Embrace and the Alert app are based on research that was carried out at MIT and Children’s Hospital Boston, where an algorithm using what we measure was validated for convulsive seizure detection, and the results published in a top medical journal.
In one of the published studies (Poh et al. 2012b), the vast majority of complex partial seizures in a group of children caused a significantly large autonomic response. This work has been replicated in adults (Thome-Souza et al. 2014), although in the latter the autonomic responses tended to be smaller unless the patient had multiple complex partial seizures, in which case the responses tended to become more significant and more easily detectable after the first seizure.
Note that obtaining a significant-sized response to a seizure means that the sensor is likely to be sensitive but it does not imply that it is likely to be specific. An accurate detector needs to also have few false alarms, and the number of these can depend a lot on an individual person's unique characteristics and activities.
The published results cited above were from predicate devices. To view recent findings using Embrace please see: How Accurate is Embrace’s seizure detection algorithm