OK - that explanation from your utility highlights the offset. For Sense, the 00:00:00 (midnight+) data point is the usage for the time between 00:00:00 and 01:00:00. You should also note that that will affect your daily calculations.
That explanation also dispels the two possible correlation error sources I mentioned, though Iām still suspicious that your utility might be hiding a 15min read cycle under their hourly numbers.
The next thing I would do is plot a bunch more hours with the corrected offset using the x-y plot you just constructed and see if there is any correlation between the size of the āerrorā and other factors like time of day (color the dots for time of day like in the plot below). Multiple days might also highlight some larger outliers that could provide the key.
When I say smaller solar component, I mean size of solar system. My system is large enough that even in the winter, my net usage from my utility and Sense (Total Usage plus Solar Production) goes negative.