It doesn't seem that hard to generate entirely plausible fake data using simple randomized algorithms, that wouldn't leave any particular signature. I don't wa

It doesn't seem that hard to generate entirely plausible fake data using simple randomized algorithms, that wouldn't leave any particular signature. I don't want to actually publish fake data, of course. But it could be fun to have a contest, where one team cooks up fake data and the other team tries to detect it. You could take real papers, scrub all the experimental data out of them, and people try to cook up fake data to support the conclusions. If the other team can't reliably decide which paper is fake, the first team wins.