The range of means by which that can be done is huge (though considerably less than "anything goes") and there is a definite role for work that is exploratory and speculative, up to and including stuff that is almost certainly wrong but worth publishing because a) the error is not obvious and b) publishing creates the opportunity for others to respond to it, hopefully putting the error to rest for good and all. Some of the early work on the "no cloning" theorem was motivated by publications of this type (it turns out if you could clone a quantum state you could use entanglement to communicate faster than light, and there was a series of papers in Physics Letters in the late '80's proposing to do just that.)
So trusting in the self-correcting ability of the discipline of science is fundamental to its progress, and therefore insisting on some alchemical standard of "proof" as the goal for publication would fatally cripple the scientific enterprise. For science to work we have to be tolerant of the publication of error.
But we need to keep the rate of erroneous publications down to a manageable level. Peer review and society membership were ways of doing this in the past. They have broken down today, and we are still casting about for new ways to keep the error rate well above zero, but not so high as to swamp everything else.