Strategic Double Clicking

[GP2]

Originally Posted by Madpoo

Thanks to GP2 for noticing I now had 2 more bad results in my stats due to the "shift count+64 < exponent" bug. Whoops.

I think shift_count > (exponent − 64) becomes increasingly unlikely the higher the exponent gets. So it's mostly only an issue if redoing a large batch of small exponents.

In the current set of unverified LL residues there are only ten results with as many as four leading zeroes and none with more than that.

exponent residue
-------- -------
42540697 00001913CA986A__
46348957 00007E82ADF8CC__
48587443 0000774BAE66CD__
51265471 000070C8B21AAA__
56958731 000014EC228B1F__
59386493 000021A9AFD6CD__
61709519 00004D157AC8A5__
67765171 0000396D20B910__
69927623 000039A4D46421__
77594761 0000457BCA1899__

These are very likely correct residues coincidentally starting with leading zeros, and in any case not a priority to double-check since by its very nature this rare bug in v28.7 and earlier cannot cause a Mersenne prime to be missed.

[Madpoo]

Originally Posted by GP2

These are very likely correct residues coincidentally starting with leading zeros, and in any case not a priority to double-check since by its very nature this rare bug in v28.7 and earlier cannot cause a Mersenne prime to be missed.

Yeah, those are probably fine. I just looked and the last actual result (besides me) to get hit by this was the bad result here:
M37830997

There appear to only be a total of 9 results that were hit by this and I'm just lucky (or crazy) enough to have 5 of those.

Besides my 5 and the other one I mentioned, here are the other 3:
163367
274147
435763

Now I do feel kind of bad for those other 4 (besides me) that got marked bad by this...

George: Do you think we should retroactively and manually mark them as good (or "factored" in some cases) instead of bad? Kind of like the manual overrides for those funky residues on the smaller exponents?

I don't know what the range is for shift counts, but given the super low occurrence of this, I suppose we can just eyeball it from time to time and make sure no new cases show up. Nothing new (besides me) since George found it.

[Prime95]

Originally Posted by Madpoo

Now I do feel kind of bad for those other 4 (besides me) that got marked bad by this...

George: Do you think we should retroactively and manually mark them as good (or "factored" in some cases) instead of bad? Kind of like the manual overrides for those funky residues on the smaller exponents?

If you would like to manually change these 9, I'm OK with that.

[Madpoo]

Originally Posted by Prime95

If you would like to manually change these 9, I'm OK with that.

Alrighty, those 9 results have been modified to reflect the appropriate "good" or "factored" status as the case may be, not "bad". In each one, the first part of the residue had the zeros, but the last part of the residue matched, indicating it was a result of the bug and not a bad run.

[kladner]

I'll take this one:

40608851    1    0    2    0    2    0    DoubleCheck=40608851,72,1

[Mark Rose]

Originally Posted by Madpoo

Here's a new list of strategic double checks.

40609993	1	0	1	0	1	0	DoubleCheck=40609993,72,1
40633921	1	0	2	0	2	0	DoubleCheck=40633921,72,1
40635191	3	0	1	0	1	0	DoubleCheck=40635191,72,1
40683901	6	1	1	0	1	0	DoubleCheck=40683901,72,1
40755161	1	0	1	0	1	0	DoubleCheck=40755161,72,1
40842203	1	0	1	0	1	0	DoubleCheck=40842203,72,1
40874411	3	1	1	1	1	1	DoubleCheck=40874411,72,1
41065957	4	1	1	0	1	0	DoubleCheck=41065957,72,1

I took these, too.

[Mark Rose]

Originally Posted by Madpoo

Here's a new list of strategic double checks.

70623359	3	1	0	3	3	0	DoubleCheck=70623359,75,1

73915319	19	2	18	0	18	0	DoubleCheck=73915319,75,1
73915801	19	2	18	0	18	0	DoubleCheck=73915801,75,1
73916057	19	2	18	0	18	0	DoubleCheck=73916057,75,1
73919411	19	2	18	0	18	0	DoubleCheck=73919411,75,1
73919803	19	2	18	0	18	0	DoubleCheck=73919803,75,1
78037517	3	0	3	6	4	5	DoubleCheck=78037517,75,1
78042049	3	0	3	6	4	5	DoubleCheck=78042049,75,1
78042199	3	0	3	6	4	5	DoubleCheck=78042199,75,1
78042233	3	0	3	6	4	5	DoubleCheck=78042233,75,1

And these. Those last ones are going to be a while.

[richs]

My DC of 37093247 didn't match. Anyone like to TC it?

[Mark Rose]

Originally Posted by richs

My DC of 37093247 didn't match. Anyone like to TC it?

Queued. I'll have it done in a couple days.

[Madpoo]

Originally Posted by Mark Rose

And these. Those last ones are going to be a while.

Those 70M should get you some nice mismatches (and I think we can be sure yours will be the correct ones). Those came from that system that had some pretty nasty runs of bad residues here and there over the past several years, usually lasting a few months at a time, then getting better, then pooping out again, etc. They come from GPUs, which reminds me I was going to look at the rate of bad results for different kinds of CPUs/GPUs...

[richs]

Originally Posted by Mark Rose

Queued. I'll have it done in a couple days.

Thanks, Mark, for the verification.