Writing
Write-ups of real findings from my projects, mistakes included.
Rerankers made my code search worse. Then the same one made it better.
A reranker made my code search worse on December 18, 2025. Two days later the same model was the biggest gain in the system. Six things changed in between, one of them my answer key.
My LLM judge missed a skipped ID check. 24 lines of code didn't.
My LLM judge caught a scheduling bot giving medical advice and missed that it booked without verifying the caller. A 24-line check on the tool log caught it.