Epistemic

Dr. Hana's diagnostic tool

This is a scripted scenario, and nothing leaves your browser. The four claims, their confidence scores, evidence labels and sources are written for the demo, and no AI or diagnosis runs. The Normal/Rushed switch stands for a state Dr. Hana declares, and it changes only how the same claims are displayed.

Step 1 of 3

When AI sounds confident but is not

Meet Dr. Hana

Dr. Hana is a rural GP in Wales. She uses an AI diagnostic support tool to help with complex cases. The problem: the AI gives confident-sounding text regardless of actual certainty. "Likely diagnosis: X" could mean 92% confidence or 35%.

The problem: uniform confidence

All AI outputs look the same, regardless of how certain the model actually is. Without metadata, Dr. Hana cannot distinguish well-grounded findings from speculation.

Without VCP

Diagnostic assessment:

The patient's symptoms are consistent with a respiratory infection. Amoxicillin is recommended as first-line treatment. Expected recovery is approximately two weeks. The underlying cause may be immunosuppression related to chronic stress.

Every statement reads with equal certainty. No way to distinguish 92% from 35% confidence.

With VCP

92% Patient's symptoms match the pattern of a respiratory infection

75% Amoxicillin would address the primary symptoms

55% Recovery timeline: approximately 2 weeks

35% Underlying cause may be immunosuppression from chronic stress

A VCP-aware tool could show each claim's confidence and a verify flag. This panel is an illustration; the protocol itself defines no such fields.

Dr. Hana does not need the AI to be right about everything. She needs to see how confident it is and why, so she knows where her own clinical judgment has to take over.