Two papers with DOIs. Five open-source repositories. Ten published measurements, including the ones that failed. A daily feed that runs in public. And, at the bottom, the list of what we have not done — because a page like this is worthless without it.
Both filed the same morning, under one researcher identity: ORCID 0009-0003-6818-6737.
Published because a protocol nobody can check is a claim, not a protocol. None of this is a moat. It is the part we would rather be judged on.
House rule for the measurements: if the null model kills the hypothesis, the write-up says so with the same volume it would have used had the result gone our way.
The split is stated because the heading used to read “ten measurements”, and four of these are not measurements. This is the page that asks you not to take a claim on trust; our own count has to survive being clicked as well.
A daily security read, published in English and Czech, free and without registration or tracking. Every edition carries its own funnel — how many items were seen, how many cleared the threshold, how many shipped — and the page states when the scoring instrument last changed, because a trend across a recalibration is not evidence. Today's edition →
If a claim about us cannot be clicked, it should not be believed — including when we are the ones making it.