Scribe
Populate a form or outcome measure from a document or consultation — every value proposed, reviewed, and verified by you before it enters the record.
Scribe turns a source you already have — a lab report, a scanned form, or the consultation you just had — into a drafted set of values, so you don't have to type them in by hand. You still do the clinical thinking. MIA does the transcription.
Scribe output is AI generated — clinician must verify. Nothing it drafts reaches the patient record until you have reviewed it, field by field.
Two ways to feed it
Scribe reads from two kinds of source, chosen from the Scribe tab on a patient record (or, for blood tests and ECGs, directly from the relevant section of the Clinical tab):
- Document — upload a lab report or scanned form (PDF, PNG, JPG, WEBP). MIA reads it and drafts the values.
- Consultation — record the consultation on this device. MIA transcribes it to text. You read and correct the transcript yourself, then run extraction on the corrected text — not on the audio.
Before you extract, you choose what the source is populating: an outcome measure (MIA lists whichever measures are active for your organisation) or investigation results (lab panels and ECG findings).
Recording a consultation requires an explicit consent checkbox before the microphone starts. No audio is ever stored — only the transcript, and only after you've reviewed it.
Every value is a proposal
This is the core of how Scribe works: nothing MIA drafts is saved automatically. Each field or result Scribe reads becomes a proposal sitting in a review queue, and a proposal only reaches the record once you act on it. For every proposed value you have three choices:
| Choice | What it means | Result on commit |
|---|---|---|
| Accept | The value is correct as drafted | Written to the record as drafted |
| Edit | You type the corrected value in | Written to the record as you corrected it |
| Reject | Doesn't belong in the record at all | Left out entirely |
A proposal stays visibly in draft — badged and dashed-bordered — until you choose one of the three. There's a bulk Accept all for when MIA was confident across the board, but it only ever touches values MIA was confident in. Anything MIA wasn't sure about is carved out from that shortcut on purpose.
When MIA isn't sure
If MIA can't read a value with reasonable confidence, it doesn't guess. That field is marked "Found but uncertain — check source" and you can't accept it as-is — you have to open the source, check it yourself, and either edit in the correct value or reject it. Confidence percentages are shown next to values MIA was able to read, and a second internal check ("verified" / "unverified") flags anything the model's own re-read couldn't confirm against the source — another prompt to look before you accept.
Every proposal shows where it came from
Select any proposal and the source pane shows you exactly the material it was drawn from:
- For a document, the original file opens alongside the field list, with the matching page and the verbatim excerpt MIA read the value from.
- For a consultation, you see the quoted line from the transcript, who said it, and the timestamp in the recording it came from.
You're never asked to take a drafted value on trust — the source is one click away, every time.
Investigation results (lab values, reference ranges, out-of-range markers) are transcribed exactly as printed on the source. MIA never calculates or judges whether a result is high, low, or abnormal — that reading is yours.
Nothing gets silently dropped
If MIA finds something in the source that doesn't match any field on the form you're populating, it doesn't discard it — it's listed separately as "Found in source, no matching field" so you can see it and decide whether it matters, even though it can't be committed automatically.
Committing to the record
The Commit to record button only becomes active once every single proposal has been accepted, edited, or rejected — there is no way to commit a draft with anything still pending review. Once you commit:
- Rejected values are left out of the record entirely.
- Accepted and edited values are written to the patient's outcome measure or investigation results, exactly as you confirmed them.
- The draft is marked Committed and locked. Like every clinical record in Qyire, a committed Scribe entry is never edited in place afterwards — corrections happen the same way they do everywhere else in the record, as an append-only addendum that preserves what MIA originally drafted and what you changed it to.
Every accept, edit, reject, and commit is attributed to you and time-stamped, giving each Scribe-populated record a full audit trail — who reviewed what, and when.
Why it's built this way
Scribe exists to save you typing, not to make clinical judgements for you. The review queue, the mandatory per-field sign-off, the visible source excerpt, and the lock on commit are all the same guarantee stated four different ways: MIA proposes, you decide. That's true whether the source is a one-page lab report or a twenty-minute consultation — the amount of material Scribe reads never changes how much scrutiny each value gets before it becomes part of the record.