Funny (probably fake...) result

[Prime95]

I'd toss the one obviously bogus/typo result. We have no reason to suspect wholesale fraud here.

[Madpoo]

Originally Posted by Prime95

I'd toss the one obviously bogus/typo result. We have no reason to suspect wholesale fraud here.

The one obviously bad one was tossed, but then I did wonder about a few others. I think I mentioned above somewhere that the user in question never seemed to do any actual work above 2^75, and yet the only time they ever reported any "no factor" stuff above 2^75 is when they also checked in a factor at some higher bit level.

It made me suspicious of those. I actually spent a little time the other day looking at this again, and there could be more, but honestly it's too hard to nail down for sure. To me, it may be suspicious if the user finds a factor by TF, supposedly, and happens to check in a bunch of "no factor found" in several lower bit ranges, all at the same time. By way of comparison, I noted that when no factors were found at all, they would check in "nothing between 2^72 and 2^73" and then a later "NF between 2^73 and 2^74" and those would be turned in at different times. It was only when a factor was found that I'd see a bunch get turned in all at once.

But again, that's not definitive, so ...

[LaurV]

I am more inclined to suspect what axn said in post #16: the user mistakenly believes that the P-1 finds the smallest factor, so when he finds one, he "helps" us by reporting "no factors" for the smaller bitlevels, so we should not waste time to check them (assuming he is not just trying to "help" his credit, mistakenly or intentionally). As long as this is not disadvantageous for the project (remember? the goal is to find primes, and any found factor eliminates a possible LL candidate!) I would vote to let him be, but keep an eye of him for the future. I won't waste time to dig into his history (I already tried to double-check his reported TF for the easy part - lower bitlevels - and I came out empty handed).

[sonjohan]

As I haven't had any GPU-TF hitting a factor found, I redid a test some time ago with a number that did have a factor, and the GPU-TF did find it again, so I hope it still works correctly.

Maybe an idea to randomly add a known factor test in the runs, just to make sure that the person submitting results isn't faking them?

Still, I haven't been doing any GPU-TF recently due to holidays, extreme heat, ...

[chalsall]

Originally Posted by Prime95

I'd toss the one obviously bogus/typo result. We have no reason to suspect wholesale fraud here.

Actually, with all due respect, we do have reason to suspect fraudulent behaviour in this particular case and this particular user.

This won't impact the end goal of GIMPS (read: finding the next MP, since missed factors below a found factor have no impact), but some would like the database to be complete as to where there are factors and where they are not.

This doesn't really matter to GIMPS, admittedly.

But some don't get out much; we're comfortable with that.

[Prime95]

Originally Posted by chalsall

Actually, with all due respect, we do have reason to suspect fraudulent behaviour in this particular case and this particular user.

I admit I haven't followed the thread meticulously. For me, the most plausible (innocent?) explanation I saw was that the user had a typo in submitting his result (i.e. mistakenly edited 2^79 to 2^89 instead of 2^80).

The innocuous explanation is the user took a peculiar interest in this exponent and simultaneously ran GPU TF and P-1 on the exponent. P-1 was successful and GPU TF ran to 2^80. The GPU TF may have taken place on another or several other machines. Rather than go to the remote machines to report the actual TF results, he found it easier to edit previous GPU results to reflect the work he had done but made an error editing the mfaktc text.

[Dubslow]

Originally Posted by Prime95

I admit I haven't followed the thread meticulously. For me, the most plausible (innocent?) explanation I saw was that the user had a typo in submitting his result (i.e. mistakenly edited 2^79 to 2^89 instead of 2^80).

Similar results have been noted for numerous other exponents by the user. Either they are "accidentally" making a typo in hundreds of results submissions, or something fishier is going on. (axn provides one explanation which preserves the innocence of the user, but even so, the records themselves should be expunged/double checked/otherwise ameliorated.)

[axn]

Originally Posted by Prime95

The innocuous explanation is the user took a peculiar interest in this exponent and simultaneously ran GPU TF and P-1 on the exponent. P-1 was successful and GPU TF ran to 2^80. The GPU TF may have taken place on another or several other machines. Rather than go to the remote machines to report the actual TF results, he found it easier to edit previous GPU results to reflect the work he had done but made an error editing the mfaktc text.

Unlikely. The user has several P-1 assignments on nearby exponents which they've duly completed (without any TF work). Most logical explanation is, they're submitting "no factor" results for their P-1 factor finds.
Maybe they could be doing TF after P-1 found a factor (just looking for smallest factor)? Perhaps. Until you realize just how incredibly time consuming these higher level TFs are.

[Prime95]

Originally Posted by axn

Unlikely. The user has several P-1 assignments on nearby exponents which they've duly completed (without any TF work). Most logical explanation is, they're submitting "no factor" results for their P-1 factor finds..

I searched for all exponents where the user submitted a factor and one or more no factor results and the CPU credit was more than 10 GHz days. I kept the factor found result (without editing the results line which is likely hand edited) and deleted the no factor results.

50 no factor records have been deleted

60 CPU credit records have been deleted

[snme2pm1]

Originally Posted by Prime95

50 no factor records have been deleted

Will the copies at mersenne.ca also need to be hand deleted?

[LaurV]

Originally Posted by snme2pm1

Will the copies at mersenne.ca also need to be hand deleted?

only James can answer to that...