The total neutrino flux from all supernovae is a barely discernible whisper amid the roar of cosmic rays and earthly interference. Each explosion hurls out neutrinos that, stretched by the expansion of space, lose energy but retain information about stellar evolution. The idea that this is how neutron stars are born was first proposed by Fritz Zwicky and Walter Baade in the 1930s, while Hans Bethe explained the mechanism behind stellar glow.

The KamLAND experiment — an underground detector with a liquid scintillator (it flashes when a particle hits) — accumulated data for 22 years. The useful signal is drowned out by atmospheric neutrinos, which are millions of times more abundant. To extract it, scientists employed a neural network called KamNet, trained on simulated events to recognize characteristic patterns — much like smart noise cancellation. After filtering, only seven suspicious events remained (with about sixteen expected by chance) — no significant excess. This allowed a twofold improvement in the upper limit: the flux does not exceed 38–43 neutrinos per second per square centimeter, the tightest constraints for low energies.