Sense has the inherent capability to distinguish between two largely similar devices, for instance that draw the same steady power once up and running because its models consider both time and frequency domains.
To get specific, amongst other things, it is analyzing switch-on transients and pulling out frequency components that will vary between two identical devices, as seen at the Sense monitor, due to production variability and different attenuation of high frequencies due to the different wiring runs for those two devices back to the panel.
The techniques for doing this one at a time are decades old. The Sense secret sauce is in the automation / learning systems side of things and the ability to capture the data in your board, do some basic analysis in the unit and then beam it back to Sense, where the really clever stuff happens.
It’s something of a work in progress, quite reasonably, but nonetheless very impressive.