The notebook
An investigation keeps competing explanations for one question, tests each against the record, and says what is still uncertain — and who it would need to go further.
Samples, not anyone’s health. Each one below is the real engine run on the app’s invented example record — the same one a new account starts with. Where a sample says “you”, it means that example.
Recovery trend
“Why has my recovery worsened?”"No real change — recovery is moving within normal day-to-day variation" leads at 0.31 of the evidence weight, left open: two records disagree about resting heart rate and sleep.
Glucose and HbA1c
“Is my blood sugar going up?”"No real change — the markers are moving within their normal variation" leads at 0.85 of the evidence weight.
How it decides
- A model may propose an explanation or an experiment. Code decides what counts as evidence, how the weights move, and whether anything may run.
- The weights are each explanation’s share of the evidence among those listed — never a probability.
- It does read-only work by itself. Anything that draws blood, starts a trial or moves an instrument stops and names who has to say yes.
- Example readings can weaken an explanation; they never rule one out, and they never resolve the question.