I see the trial and error solution but I doubt it's possible to solve it in general...lot of cases based on the differences between the guesses and how far off the #s are.
I don't have the exact numbers I first saw for this problem. This is how it goes.
Say there are M jelly beans in a jar, and say 5 people guessed 107, 96, 122, 120 and 106. Someone tells you that the numbers are off by 11, 26, 5, 15, and 16 (these numbers probably don't work). You don't know which error goes with which guess. With this size problem we can solve it using trial and error, but how would you solve it in general?
Well, I suppose in general is the wrong phrase. There should be some way to do it if you had 1000 guesses and 1000 error reports. My initial thoughts are to set up a linear system, where your unknowns would be -1, 0, or 1 and your coefficients would be the different errors. but I can't quite get it too fit