The doctor signs

Everything the model does in a consultation is a draft until one thing happens: the doctor signs. The briefing, the suggested note, the flagged interaction, all of it sits in the category of proposal until a licensed human reviews it and puts his name on the result. The signing is not a formality at the end. It is the event the whole workflow exists to produce.

And signing here is literal. The doctor approves the note and it is signed with his cryptographic key, the same key that identifies him across the network. In that moment two things happen at once. The draft becomes a decision, and the decision becomes traceable to a specific accountable human. Those are the same act, and neither can happen without him choosing it.

This is why the signature is the safety mechanism, not a feature beside it. A clinical AI is safe to use precisely to the degree that nothing it produces becomes an action without a human deciding to sign. Glyph is built so the model's output is always, structurally, a draft, and the doctor's key is the only thing that can promote a draft to a decision. The model cannot reach around him, because it has no key to sign with.

It also means the accountability is never diffuse. When a Glyph prescription is questioned later, there is no shrugging at the algorithm. A named doctor signed it, with his own key, and the record proves it. That is heavier for the doctor than an anonymous paper scrawl, and it is meant to be. A system where a clinical act traces to an accountable human is the only kind that deserves trust, and the signature is how Glyph guarantees there is always one.