There's an explanation for this.
Thanks to previous searchers, we have some data available on the internet. Referring to Louie's comment as quoted above, I've collected the data available on the net for previous searches. I then reorganized and parsed them so that we can use their residues data for our purposes.Originally posted by jjjjL
i realize that this will represent a triple check of some of the very low numbers, but in most cases, those numbers even if they have residues will take very little time to redo. if you still have contact w/ any of the previous searchers, they are welcome to email me residues for comparison and that would help me avoid triple checking. the whole idea is just to get a pair of matching residues so that we can eliminate the number for good... and eliminate it for sieving too.
The parsed data includes 6279 residues for;
k=4847, 1000431 < n < 2099751, 4083 residues,
k=19249, 1046318 < n < 1192802, 252 residues, and
k=28433, 1150393 < n < 2003785, 1944 residues.
But, this does not mean we've eliminated 6279 candidates for secret tests. We've eliminated 5277 secret tests in this manner. I guess, the main difference comes from the difference in sieving efforts. Some of the candidates previously prp tested by other searchers are already eliminated with our sieving efforts.
Also, there were a couple of cases where we've already tested some of the candidates. And, the good news is, the residues match for these couple of tests. So, there seems to be no error while parsing and importing the data.
Anyway, in terms of what we'll do next, the remaining is straightforward. These numbers will be checked by supersecret only when the time comes, instead of being tested by both secret and supersecret. If the residue of supersecret will match that of previous searches, then we'll be done with that specific candidate and there will be no need for a third test. If the residues will not match, then a third test will be done.
PS: If you look at the highest untested figures, you'll see that the n numbers do not necessarily match to the data I've given above. For example, we have n=1186863 for k=4847, simply because it is the highest n figure where previous searchers did not make the residue data available. (Check it here.)
I hope this explains the change in secret stats.