[QUOTE=GP2;486231]There are 81 self-double-checks and 25 self-triple-checks by "Phil Frakes" and "pfrakes", assuming this is the same person.

I am retesting ...[/QUOTE]

Could there be less obvious cases of self double-checking? Can the project detect when user A's tests are always/mostly/frequently double-checked by user B?

Two from Curtisc:
[LIST][*][m]72391931[/m][*][m]67768007[/m][/LIST]

[QUOTE=masser;486263]Could there be less obvious cases of self double-checking? Can the project detect when user A's tests are always/mostly/frequently double-checked by user B?[/QUOTE]

1. Yes

2. It could get complicated. Suppose there's implemented on the server, a filter that flags A and B running more than x% of common exponents out of their individual totals. Depending on the number of active users and frequency of checks against x%, that could get computationally expensive. (If limiting to the past year's contributors: Cache ~7382 lists of exponents run. Check N (N-1)/2 = 7382 * 7381 / 2 = 27,243,271 combinations of two accounts for commonality of primality tests. Doing a quick eyeball average, ~500 LL & DC results per day for the past year; 500 * 365 / 7382 is 25 per user average. Search User A's list for User B's nth exponent, for 1 thru n, counting matches and entries; for each of the 27.2 million user combinations; around 7.4 billion comparisons.. It gets worse by far if doing 3-way checks, etc.)
A user notices that x=80% produces early triple checks, or an end to credit for self-checks. The user adjusts by taking out another account to get under the x% threshold by spreading the action among accounts A, B, and C. So x gets dialed downward to 40%. Rinse and repeat. At some point the net starts snaring legitimate innocent users, and causing a lot of increased overhead for the administrators.

[QUOTE=kriesel;486421]2. It could get complicated. Suppose there's implemented on the server, a filter that flags A and B running more than x% of common exponents out of their individual totals. Depending on the number of active users and frequency of checks against x%, that could get computationally expensive. (If limiting to the past year's contributors: Cache ~7382 lists of exponents run. Check N (N-1)/2 = 7382 * 7381 / 2 = 27,243,271 combinations of two accounts for commonality of primality tests. Doing a quick eyeball average, ~500 LL & DC results per day for the past year; 500 * 365 / 7382 is 25 per user average. Search User A's list for User B's nth exponent, for 1 thru n, counting matches and entries; for each of the 27.2 million user combinations; around 7.4 billion comparisons.. It gets worse by far if doing 3-way checks, etc.)
A user notices that x=80% produces early triple checks, or an end to credit for self-checks. The user adjusts by taking out another account to get under the x% threshold by spreading the action among accounts A, B, and C. So x gets dialed downward to 40%. Rinse and repeat. At some point the net starts snaring legitimate innocent users, and causing a lot of increased overhead for the administrators.[/QUOTE]But the motivation is lacking. Primenet credits have no monetary value. There is not an endless line of females/males/whatevers desperate to "make a connection" with high credit value users. There is no fame for having big credit numbers. So one wonders what could possibly be the motivation for such behaviour?

I'd much rather the admins spend their time on other things than improbable scenarios.

Edit: If one thought the possibility was real, then there is probably an easier to find a cheat. Look for user that has all of their first time tests verified by users that have manually selected the exponent. A real user will have the verification done by some random person chosen by the server. But a cheat would have to manually select which exponent to verify to ensure they get the exponent they want.

[QUOTE=GP2;486231]There are 81 self-double-checks and 25 self-triple-checks by "Phil Frakes" and "pfrakes", assuming this is the same person.[/QUOTE]

Good catch. I did some work to find cases where exponents were done by the same person if they had "synced" their Primenet v4 and v5 accounts, but not everyone had done that apparently.

[QUOTE=ixfd64;486193]User "SalVi" just self-verified a few exponents:
[LIST][*][m]83406347[/m][*][m]83406263[/m][*][m]83406203[/m][*][m]83406289[/m][/LIST][/QUOTE]

Yeah, I've noticed a few users doing their own verifications, on purpose (not a "by accident" getting the same exponent assigned). Sigh... I had been keeping up with those with my old posse of servers but there were only a handful per month so it wasn't too bad.

It'd be great if people didn't do that because I tend to think it ruins the appearance of integrity (even if it's all on the level, it just doesn't look good) of the result, unfortunately.

[QUOTE=retina;486457]But the motivation is lacking. Primenet credits have no monetary value. There is not an endless line of females/males/whatevers desperate to "make a connection" with high credit value users. There is no fame for having big credit numbers. So one wonders what could possibly be the motivation for such behaviour?

I'd much rather the admins spend their time on other things than improbable scenarios.

Edit: If one thought the possibility was real, then there is probably an easier to find a cheat. Look for user that has all of their first time tests verified by users that have manually selected the exponent. A real user will have the verification done by some random person chosen by the server. But a cheat would have to manually select which exponent to verify to ensure they get the exponent they want.[/QUOTE]

More likely, they don't even get an assignment at all and just run the test and turn in the result. When it comes to people who deliberately self-verify, I've seen both cases. Getting the assignment (by starting it and then checking in which "forces" Primenet to create the assignment when normally it wouldn't), or by just running it again "off the books" and submitting the result.

And that's for cases of the same user doing both runs. To find cases where different users, that may just be sock puppets of the same person, turned in results for the same exponent, it would be trickier. If they used the same pair of users, as mentioned, it's something we could query for and say "hey, that's weird, these two users seem to verify each other's work quite a lot". That's something I may explore...

[QUOTE=ixfd64;486235]I'll rerun P-1 on the poorly factored exponents:[/QUOTE]

I've finished rerunning P-1 on the exponents. Didn't find any factors.

I managed to double check my own result again by mistake :sad:

[url]https://mersenne.org/M47839741[/url]

[QUOTE=Uncwilly;486197]I'll take the first 2.[/QUOTE]They are on track to complete Tuesday (subject to IT rebooting the PC during maintenance.

[QUOTE=ATH;486751]I managed to double check my own result again by mistake :sad:

[url]https://mersenne.org/M47839741[/url][/QUOTE]

Avatar checks out. :P

On another note, I'll rerun P-1 on that exponent on the off-chance that the original run missed a factor.

[QUOTE=ATH;486751]I managed to double check my own result again by mistake :sad:

[url]https://mersenne.org/M47839741[/url][/QUOTE]

I will do this.

I'm frankly amazed that the shift count for your second double check was a mere "284". What are the odds?

I would work on some of the other exponents mentioned here recently, but spot prices for c5 instances in AWS us-east-2 have soared lately and it's rather discouraging.