Prometheus Entity Management API
    Preparing search index...

    Per-runtime state container.

    Both fields were previously module-level singletons shared by every runtime in the process. That was not merely shared state — hydrateGraphFromStorage clears the pending set before repopulating it, so a second runtime hydrating erased the first runtime's un-settled actions. On an account or practice switch that is silent write loss, which is why this is scoped per runtime (ADR-009 G1).

    interface RuntimeScope {
        pendingActions: Map<string, GraphActionRecord>;
        statusStore: StoreApi;
    }
    Index
    pendingActions: Map<string, GraphActionRecord>
    statusStore: StoreApi