How Solanto will keep getting better.
Every scribe looks good in a demo. What matters is the note you would have to fix. Here is how each Solanto version will be checked before it reaches you, and how every mistake will become a test the next version must pass.
One question first: would you sign it?
Truth comes before style. Before we ask whether a note reads well, we will ask whether it says anything that did not happen in the room. A version that invents one dose, one finding or one plan will not ship, however polished it reads.
Every change a scribe makes to what was said falls into one of four kinds.
Tidied.
The same facts, said more clearly. That is the job.
Filled in.
Something likely, but not said. It counts against the version.
Jumped ahead.
A conclusion you did not reach. It counts against the version.
Made up.
Something that did not happen. One is enough to fail.
Stress tests: we test where scribes break.
A demo is five calm minutes. A clinic day is not. So each version will be run against the conditions where notes go wrong.
Long visits.
Past fifteen minutes, notes start to drop facts. We run full-length visits, not highlights.
Numbers and doses.
Seventy-five milligrams must stay 75 mg, never 750. Every dose, date and number is checked against the recording.
Corrections.
“Sertraline. Sorry, no, escitalopram.” The note must keep the correction, not the first word.
Who said what.
The patient’s words stay apart from your findings, and every line traces back to who said it.
An older phone, running hot.
Private Mode will run on your phone, so we test on the oldest iPhone we plan to support, after a morning of visits, not a cold start.
The Watch on your wrist.
A full hour on Apple Watch with the screen off, the way it is really worn.
Visits it has never seen.
Every fix is confirmed on recorded mock visits held back from the work, drawn from a public set of mock consultations (PriMock57), so nothing is tuned to the test.
RSI testing, in plain words.
RSI stands for recursive self-improvement. In AI research it means a system that improves the very process it uses to improve. We use a narrower, safer version: a testing loop where every mistake makes the next round harder to pass.
1. Find the worst mistake.
Start with the single most harmful error from the last run.
2. Turn it into a test.
Write a case that catches it. The case stays in the set for good.
3. Try one fix.
Change one thing, and write down what we expect before we run it.
4. Judge it blind.
A reviewer compares the old note and the new one without knowing which is which.
5. Confirm on fresh visits.
Run it again on visits the fix has never seen, so it is not just memorizing.
6. Keep it only if nothing got worse.
A fix that helps one thing and hurts another is thrown out.
Because each mistake becomes a permanent test, the set will grow with every round. The bar only moves up.
The loop proposes. A person decides.
Memory
It will learn you, too.
Getting better is also personal. Across visits, Solanto will remember the patient and learn how you write, so the next note starts closer to yours. Tap any point to see what it will keep.
Run it on us too.
You should not take our word for any of this. Our free test visit gives you a six-minute script, twelve planted checks and an answer key. Run it on any scribe, including ours.