Do you publish an accuracy percentage?

Not yet, because an accuracy number without methodology is marketing. We publish the evaluation method, examples, dates, and limits first. Formal benchmark numbers should come with dataset size, review criteria, and external validation.

What is the key evaluation difference?

We evaluate the raw clinical facts as a first-class output. A note that looks polished is not enough; the facts behind it must be visible and usable by the clinician.

How often should these evaluations be refreshed?

Quarterly for comparison and methodology pages, and whenever the extraction model, note generator, or supported language set changes materially.

Hoe wij het hallucinatierisico van klinische AI evalueren

Het bewijsverhaal van Notat is geen magisch nauwkeurigheidspercentage. Het is een transparante methodologie: eerst klinische feiten extraheren, notities genereren op basis van die feiten, de ruwe context aan de clinicus tonen en elke output tegen het bewijs toetsen.

Bekijk FactsContext Vergelijk leveranciers

Evaluatiemethode

1. Build the reference facts

A reviewer reads the encounter material and creates a reference set of clinical facts: symptoms, negatives, medication changes, assessment, plan, safety-netting, and coding-relevant details.

2. Compare transcript-direct output

We generate a note from transcript-style input and mark unsupported statements, missing high-salience facts, incorrect attribution, and invented certainty.

3. Compare FactsContext output

We generate documentation from extracted facts and review whether each sentence is supported by a visible fact. The clinician-facing fact list is evaluated as part of the output, not hidden.

4. Record limits, not magic numbers

We do not publish fake accuracy percentages. Each evaluation note includes dataset scope, review date, known limitations, and examples of what the system still requires clinicians to verify.

Wat beoordelaars markeren

Unsupported clinical assertion

Wrong medication, dose, frequency, or route

Missing red-flag negative or safety-net advice

Wrong diagnosis certainty: possible vs established

Wrong speaker attribution

Unsupported ICD-10 suggestion

Clinician-visible evidence for each key statement

Facts reusable for notes, codes, referrals, and patient instructions

Gedateerde interne notitie

Huidige evaluatienotitie, 2026-07-01: deze pagina documenteert de methode en de kwalitatieve voorbeelden die worden gebruikt om de FactsContext-architectuur te evalueren. Er wordt geen externe validatie of universele nauwkeurigheid geclaimd. De volgende bewijsmijlpaal zou een geblindeerde, per specialisme gestratificeerde beoordeling moeten zijn, met openlijke rapportage van datasetgrootte, overeenstemming tussen beoordelaars en het percentage niet-onderbouwde uitspraken.

Feitenextractie versus direct genereren uit transcript

Medication change discussed twice

Risico van direct transcript

“Increase amlodipine to 10 mg and stop lisinopril.” The transcript contained a correction: the clinician first considered stopping lisinopril, then decided to continue it after reviewing renal function.

FactsContext-output

Facts: amlodipine increased to 10 mg daily; lisinopril continued; renal function normal; review in 6 weeks. Note generated from those facts only.

Waarom het ertoe doet

Transcript-direct generation can smooth over corrections. FactsContext preserves the final decision as a discrete fact before writing.

Negative finding matters

Risico van direct transcript

“No neurological symptoms.” The actual encounter only documented no saddle anesthesia and no bladder symptoms; leg radiation was present.

FactsContext-output

Facts: left leg radiation to calf; SLR positive left; no saddle anesthesia; no bladder or bowel symptoms. Note keeps the negatives specific.

Waarom het ertoe doet

Broad negative statements are risky. The fact list keeps the clinical context granular and reviewable.

Code suggestion requires evidence

Risico van direct transcript

Suggested J44.1 for COPD exacerbation without showing the symptom or treatment evidence.

FactsContext-output

Facts: increased breathlessness, purulent sputum, prednisolone burst, antibiotics started. Suggested J44.1 with those facts as evidence.

Waarom het ertoe doet

The code is easier to verify because the reason for the suggestion is visible, not buried in prose.

Bekende beperkingen

Clinician review remains mandatory. Notat drafts; clinicians verify and sign.

The method reduces unsupported statements by architecture, but no clinical AI should claim zero hallucinations.

Small internal evaluations are useful for engineering direction, not a substitute for external clinical validation.

Specialty, language, audio quality, speaker overlap, and local coding rules can change performance.

Lees het bewijs en controleer daarna je eigen feiten.

De evaluatiemethode is eenvoudig omdat het product is ontworpen om inspecteerbaar te zijn: eerst feiten, dan de notitie, altijd klinische beoordeling.

Probeer Notat gratis