MetricBraid: AI wearable-data reasoning
An open-source ruleset that helps AI reason more carefully about wearable data and source quality.
Reasoning across Garmin and Oura data
MetricBraid grew out of using a Garmin watch, an Oura ring and a Garmin chest strap while training. Those systems can describe the same workout differently, and they do not measure every metric in the same way. Giving an AI access to all of them does not automatically make its answers more reliable.
The ruleset makes source preference, provenance and confidence explicit. It addresses duplicate workout records, competing measurements, individual baselines and the difference between an association and a causal claim.
Making uncertainty part of the answer
The goal is to constrain what an AI can infer before it starts constructing an explanation. Two observations should not become a trend simply because a model can describe them fluently. Conflicting signals can be worth preserving rather than forcing into a single answer.
MetricBraid remains an evolving experiment in evidence-aware AI reasoning. It draws on published independent research where evidence exists and makes uncertainty explicit where it does not. The related essay explains these choices through examples from my own wearable data.
Where the project is now
An evolving open-source framework for interpreting wearable metrics with more explicit reasoning about confidence and source quality.