Engineering intelligence, quantified.

Claude Told Me a Family Member Had Kidney Disease. They Don't.

I used a frontier model to reconcile years of a relative's medical reports. It was genuinely useful — and it made five mistakes that, clinically, would matter.

Self-review, a second model, even a stronger model missed them. What caught each error wasn't a smarter model — it was checking every claim against its source. A real, de-identified case. Not medical advice.