The Third Verdict against Dynamical Geometry: Is the Modular Record of Matter Coupled to a Quantum Incidence Register Irreducible to an Effective Background?

Phase VI closed the fixed-background branch with a structural obstruction (Proposition 1): for a field on a fixed classical background the state is a functional of the background data, hence so is every modular observable, and grade (ii) is the ceiling. Exactly two bullets survive: (a) genuine interaction, and (b) dynamical gravity — the case in which the metric ceases to be a c-number. Phase VII/VII-b closed Exit (a) with a certified collapse. This paper attacks the last bullet, Exit (b), in the minimal form in which it is tractable and honest: the incidence relation itself — the bond couplings of a 1+1 lattice — is promoted to a quantum register entangled with matter, with its own transverse dynamics. No continuum gravity, no holography, no islands, no JT. Two structural results precede any sweep. Tier 1: any self-consistent c-number geometry reproduces, to machine precision (|1−F| ≤ 4. 4×10⁻¹⁶), the matter state prepared on the same profile imposed externally — semiclassical backreaction is a background functional and is closed by construction. The blindness theorem: Trg ∘ Δg = Trg for any dephasing Δg of the geometry register, so every single-time matter observable — including this phase's own preregistered discriminators — is invariant under classicalization of geometry, and its grade ceiling is (ii) by construction. The price is not hypothetical: on the single-time grid the frozen primary Δr clears its threshold by two orders of magnitude in all twelve cells, so a phase that had read it would have reported a signature. This is the Bonder–Okon–Sudarsky critique of single-time signatures, reproduced inside our own instrument, and it forces the decision onto the programme's founding principle: the signature lives in the record, never in a single-time reduced matrix. The deciding layer is therefore a quench record: the quantum branch against a classical-geometry twin (the central incidence register dephased at t = 0), with a record divergence R (t) (trace distance) and a history-irreducibility Dₕist (the cost, in the GNS metric, of forcing one shared effective background to fit the whole history). Teeth pass: R ≡ 0 at g = 0 by exact decoupling, and positive controls register above 10σ. At N = 10 the record collapses. The preregistered falsifier then ran at N = 14 (2²⁷ amplitudes, GPU): the full ladder g ∈ 0. 2, 0. 4, 0. 6 × Ω ∈ 0. 5, 1, 2, 4, thresholds recalibrated from four decoupled controls, with a noise floor three orders of magnitude below the N = 10 instrument. The record is loudly alive — Rₘax reaches 223 θR — and yet all twelve cells return collapse (b) history-functional: Dₕist decreases with g on every ladder, by up to a factor of 37, and never approaches its threshold. The declared reopening condition (Dₕist growing with g or with N) does not occur; the deciding trend is size-stable. Verdict: collapse (b). Matter–geometry entanglement, at this order and in this window, is a functional of an effective, dynamical background: coupling the modular structure to a quantum incidence register makes its record more background-reducible, not less — the exact echo of Phase VII on the gravity axis. The grade- (ii) ceiling survives both exits; the enclosure holds, symmetrically, and what remains open is named, not hedged. Preregistration SHA-256 (§1–6) = 1843258fb86d8dca279bd0a6f294f2d5fbdde722f27d5a768c774f11e8dfd9dd. Code, preregistration, append-only deviation log, per-claim ledger and emitted data: github. com/PauDD/Simulador-Fase-VIII.