition carried.

      A successor whose join is not yet satisfied is simply not returned. It becomes eligible when its remaining predecessors settle and this runs again for one of them — so the join is evaluated on every completion rather than parked on a waiting list that could go stale.

      Parameters

      Returns { NextNodeIds: string[]; Selection: EdgeSelection }