A glucose trace is a sequence of measurements. Understanding it may also require information about meals, recorded insulin exposure, activity, sleep, illness and the measurement system itself.
Our research explores how authorized observations can be understood together. Missing information remains missing. A reported dose, a prescription and confirmed delivery remain different events.
The research challenge is not simply fitting a curve. Different underlying processes can produce similar observations. A model that forecasts well has not necessarily identified a biological mechanism.
We plan to test models against transparent baselines, across independent people and settings, with uncertainty reported alongside performance. Explanations should help a qualified person inspect the evidence, including reasons the model may be wrong.
