The "open ended stops when you win" method, has to take into account contrary evidence into account. If you make 10 experiments, and 3 of those say your new and

The "open ended stops when you win" method, has to take into account contrary evidence into account. If you make 10 experiments, and 3 of those say your new and improve medication doesn't work, you have to take those into accounts, or else you're just cheating. The only remedy against publication bias based cheating is to publish everything, including the failures. That will take care of the "wait until I get a 1 in 20 fluke" just so you can get past the p<0.05 threshold.