## Problem To Solve When an entity fails during processing, the entire batch may need to be retried from scratch, wasting resources and time. There's no mechanism to resume from where things left off, which means a single failing entity can cause a lot of unnecessary rework. ## Proposed Solution Implement per-entity watermarks that track the last successful checkpoint for each entity, allowing retries to resume only the failing entity from its last known good state. If time permits, investigate cursor-based checkpoints for SDLC entities to enable even more granular retry behavior and reduce the blast radius of failures further.