I do understand the way that % credit is given, in that a quorum of results is established and those with closer results get more% of the credit. If 1 of the 3 is far off, then it is sent out to another again to re-run the wu. Typically, this is never an issue, except for those trying to cheat the system, or on the rare occurrences that a computer does compute incorrect results.
My question is as follows:
Would it benefit BOINC or even a specific individual, if that individual did generate some results that were inaccurate, or far enough out of the 90%tile credit, to be informed of such? I know we could discuss ways to do this, after the 27th time of an inaccurate result, or the first, or you do it on some log scale, or..etc., or KISS (keep it simple stupid), and do a return mail on each >75% inaccuracy, etc.,
But, this question is limited to only: Does it make sense to inform individuals if their computers are so inaccurate that their data is not used?
I don't know the # of computers that are giving bad results, but it is probably very low.

---

Thanks, and Keep on crunchin'
John Henry KI4JPL
Sevierville TN

I started with nothing,
and I still have some of it left.
<img src="http://www.boincstats.com/stats/banner.php?cpid=989478996ebd8eadba8f0809051cdde2">

Only results passing validation gets credit, and all results for a wu passing validation gets the same credit.
The Validator doesn't care whatever cpu-time or claimed credit for a result are when it comes to validating a result or not.

Users can look-up all their resently returned results, so anyone with many invalids can check this.

As for error-rates, don't know.
Still, according to THE redundancy question roughly 50% of all wu needs more than 3 results to be sent out before validated.
But, download-errors and results never returned before deadline is nearly all of these, with an ocassional crunching-error. With the fairly resent change with decreasing a computers daily_quota on error/missed deadline, and more users upgrading from v4.13, these errors should decrease.

> I do understand the way that % credit is given, in that a quorum of results is
> established and those with closer results get more% of the credit. If 1 of the
> 3 is far off, then it is sent out to another again to re-run the wu.
> Typically, this is never an issue, except for those trying to cheat the
> system, or on the rare occurrences that a computer does compute incorrect
> results.

Credit is given, or not given, based on how well work units match.

A perfect match is not required because an Athlon's FPU may produce slightly different results than a Pentium (or even different answers between a P2 and a P4, for example), but basically, the quorum consists of results that are "close enough" and results that are not close enough are thrown out.

Then the highest and lowest claimed scores are thrown out, and the granted credit is the average of the remaining scores.

Someone can "finesse" the scores slightly by always claiming an unreasonably high score. Their claimed score will always be thrown out, and the 2nd highest score will be used as part of the average.

For example, let's say four work units are returned, and the claimed scores are 100, 20, 22 and 28. The 100 will get thrown out, the 20 gets thrown out, and all four machines get 25. If the machine claiming "100" should have claimed 16, then we've got 18, 20, 22 and 28, and the granted credit should have been 21.

Once the granted credit is determined, everyone returning valid results by the deadline will get that score.

---

Hi all,
All of the quorum, and how it gets applied, when, etc., I do understand.
I just wondered if it wasn't viable to tell those that aren't sending inv alid results, that there may be an issue.
Me for one, I really don't know if I am making valid results or not.
I can only assume that if I get credit, then I have sent a valid.
But after checking 200 results, I prefer to simply leave it running, and assume all is fine, without checking results.
I know this number of people that send back results that send back bad results is very very small, but thought it might help, in case someone changes to a new machine, or changes a process on their computer, etc., it might help them get back in line quicker.
No responses needed to this, as it is only a suggestion, and can be trashed into the bitbucket at will.

Thanks,
John Henry

---

Thanks, and Keep on crunchin'
John Henry KI4JPL
Sevierville TN

I started with nothing,
and I still have some of it left.
<img src="http://www.boincstats.com/stats/banner.php?cpid=989478996ebd8eadba8f0809051cdde2">

> Hi all,
> All of the quorum, and how it gets applied, when, etc., I do understand.
> I just wondered if it wasn't viable to tell those that aren't sending inv alid
> results, that there may be an issue.
> Me for one, I really don't know if I am making valid results or not.
> I can only assume that if I get credit, then I have sent a valid.

That is the assumption we all make.

> But after checking 200 results, I prefer to simply leave it running, and
> assume all is fine, without checking results.
> I know this number of people that send back results that send back bad results
> is very very small, but thought it might help, in case someone changes to a
> new machine, or changes a process on their computer, etc., it might help them
> get back in line quicker.
> No responses needed to this, as it is only a suggestion, and can be trashed
> into the bitbucket at will.
>
> Thanks,
> John Henry
>
I agree with what you are saying. Some people with large farms are unable to check all their computers to see if they are working properly. I have seen some computers putting out nothing but error results and the owner probably didn't even know. Maybe the developers can find some way to let people know if there is a problem with a computer.