[yosito]

I agree with you that this is p-hacking, but this makes me curious. Could p-hacking be used in a positive way to find obscure correlations which can be used to generate novel hypotheses which can then be replicated/tested independently?

[pas]

A p value close to the significance level ought to mean we need more data. Journals should start to reject papers that do experiments with bad designs.

If there's a real effect, with more data the p value will decrease, or the signal will go away. Uncertainty decreases.