I'm wondering of how does SETI @ Home server verify that the data returned by the clients are actually correct and not fake? That is, how does it know that the computing power has been really spent on calculating the actual data and not just randomly generating them? Does it send the same sample to different clients to see whether they return the same reply? Or is there some other, better mechanism of verification?

Comparison with others' results is the only verification that I know of on a workunit-by-workunit basis. It has also been verified that there are test signals injected into the data to ensure that they are being detected and reported, to show that the system is working from start to finish.

Each unit is sent to a minimum of two random hosts. If their results agree then the unit is marked completed. If not is it is sent to a third, then a fourth, to a maximum of ten if I remember correctly until there is quorum. If there is not by then it is flagged as erroneous and removed.

---

Thanks, sounds reasonable. And the test signals is a good idea too.
Is there any documentation about these verification mechanisms and how is the credit calculated then?

How credit is calculated. This system "CreditNew" pays out less than the old system (the old one from the Cobblestone definition that a gigaflop machine will earn 200 credits a day we get that each credit is 432 billion floating point operations) so it has led to much complaining. :^) (It pays as little as half depending on platform.)

---