As promised, I kept an eye on the native detection. I now have three months of data to compare. Using hourly export, I compared native to integration. Here is the result, plotted as events per week.
If the two measurments were within 50% of each other, either over or under, I called it a Good measurement. If the native measured less than 50%, I called it a Missed On. If native was more than 50% over, I called that a Missed Off. I manually removed duplicate events caused by the device running across the top of the hour. Those duplicate events showed up in the data export as a small usage in an hour either directly before or after a larger usage. Finally, there was one case of the HS110 being offline for the afternoon and the native device picking up usage, so I called that special case Better.
Slightly more than half of measurements are Good. Unfortunately, that means that the native detection is missing nearly half the usage, so I don’t get to move my HS110 to another location. The bottom line is that this device of 108 Watts is right on the edge of what Sense can reliably distinguish. One good trend in the graph is that Missed Offs seem to have improved with time, which surprises me because of the way this physical device turns off in stages. Those stages are explained in this post: