In theory, you should count only those with score > 0 (see the first post here).
I rechecked and found that my number N=3350 was wrong (I counted those with more than 0% completion).
My bad, but those with positive score are 3356, the difference is insignificant.
And I did try the hypothesis that the rule changed and one should count all participants (so that N=3963).
This would give 3761 CP.
None of these fits.