The problem is the template sensors feeding the statistics sensors. We made them go unavailable so the statistics average would not be affected when the template sensors were not measuring.
I could make the two template sensors hold the last value, which will affect the 7 day average. Or hold the statistics average, but again this would be self referential and just reinforce the last calculated average.
How constant are the 7 day averages turning out to be?
The other option is to make two more template sensors that hold the last value of the statistics sensors. These will always display the averages but not affect them.