The issue with the COP graph (9:05) is that it's comparing the mini-split system to the prior system. The mini-split doesn't have radiators in the same place as the prior system, so he runs it 1-2˚ hotter (7:50) to get the rest of the house to the same temperature. Also, the system isn't using much electricity on warm days (8:48), so the inefficient placement of the minisplit is more dramatic.
The more interesting graph is the total kWh saved per day (8:07). The colder it gets, the more energy he saves. That's because more heat needs to be pumped into the house, so a worse COP multiplied by more total heat equals more electricity savings (at least for the temperatures he's measured).
