Arguments are presented in favour of the interpretation of cosmic ray jets as composite collisions, in a single act, of the primary with a compound of target nucleons. In particular, it is pointed out that the identity of «successive» collisions is lost at high energies, because the expanding meson cloud enwraps the compound before ejecting the bulk of the secondaries. According to this view, the primary energies assigned to the jets by the customary procedures are too low in many cases, and especially for events with large numbersN h of heavy prongs. The resulting difference in the true primary energies of the jets with, say,N h ≤ 3 andN h > 3 accounts for the apparently excessive frequency of the events of lowN h . But if one wishes to retain the model of the nucleus as an aggregate of corpuscles, the cross section of the avalanche of interacting nucleons will increase with penetration, and the usual picture of the interaction volume shaped like a cylindrical tunnel must be replaced by that of an inverted funnel. This «funnel model» predicts a stronger dependence of the multiplicities on the atomic weight of the target nucleus, and seems to fit the—rather scanty—experimental data somewhat better. It is then shown that according to this model, the larger part of the observed multiplicity spread is not due to the «fundamental fluctuations» in the emission of secondaries in a nucleon-nucleon collision, but to the randomness in the number of particles involved. For nucleonnucleon interactions at about 5·1012 eV, the «fundamental fluctuations» can be approximated by a Gaussian with a widthσ = 6.5.