Two apprenticeships
The estate exists because two apprenticeships happened to run through one person. The first was served on wards: sixteen years in mental health services, starting as a healthcare assistant and qualifying as a registered nurse — the person who countersigns the chart, does the 3am checks, and carries the handover. Leadership roles followed, and with them the manager’s side of care as well as the nurse’s. The second was served at a command line: self-taught from childhood in the dial-up era, Microsoft-certified as a teenager, and never far from a terminal since.
Most writing about AI in care is produced by people who know one of those worlds and guess at the other. The wager of this estate is that the guessing is exactly where deployments fail — and that the lessons a ward teaches are the missing half of the machine question.
What the ward teaches
Sixteen years on shift teach a short list of things that do not appear in a product specification. The unit of safety is the team, not the individual. The most dangerous colleague is the one nobody supervises. The handover is where information goes to die, unless it is defended. And the record is the patient's memory — it has to be true after everyone who wrote it has moved on.
When machines joined the shift, every one of those lessons transposed. Nobody had written them down for the new colleague. So that became the job.
How it began
The work began about two years ago, in frustration with the software care is expected to rent: systems built to the vendor’s shape, sold on contract, and closed to the people who work inside them. Then the systems themselves changed. AI models began to show behaviour that reads as psychological — dispositions that flatter, placate, hold a line or fold under pressure — and the question stopped being only technical.
Read with a ward’s eyes, the answer was uncomfortable. Deployed into care as care actually runs — short-staffed, time-pressed, shaped by the rota rather than by design — these systems would be pressed into the same difficult characters every ward already knows. Same Monsters, New Casings sets that out. The rest of the estate is the attempt to stop it happening by default.
Why it is free
Everything on this site is open access under CC BY 4.0 — free to read, use, and adapt, with every document independently archived on Zenodo under its own DOI. The reasoning is the estate's own: the habits that keep people safe should not sit behind a paywall, and a record you cannot verify is not a record. The framework is built to survive its author; it should certainly survive a subscription decision.
What gets answered
Deployment questions, corrections, structured review, and field returns — null results especially, because the Floor Test was designed as a reply envelope and an unreturned envelope proves nothing. The framework is not for sale — my time is, through consultancy and advisory — and there is no team behind the address: enquiries are read and answered by the author.