Definitely don’t think you’re being a jerk here. We totally appreciate the critical feedback. We’re just dealing with a much harder problem than can be succinctly handled with a “it won’t work in homes built before 1980.” Because it might work great in a home from the 30s and poorly in a home built last year. ML-based device detection is just far more complex than that and is based on too many variables to have such a simple disclaimer. Similarly, this is why manual intervention just isn’t all that feasible. We’re not just identifying some pure device signature, tossing that in a database, and then letting your Sense monitor compare what’s happening in your home to those pristine signatures. That’s certainly part of the process, but your devices rarely ever look exactly the same across instances. I joked earlier with a user that we could certainly create a training process as long as they’re OK with us not covering their carpal tunnel surgery…because that’s what it would be. Sense needs a lot more data, and a lot more specific data, than you might imagine. There’s a world of difference between you/me/our data scientists recognizing something in the Power Meter as a particular device and Sense recognizing that same signature on its own, at different times and in different contexts, and doing that consistently. When you look at the Power Meter, you’re only seeing a small window into what’s really happening in your home (it updates at .5s intervals, we pull data 1 million times every second).
I think it’s always worth keeping in mind that this tech is on the bleeding edge. We’re really in new territory here. On top of that, we’re still a teeny company and Sense is only about two years old. We’re young! We have improved drastically since 2016 and will continue to improve. We also offer a 60-day return period, no questions asked. For most users, that should be enough time to know if Sense is going to work out for them.
This thread has gone way off the rails so, in fairness to @mrtimo, I’m going to request that any talk about training Sense be moved to We should be able to TRAIN this!