Signal Quality Is the Whole Product: Lessons From 14 Months of Continuous Monitoring

A wearable biosensor that performs beautifully in a clinic and poorly in a shower is not a medical device — it is a demo.\n\nOver 14 months of continuous multi-parameter monitoring across an ambulatory cardiology cohort, the dominant failure mode was never the sensor element. It was adhesion, motion artifact, and charging adherence. Roughly 70% of unusable data traced back to those three causes.\n\nOur response was to stop treating signal quality as a post-processing problem. We moved artifact detection onto the device, gated uploads on a confidence score, and gave patients a single unambiguous indicator of whether the device was working. Usable-data yield went from 62% to 91%.\n\nThe clinical takeaway: report coverage alongside every derived metric. A trend line built on 40% coverage should never look identical to one built on 95%.