How this is different
A trail you can check, not an answer you have to trust
Most tools like this work like a well read assistant. You ask a question and it writes a new answer each time, so the same question can get a slightly different answer each time. It can be hard to know exactly why the tool said what it said.
Our engine works a different way. It is built like a recipe, not like an assistant. The same patient information always produces the same result, every time. Along with the result, it shows its work: a step by step list of why it reached that answer, how sure it is, and what would have to change for the answer to change. This is called a reasoning trace.
No names or contact details ever reach the reasoning engine. It works from a record id and the clinical facts, under a signed BAA with row-level audit logging. That keeps the most direct identifying information away from the part of the system doing the thinking.
In short: other tools explain an answer after they generate it, in a way that can shift each time you ask. Our engine shows its reasoning as a fixed, repeatable trail, and keeps the patient's name and contact details away from the decision making core.
This is a real sample output from the engine, made with synthetic data. It is not made by a language model. With the same structured input, the engine gives the same output every time. Patient names and contact details do not enter the engine. It works from a record id and the clinical facts.
Sample trace from a synthetic patient.
This example shows how the engine explains one result. It is not medical advice.
1. The trigger
One result was higher than normal
This patient's inflammation marker was higher than normal.
- Result
- 2.4 mg/L
- Normal range
- 0 mg/L to 1 mg/L
2. The reasoning steps
How the engine reached one sample recommendation
The engine found hs-CRP, an inflammation marker at 2.4 mg/L. That is above the normal range of 0 mg/L to 1 mg/L. It was 85% sure about this finding.
The result pointed to body-wide inflammation linked with gut health. The engine was 78% sure about that link.
Based on the earlier steps, the engine selected BPC-157. It found 487 supporting research records and was 70% sure about this choice.
3. The confidence score
A score you can see and question
The score at the top brings together the strength of the research, the lab result, and the link between them. It is for this one sample recommendation, not for every patient.
4. The counterfactuals
What would need to change for this result to change
Change 1
If the inflammation marker fell below 1.0 mg/L and returned to its normal range, this recommendation would drop. Its main trigger would be gone.
Change 2
If the supporting research count fell below 50, this choice would move down because it would no longer meet the first-choice bar. The engine could consider KPV instead.
Change 3
If the source rating fell from B to D or E, this choice would move down. The preferred way to source it would no longer be available.