Our models rely on signals that are broadly available across consumer-grade wearables: heart rate and respiratory data, along with sleep metrics (duration, phases, heart rate and respiratory rate during night). We've scoped our signal requirements to inputs that are common across the major wearable ecosystems to maximize compatibility.

The signal set we depend on is shared across most major consumer wearables, so the model is designed to generalize beyond a single brand.

Thanks to a preprocessing layer designed to normalize and harmonize signals coming from different sources, we don't enforce a strict minimum sampling rate or data quality threshold, the system is built to handle variability across devices. That said, higher-quality and higher-resolution input signals translate into more precise output scores.

The model can optionally leverage up to five previous sessions to provide more personalized and context-aware assessments. Each previous session consists of the original payload sent to the API together with the corresponding API response. Sensia does not store or manage this historical data. Clients wishing to enable personalization are responsible for maintaining the session history and including up to the five most recent sessions in each API request. If no previous sessions are provided, the model will generate its assessment based solely on the current input.