My routine said it ran. It was lying.

Short summary

Your scheduler says the job ran, but did it actually work? The author debugged an AI routine that showed green checkmarks while producing no output—it silently failed before the write step. Key lessons: (1) verify output artifacts exist where claimed, (2) check they're fresh and non-empty, (3) read raw transcripts instead of summaries from the system that failed, and (4) fix it by persisting output first.

• •Scheduler success ≠ job worked—check disk artifacts, not status dots
• •Read raw transcripts, not summaries; the summary is written by the thing that failed
• •Fix: persist output as first action before any fail-able step

Generated with AI, which can make mistakes.