A physician asks a patient how their breathing has been. The patient says fine. The physician listens to their chest and writes decreased air entry bilateral bases. None of that information was spoken in the conversation. It was observed examined and interpreted.
This is the part of clinical documentation that transcription based tools cannot solve. They can tell you what was said. They cannot tell you what was found.
How we approach clinical context
Diagnose is not a transcription tool that reformats its output into SOAP structure. The model was trained to understand the difference between the subjective account a patient gives and the objective findings a physician documents. These are not the same source of information and they should not be treated as one.
During an encounter Diagnose listens for linguistic markers that indicate a shift from reported to observed. When a physician says the patient reports their pain is a seven out of ten the model handles that differently from when a physician says tenderness on palpation right lower quadrant. One is patient reported. One is clinician observed. The note structure reflects that distinction.
What the model still cannot do
We are direct about the limits. Diagnose cannot document what a physician does not say or indicate during the encounter. If a physical finding is made without any verbal acknowledgment it will not appear in the note. This is intentional — we do not generate clinical content that was not present in the encounter.
Physicians who use Diagnose develop a light verbal habit in the first week. A brief narration of physical findings as they examine. It takes less than ten seconds per section and it closes the gap between what is observed and what is documented. Most physicians tell us it feels natural within three or four encounters.
