When you paste a stack trace and ask an AI to fix it, it usually patches wherever the error surfaces not where the bug originates.
I built a prompt that forces the model to work the problem in order: read the symptom literally, rank hypotheses by likelihood with confidence levels, identify the actual origin (not just the throw site), then propose a fix and a way to verify you actually found the root cause not just made the error move.t: instantly share code, notes, and snippets.
