A misplaced UUID caused four weeks of silent data pipeline failures on Mondays

A data pipeline built on Cloudflare Workflows silently failed every Monday for four weeks, reporting zero errors while leaving gaps in a public time series. The root cause was a UUID generated outside a persisted workflow step, causing it to regenerate on replay after a sleep, so finalize operations targeted a database row that never existed. Because updating a non-existent row returns a success response in D1 with zero affected rows, no error was raised and downstream readers simply saw missing data. The bug only surfaced on Mondays because the weekly code branch was the only one containing the sleep call that triggered replays. The fix involved moving identity-bearing values like UUIDs inside persisted steps and adding explicit checks on database row-change counts to catch silent no-op writes.