My take is that you are doing the right thing and what most users try to do, especially in the early going with Sense - do the detailed accounting to resolve bottom up detection and smart plug data with the top down house-level Total Consumption and Always On numbers. Plus, trying to reduce Other while also trying to reduce overall usage.
Just a couple thoughts on how not to go out of your mind dealing with “fuzzy numbers”:
- Look at daily, weekly or monthly data instead of instantaneous (bubble) data when trying to do the accounting. Instantaneous data bops around too much for easy accounting, as you suggest.
- I actually prefer working with Sense hourly exported data and re-aggregating myself for these calculations since it enables me to notice things like “missing hours” (zeros and NA data) that the Trends view hides.
- I’m getting ready to take a look a a full year of 2020 data from a weekly perspective. But for me, that means aggregating some additional bottom-up data from my second Sense and from my own EV charging “detector” algorithm, then cleaning out duplicates, before analyzing.
As for the known undetected and mysterious and unknown detected devices - learn to live with a little uncertainty but continuously keep trying to fill in the blanks. Setting an alert on ‘Central Heat’ and looking at the Device Power Meter waveform might give you enough info to find the mystery devices. I had a mystery ‘Heater’ device like than until one day I saw an alert pop up the exact second I heard a slight noise in my kitchen - I discovered the ‘Heater’ was my much forgotten instant hot water heater.