Cells are of the form Rate (X%) where:
- (for crashes) is the number of that type of crash divided by the usage_khours
- (for usage_khours) is just the number of usage_khours
- (for crashes) is the percent that this week's number (not the rate) of crashes is vs. last week's number (not rate) of that crash
- (for usage_khours) is the percent that this week's number is of last week's usage_khours
Since crash rates are #crashes / usage_khours a low (X%) in usage_khours can result in an inflation of crash rates.
To estimate how much of an inflation, take a relatively-stable crash type (like P) and compare its (X%) to usage_khours'.
As of yet, no model has been developed to do this properly, so YMMV.