Free · open · no account
A safety architecture for AI-assisted clinical documentation. Six named protocols, published openly so they can be cited, taught and argued with, whether or not you ever buy anything from us.
Each one is a rule you can adopt today without any file from us. The paid toolkit is the implementation manual; the standard itself is free.
No AI-generated content enters a record, reaches a patient, or goes to a payer until a licensed clinician has read every word and taken responsibility for it. Not skimmed. Read.
Before anything is typed, the task goes in exactly one box. DON'T, diagnosis, treatment, consent, capacity. DOUBLE-CHECK, coding, guidelines, anything with a number. DRAFT, notes, letters, appeals. DELEGATE, reformatting, reading level, admin.
A three-tier classification of tools by the data they may receive, a de-identification protocol, and the mosaic test, because a rare diagnosis in a small town identifies someone no matter how many fields you strip.
Every prompt ends by forcing the model to list what it added that you did not supply. You are not asking it to be honest. You are requiring a diff. Trust is not a prompt technique.
Every task carries one of three marks. GREEN use as written · AMBER verify the output against source before it goes anywhere · RED never delegate this. A library without triage is a pile.
Five checks between the draft and your signature: did it invent a finding · did it invent a number · did it harden a hedge into certainty · does every wildcard have real content · would this read the same for the next patient. If yes to the last one, it is a cloned note.
No email required for anything on this row. We would rather you used them and never bought anything than guessed.
Paste clinical text; it flags and masks all eighteen HIPAA Safe Harbor identifiers before you put anything into an AI tool. Runs entirely inside your browser. The page is served with a Content-Security-Policy that makes an outbound request impossible, not merely absent. Disconnect from the network and it still works.
Open the tool →Forty-seven specific ways AI-assisted clinical documentation goes wrong, in six classes: the completed exam you never performed, the confident dose you never said, the phantom negative, the borrowed vital sign, the confabulated citation, the manufactured consent. Every one is cross-referenced to the protocol clause that catches it, so a department policy can cite AS-06 check 1 and mean something specific. 26 pages. Free, no signup.
One email, occasionally
And the beta study result. The measured one, with its method attached and its N stated, however unflattering the number turns out to be. That is the whole list. No drip sequence, no funnel.
We never ask for and never want patient information. Unsubscribe in one click, in every email. Full detail in the privacy notice.
The standard above is free forever. The toolkit that implements it, handbook, 100 badged prompts, plain-text EHR note structures, templates and trackers, is $99, one time.
See what's inside →