In message <x7ema6xwp7.fsf@capsicum.wsrcc.com>, Wolfgang Rupprecht writes:
>Couldn't they just throw in a handful of known test cases in the
>problem mix and see if the people calculated the right answer?  Any
>answer from bogus / malicious hosts could be discarded.

Short answer:  The "answer" might well be right for any given collection
of test cases, even though a client would introduce errors on *some* set of
inputs.

Long answer:  Read the thread in alt.folklore.computers.  Basically, they're
happy with their current technique, and it's probably a good one.

-s