Onset and Threefold Classification of Outcomes of Spontaneous Splitting in N-Body Relational-Wave Closed Systems: Exponential Decay of the Parent State at a Seed-Independent Rate under Generator Unfreezing, the Scale-Anonymity Axiom, and Outcome Stratification by the Conserved Ratio c

Paper 4 of this series established a no-go: under a fixed generator, neither closure nor energy-like quantities move between rotation planes, so spontaneous splitting of a state cannot occur. Starting from this boundary, this paper asks whether waves can split spontaneously under the closure condition and whether the subsequent outcomes can be classified. Three preparations are made. First, scale anonymity is introduced as Axiom 0.5: no construction may reference an absolute scale, and everything must be covariant under Z -> lambda Z; the zero right-hand side of the quadratic closure is the unique choice compatible with this axiom within the family sum x_n^2 = c. Second, it is proved that every smooth closure-preserving flow has the form dX/dtau = K(X)X with K skew-symmetric, so the fixed generator is a special case and the search space for minimal extensions is closed. Third, the minimal intrinsic extension is identified: the phase-difference sine generator is reconstructed each step from the current phases (unfreezing); being zeroth-order homogeneous, it satisfies Axiom 0.5 without adjustment. Spontaneity is defined operationally as dynamical instability. Numerically (N=4, seed amplitudes delta = 1e-2 to 1e-5), the dormant fraction grows exponentially at a rate independent of delta over three decades; absolute-threshold crossing times are linear in log(1/delta); and even the delta = 0 exact periodic orbit collapses from floating-point rounding noise alone, amplified over about 20 orders of magnitude, while the real and imaginary parts of the complex bilinear closure are conserved to machine precision. A large-N onset experiment (N = 40, 300, 1000; up to 499,500 relational waves) reaches seed depths f(0) ~ 1e-30, measured without cancellation via projection onto the orthogonal complement of the parent plane, and observes more than 20 decades of constant-rate exponential amplification; the rate depends only weakly on N. Outcome-classification experiments (N = 3-6, 20,000 steps) find three classes - expansion, bounded mixing, and recurrence - with a sharp N phase structure (no expansion from zero-closure states for N <= 4; expansion in all trials for N >= 5). A controlled contrast experiment shows that the conserved dimensionless ratio c = |Z^T Z| / ||Z||^2 stratifies the outcome classes at small N: closure is not only a conservation law; its conserved value selects the outcome. A spontaneous exponential decay of the second singular direction observed for N = 4 suggests that splitting can also stop without external freezing, which is the subject of the sequel.