EventAction<'T>Union

Persist the event to the journal. The actor's state will be updated using the event handler *after* persistence succeeds.

Persist several events from one command as a single journal AtomicWrite: all land or none do (one DB transaction in the SQL plugins), and no state update, publish or saga wake-up happens until the whole batch is durable. Versions are allocated sequentially. Atomic for THIS aggregate instance only — atomicity across aggregates is saga territory. Note the read side still projects the batch event-by-event (durability is atomic, read-model visibility is not).

Persist the event and, once it is durable and folded, immediately save a snapshot — a manual checkpoint, independent of the SnapshotPolicy cadence.

Publish and fold the event in the live actor without storing it or incrementing the persisted version.

Publish the event immediately to the mediator without persisting it. The actor's state is not updated.

Ignore the event or command completely. No persistence, publishing, or state update occurs.

Indicate that the command or event could not be handled in the current state.

Indicate that the state of a saga has changed (used internally by sagas for persistence).

Dispatch an async side effect (e.g. an AI/oracle read) as a "mini saga" without the saga's persistence ceremony. The payload is an INSPECTABLE DATA description of the effect (boxed) — NOT a closure — so `decide` stays a pure `(command, state) -> effect` function you can unit-test by asserting `decide cmd state = RunAsync (box (ClusterThemes texts))` without any runtime. The oracle lives in the runner registered at `Fcqrs.aggregateWithEffects`, which maps the description to a command; that command is sent back to THIS aggregate (reusing the originating CID) and re-enters `decide`, re-validated against current state. EPHEMERAL: the in-flight work is process state, NOT journaled. A crash, restart, or shard rebalance while it runs loses it SILENTLY — nothing re-issues it. Use this only when that loss is tolerable; when the result MUST survive a crash, use a saga (which persists its intent). TOTAL: the registered runner must map every outcome (oracle error, timeout) to a command (e.g. `ClusteringFailed`), never let an exception escape — `Fcqrs.total` helps. An escaping exception fail-fasts the process, like a throwing fold.