Reasoning. The renormalization ambiguity in T_ab is handled by analogy with gravitational-energy pseudo-tensors: expressions differing locally agree when integrated over large surfaces. Choosing normal ordering with respect to the late-time Killing vector near future null infinity, the average outward flux between retarded times evaluates to (1/2pi) Integral d(omega) omega Gamma_omega / (e^(2pi*omega/kappa) - 1) — exactly the thermal emission rate of §2. Conservation transports this to every constant-r surface, so an equal negative flux crosses the horizon (a local weak-energy-condition violation, unobservable locally, of the size allowed by the ~M^-4 indeterminacy of the local energy density there). Consequences: the classical area theorem is evaded, the mass decreases slowly (evolution rate << light-crossing time while M >> Planck mass, justifying a sequence-of-stationary-states treatment), and integrating dM/dt ~ -M^-2 gives lifetime ~10^-28 M^3 s. Baryon and lepton number of the collapsed body are not returned: the rest-mass energy comes out as thermal radiation. The particle creation is shown to be a global, not local, process: an observer falling through the horizon sees no divergent flux (occupation of the smooth horizon-crossing modes is small), removing the objection that emission would have to originate implausibly from the collapse epoch.
Step 6 — validity verdict
approved / checked. Reconstruction — premises: a renormalized stress tensor that is conserved, stationary at late times, and agrees with normal ordering near future null infinity; the thermal outward flux of A-38. Load-bearing steps traced: (1) conservation plus stationarity transport the same integrated flux through every constant-r surface, hence an equal negative flux across the horizon — valid bookkeeping; (2) dM/dt proportional to -M^-2 integrates to lifetime proportional to M^3, and the 1e-28 s g^-3 coefficient is consistent dimensional bookkeeping (it reproduces the standard ~8e71 s solar-mass lifetime and the ~1e15 g decayed-by-now threshold); (3) evolution rate << light-crossing rate while M >> M_Planck justifies the quasi-stationary sequence, and the derivation honestly stops claiming validity near the Planck mass. The deep premise — that all admissible renormalizations agree on large-surface integrals — is assumed, not proven here; its truth is priced downstream. Conditional on it, every step checks.