Hi, I have been using GB since GB3, and am not usually interested in individual workload scores, and since 2020 most of my devices have been Folding at Home with GPUs only. However, I was curious to see the relationship between GB6 CPU scores and FaH Point Per Day (PPD) for CPU folding, The PPD per Logical Processor correlates about 0.6 with GB6 1core across those of the top-ranked CPUs at https://folding.lar.systems/cpu_ppd/overall_ranks (where there is a corresponding entry in the GB6 processor charts) but it occurred to me that the performance of the GROMACs cores that do the folding might be more closely related to some GP6 workload scores rather than others. I asked my local AI (Google) and got various answers, which led me to check out the scoring algorithm in GB6. The GB6 documentation says that there are two overall subsets (Integer and FP); within each the Geometric mean of the workloads is calculated, and then combined with a weight of 65% to the "Integer" GM and 35% to the "FP" GM. However, nowhere is it explicitly stated WHICH workloads contribute to which GM. I asked the Google AI , and it listed out the first 8 workload scores from the usual GB6 report as "Integer" and the last 5 as "FP" and said nothing about "Object Detection" at all. When I apply these to calculate two GMs combined in a 65%/35% proportion I do not get the same answer as GB6. The GM of all workloads without differential weighting is closer but the differences are still large. Please explain?