We did $old - $new instead of $new - $old so positive changes where good and negative where bad T-T
This allows us to analyze the report more easily. There is experimental support for comparing against a base report, but we don't expose that functionality to the action.