AI project lessons: share a failure and the recovery path

What went wrong in an AI project, and what did you learn from investigating it? This thread welcomes small, specific lessons as well as larger project retrospectives.

• Goal: What the workflow was meant to do.
• Symptom: What failed and how you noticed.
• Evidence: The logs, outputs, or observations you examined.
• Investigation: What you tested, including unsuccessful attempts.
• Change: What you modified and why.
• Verification: How you checked the result and what remains uncertain.

Keep private data and credentials out of examples. If the issue is unresolved, say so and identify the next observation you need.

Replies should distinguish a confirmed cause from a plausible explanation. A useful lesson can be a better test or recovery procedure even when the project is still unfinished.