Use this rather than a plain batch only when a multi-step write needs logic between steps, such as create → check capacity → finalize, where a zero-row guard must abort and undo everything.

Concurrent calls serialise on the shared connection. A retry re-runs work on a fresh transaction, so work must be safe to run twice.

Cache invalidations fire once the commit succeeds. Settings writers that synchronize their own cache can explicitly preserve the loaded snapshot.