A frontier model’s access was revoked overnight. What it built in a few days — and why the method, not the model, is the part that lasts.
“A method, unlike a model, cannot be withdrawn by directive — and this one was never really about law.”
Queensland raised a levy for mental health, then built no mechanism to confirm the money arrived and no way to know whether it worked — a textbook failure of unmeasured, ungoverned spending.
“A levy without that control is indistinguishable, in operation, from general revenue wearing a more sympathetic name.”
Compliance is the one setting in which a plausible answer and a correct answer are not the same thing. Why durable automation here keeps the language model away from the decision.
“A system that changes its mind without the facts changing is not a control. It is a liability.”
Every transformation programme is a success — until you ask by how much, compared to what, and how you know.
“Transformation without measurement is theatre: expensive, well-attended, and impossible to evaluate.”
When Toowong Private Hospital closed its doors on 11 June 2025, the failure that mattered was not clinical but structural — and that distinction is exactly where disciplined governance earns its keep.
“A hospital can be fully accredited and still cease to exist within a month of a creditor’s meeting.”
A model that is 95% accurate is reassuring until you learn what the 5% is made of. Designing a clinical automation experiment around the failure that matters.
“Accuracy is an average. Safety lives in the tail.”