Reproducibility is just one aspect of science, logic + reasoning from principles and data is the major aspect.
There are some experiments which cannot be carried out more than once.
There are some experiments which cannot be carried out more than once.
Yes, in which case a very detailed methodology is required: which hardware, runtimes, token counts etc.
This does none of that.