LLM systems need evaluation before they need mythology
The practical question is not whether a model feels intelligent. It is whether the workflow is measurable, bounded, and worth trusting for the task at hand.
Drafts & essays
AI · measurement · baseball
Why healthcare analytics starts with data contracts, not dashboards
Claims, registry, and EHR-derived systems only become useful when the definitions underneath them are stable enough for operators to act on.
Baseball is still one of the best laboratories for uncertainty
Forecasting baseball performance forces you to confront variance, calibration, model drift, and the limits of narrative confidence in a way that transfers surprisingly well to other domains.