Case Files
The bugs, the retros, the wrong turns
One system, one wrong turn, one thing to keep. Every case file has an explicit "we shouldn't have done that" section — the part you rarely see in tech blogs.
- The eval pipeline that lied for a month· an offline LLM evaluation harness
A team shipped model changes against a green eval suite for a month while quality quietly dropped. Here's the decomposition — and the wrong turn that hid it.