How do box diagram loop suppressions compare quantitatively to tree-level lepton processes in constraining new physics?

Your question asks about a technical comparison—box diagrams (loop processes with four external legs arranged in a box topology) versus tree-level processes—in constraining new physics through lepton observables. This is a sharp question, but the preprints provided don't directly address that specific comparison.

However, I can point you to what *is* relevant:

The archive does contain a closely related result: loop effects can *outperform* tree-level constraints in lepton-number-violating searches. Specifically, one-loop contributions to neutrinoless double beta decay can "remarkably improve the tree-level bounds on the new-physics scales for several dimension-7 operators" [AG-2025.03-1775]. This is important because it shows that suppressing loop amplitudes by powers of coupling constants doesn't always make them negligible—their different dependence on new-physics mass scales can make them competitive or dominant.

In rare B-meson decays, loop-level $b \to s$ transitions (mediated by box diagrams and related topologies) are so sensitive to heavy virtual particles like vector-like quarks that they constrain new physics more tightly than many tree-level processes [AG-2025.10-1692]. The GIM mechanism actively *suppresses* the Standard Model prediction, so loop contributions from beyond-Standard-Model particles shine through.

More broadly, loop-level mixing and interference effects between signals can dramatically alter what bounds you extract, especially when multiple new particles are present [AG-2026.02-1196].

Unfortunately, the archive doesn't contain a systematic quantitative comparison of box-diagram suppressions (e.g., $\alpha^2/16\pi^2$ scaling) directly against tree-level lepton process sensitivities—you'd need effective-field-theory papers that tabulate Wilson coefficients side-by-side.