Indeed @ben it starts to feel like stereotypical Internet tech support: “Just shut it down for 15 minutes and then restart, that should fix it!” 
@jkish I wish somebody would come up with a really good quick method of explaining this … I’ll add to @ben’s and others’ attempts and try and keep it generic:
- Imagine your Sense has detected a 10w lightbulb (we’ll call it Light1) as a Device. Great!
- You switch the light on and off and see it perfectly in your Bubbles and in Light1’s Power Meter.
- Also, when you look closely you see that the overall Power Meter goes up and down 10w with the light being on and off. Wonderful!
- Then you install another 10w light, we’ll call this one Light2. Here we go!
- You switch Light2 on and off and the Light1 Bubble and Power Meter show it perfectly. Oops!
- What to do?
This may seem like a trivial example (it is) but it represents the near impossibility of Sense reliably tracking all devices. In this example, any device (e.g. a fridge) that has a 10w fluctuation in usage (not uncommon) could fool Sense into thinking the light just went on … or off.
To stretch things a little further … even though a 10w light has a straightforward electrical signature that doesn’t fluctuate up-and-down, because it has (in the “average” home) low power usage it is actually very hard for Sense to separate it from the other noise in the home like a fridge or a washer dryer or another light. When something like a hot water tank comes on and uses 4kW it’s usually the only device doing that so eventually Sense has a good chance to recognize it.
And one more step into the beyond … Sense’s learning could of course (and does, as far as I understand … please correct me techies) incorporate usage patterns. For example if Light1 above only gets used on Monday-Friday and Light2 is Weekends-only, then in theory Sense can (does?) use that information to differentiate Devices. Not that that helps necessarily if the fridge is cycling on and off all week.