I'd also love to see a journal that agreed to publish work before any results are known. Researchers would submit hypotheses and methodology for review, and the journal would publish the results after the experiments were conducted, regardless of their outcome.
It would still incentivize interesting hypotheses, but wouldn't lead to results-biased publications.
There is a workshop[1] at NeurIPS this year (2020) for experimenting with this model. I hope it is adopted more widely, especially in the ML community to disincentivize the +1-2% performance increase in SOTA papers.
You'd do the work, then sign up, wait a bit (while working on your next project) then you submit the results.
This is what happens with many grants anyway. You propose to research the stuff that you already understand, you get funded and can finally work on new stuff you don't understand well enough to get funded for.
It would still incentivize interesting hypotheses, but wouldn't lead to results-biased publications.