We describe a modeling approach that links user-logged exposures with continuous physiological responses to surface likely triggers, while remaining transparent about uncertainty.
The trigger problem
Identifying triggers by memory is error-prone: responses can be delayed, cumulative, or masked by other exposures. A model that reasons over time can propose candidate triggers that a person might never connect on their own.
Method
By aligning logged foods, environments, and activities with subsequent physiological shifts, the model estimates the strength and consistency of each candidate association, ranking them for the user and their care team.
Responsible design
Trigger suggestions are framed as hypotheses to test, not verdicts. Communicating uncertainty is a core design requirement, not an afterthought.