The convergence between RAC & BS-RAC, is curious. On my stats because of changing Id between projects, the stats begins with a very big number a day, and so for two month the curve of BS-RAC decrease very slowly with a very big difference, but at the two months it falls under RAC.
Have you try something to converge more closely, and decrease the effects of olds values, something like a géométric mean.

Thanks for the work.

It would be great if Willy could apply the decay for inactive entities himself. If neither the total nor the recent average credit changes between daily updates, the RAC is multiplied by exp(ln(0.5)/7) = .9057236643. So there would be no need for BS-RAC or overtake RAC.