Many "re‑runs keep getting harder" problems in GBase 8a batch processing don't come from the scheduler — they come from commit granularity, batch boundaries, and rollback strategies that were never explicitly designed. A multi‑step script fails halfway; nobody can say which steps are already committed and which are still in the session; re‑running either doubles the results or leaves stale data, and teams end up fixing things by hand. Why Transaction Boundaries Matter in Batch Work Analytical da