Even in the hard sciences, the results are fundamentally about statistics. And they often require very expensive machines for reproduction.

I'm not sure if it is a good idea to try to optimize the scientific process for fraud resilience. In the end, we still rely on people doing the right thing most of the time.