Short papers on the practice of structuring and deciding: what our models show, where methods disagree, what a case taught us. Numbered, dated, and kept. No news, no commentary, no marketing.
When every model is pointed at the same case, they rarely return the same order. Most treat this as a defect to be hidden behind a single number. We treat it as the measurement itself: the spread between methods is the honest width of the uncertainty, and a leader is entitled to see it.
Robustness is not a feeling about a result, it is a distance. We measure the smallest perturbation of the weights that reverses the ranking, and we publish it with the conclusion. A conclusion that flips under a rounding error is not a conclusion, it is a coin toss with extra steps.
The counterfactual is the most useful page of any arbitration file, and the least often written. It tells the winner why it won, the loser what was missing, and the judge that the choice was examined rather than assumed.
Bulletins are published when there is something to say, not on a schedule.
