Haar subleading structure

Beyond the leading $d_B^{-3/2}$ scaling of Result 3, the Haar-class disagreement exhibits subleading structure of the form $\mathcal{O}(d_B^{-2})$ that fits numerical data well across the tested dimension range.

We have not been able to derive an analytic form for this subleading term.

We chose to include this result anyway, with a clear statement that it is empirical, not analytic. The alternative would have been to hide it and leave the field to rediscover the structure later. That felt dishonest. Listed as Open Problem 3 in the paper.