TRACE-TS Grounds Human Activity Reasoning in Sensor Data

TRACE-TS uses attribution from expert classifiers to create DAG reasoning traces for wearable sensor time series. This framework allows compact language models to output both activity predictions and verifiable, signal-grounded explanations.