The research state
We don’t
know yet.
Eight candidates. Each one has a computation attached to it and a decision rule that was fixed before any data existed. The statuses below are recomputed from storage every time this page loads — including when the answer is that there is not enough yet to say anything.
There is no proven column, and there will not be one. A claim that has survived everything thrown at it is supported — which is a statement about what was thrown.
Computing from the dataset…
What makes one model more justified than another when the evidence does not decide?
And how would we know whether a structure we keep seeing was discovered or imposed?
If this works, what will exist that doesn't now
If this works, what exists at the end is evidence about which criterion people actually reach for when the evidence runs out — whether that criterion is stable across domains, whether people distinguish confirming from discriminating observations, and whether a language model given the same material reaches for the same one. A clear null result on any of those is a result.
Why every candidate carries a rival
A candidate with no rival is not a finding — it is a description of the data that happens to be phrased as a discovery. The rival field is the cheapest available guard against that, and in three of the eight cases above the rival is currently at least as plausible as the claim.
The method page has the pre-registration and the limitations. The audit is where the connections themselves get attacked.