Writing — digline https://digline.dev/blog/ Occasional posts on measuring LLM systems: what the numbers do when nothing has changed, and what it costs to find out. en Fri, 18 Sep 2026 00:00:00 +0000 The case that errors is usually the case that was failing https://digline.dev/blog/denominator-trap/ A case that errors drops out of the denominator, so precision and recall get measured over a smaller suite than the run you are comparing against. The score goes up. Fri, 18 Sep 2026 00:00:00 +0000 https://digline.dev/blog/denominator-trap/ Bad evals, my own: five exercises from two LLM judges https://digline.dev/blog/bad-evals-my-own/ I applied the reading Dan Luu applies to other people's benchmarks to my own two LLM judges. Numbers first, explanations after. Sun, 13 Sep 2026 00:00:00 +0000 https://digline.dev/blog/bad-evals-my-own/ My LLM eval cried wolf. Here's what I measured. https://digline.dev/blog/my-llm-eval-cried-wolf/ A case went from 5/5 to 2/5 with nothing changed. How I measured the noise floor of an LLM-judged eval, what it caught the week after, and where it still can't see. Wed, 09 Sep 2026 00:00:00 +0000 https://digline.dev/blog/my-llm-eval-cried-wolf/