Scientists from the Belle and Belle II collaborations used collisions of electrons and positrons to look for extremely rare decays of B mesons that produce invisible particles (neutrinos) or violate the rule that lepton types do not mix. Working with a combined data sample of about 1.2 inverse attobarns (the paper cites datasets corresponding to 711 fb⁻¹ from Belle and 365 fb⁻¹ from Belle II), they searched for several rare processes and report evidence for the decay B⁺→K⁺νν̄ at a significance of 3.3 standard deviations. They also set new upper limits on other channels that remain consistent with the Standard Model or are still too small to see with the present data.

The experiments run at the Υ(4S) resonance, which almost always produces a pair of B mesons and gives a very clean environment for these searches. Because neutrinos do not leave a direct signal in the detector, the teams infer their presence by fully or partially reconstructing the other B meson in the event (the “tag” B). Two tagging strategies are used: inclusive tagging, which keeps more events but has more background, and hadronic tagging, which fully reconstructs the other B with the Full Event Interpretation algorithm and gives stronger kinematic constraints but with low efficiency (often below 1%). Machine‑learning classifiers called boosted decision trees are used to suppress backgrounds and to extract the signal.

For the B⁺→K⁺νν̄ channel the groups reinterpreted an existing analysis in a model‑agnostic way. They used the Weak Effective Theory, a formalism that parametrizes possible new interactions by a few numbers (called Wilson coefficients), and reweighted simulated signal events to test different possibilities. Their fit prefers an enhanced vector contribution together with a nonzero tensor term compared with the Standard Model expectation. The fit peak in the three‑parameter space is reported and 95% credible intervals are given for each combination of coefficients. The observed excess corresponds to a 3.3σ significance, which is evidence but below the usual 5σ threshold used to claim a discovery.