You failed to make your point there, too :-)

Any error could potentially be important. The fact that most of the results are irrelevant doesn't change that.

Good science is worth infinitely more than bad science. Let's do this thing properly.

one of the basic tenets of science is the experiment has to be "reproduceable" so repeating the experiment and getting the same results is very important. i doubt that less than 3 work units will ever happen, wot if u ony do 2, how do you know which one is best, because we are using computers from all over the world with different setups, some of which may be dodgy then it is reaonable to insist on multiple jobs. if it was beig done on an "inhouse: supercomputer where inhouse tech experts monitor everything then maybe the quorum method would be unecessary. IMHO i tend to agree with you but the scince has to be sound , had to be trusted by scientists and by nature they are difficult to convince ( which is a good thing) ..if we can do this and get fabulus results with qorum three then who knows maybe there will be some other way to validate results.. one small error could be a disaster, wot if somwe dodgy computer failed in a job and missed "the only possible cure for disease x etc" ?? all the millions of years of crunching wasted because of stuffed up result :(
i know i started this thread and am keen for improvements in efficiency also but they have already gone from quorum 4 to qorum 3 which is huge improvement in efficiency but then only after proving wot a good job the crunchers were doing , keep up the good work cheers

Hello dhatz,
I thought about mentioning Rosetta@home, but decided that would be unnecessary. I was wrong. So here goes . . . Rosetta@home is testing new algorithms on proteins of known structure. Their main interest is getting a predicted structure as close to the known structure as possible. So they run with no duplication, then pick the best prediction and verify it by rerunning it on their computer. So they are no more trusting than anybody else. Every result that they care about is validated this way - - 100% duplication. We are folding proteins without a known structure, so we have to validate every prediction, not just the best prediction. We cannot rerun everything on known good computers.

Right now we need 3 identical results for validation. In theory, we could accept 2 identical results, sending out another 2 copies only if the first 2 disagreed. (This would effectively be the same as Rosetta@home, except that we would care about every result, not just the best result.) But there is no way we can get below 100% duplication. I have no knowledge of just what our error rates are, but I am glad that we have dropped from 4 to 3, since I have always considered 3 to be a 'gold standard' for validation.

Lawrence