Built a fault-tolerant orchestration layer for long-running data pipelines in TenderScope. The system manages the full run lifecycle — running, success, partial success and failed — while preventing duplicate triggers and recovering safely from interrupted or stale runs. I implemented lease-based execution and reclaim after restarts, stale-run recovery, coordinator locks and PostgreSQL-backed concurrency controls, then validated the behavior with real database concurrency tests. The result is a pipeline architecture designed to recover from failures without leaving jobs permanently stuck or unintentionally processing the same work twice.