DIRECTIONAL CMB PARITY ASYMMETRY AND THE MTN-G WINDOWED POWER SPECTRUM: A Pre-Registered Spherical-Cap Patch-Scan Analysis of Planck 2018 PR3 Data

Abstract The Non-Hermitian Algebraic Gravity framework (MTN-G) derives spacetime geometry as a thermodynamic phase transition from a Type III₁ von Neumann algebraic substrate. At the cosmic Exceptional Point (EPᵈ), eigenvector coalescence archives pre-geometric modular coherence into the Sparse-Reservoir Substrate Memory network, leaving a spectral scar: a k²-suppressed primordial power spectrum at wavenumbers k < k_crit (the D5 windowed template). The Temporal Layering (TL) conjecture further predicts that spatial heterogeneity in the network channel density N_channels(x) — modelled as a Poisson random field with mean λ_TL — produces directional imprints on the CMB: localised TT suppression correlated with enhanced E/B parity asymmetry in the same sky region. Here we report a pre-registered spherical-cap patch-scan analysis of Planck 2018 PR3 published band-powers against this framework. The D5 windowed spectrum provides a substantially better fit to Planck 2018 TT band-powers at ℓ = 2–20 than the ΛCDM best-fit; the v2 internal-consistency audit (§III.3a) bounds the BIC-corrected Bayes factor at ln B ≈ 1.5–3 (best-fit k_crit = 2.89 × 10⁻³ Mpc⁻¹, ℓ_crit ≈ 40). The preference is weak-to-moderate — at or slightly above the Jeffreys threshold — and provisional pending re-derivation with the full Plik likelihood; the structural preference over the standard dipole modulation model is unchanged. The TL directional amplitude ε is derived from the Poisson variance: ε = 1/√λ_TL ≈ 0.77 (λ_TL ≈ 1.69 channels per coherence volume), consistent with the observed quadrupole suppression ratio. The pre-registered hypothesis tests H1 (TT suppression > 2.5σ) and H2 (E/B parity asymmetry > 2σ) both pass from published Planck 2018 VII results (3.5σ and 2.1σ respectively; foreground contamination bounded to < 0.3σ by PR3 null tests). H3 (Bayes factor for the directional model) is θ_cap-dependent and cannot be resolved without MultiNest sampling on the full Planck PR3 Q/U polarisation maps. Two additional falsifiable predictions are advanced: H4 (T-E cross-correlation suppression R_TE < 0.80 in the TL island) and H5 (positive E-mode angular skewness S_E > 0 at ℓ = 2–5), both testable on existing Planck data once the island direction n̂₀ is determined. The MTN-G core framework (D1–D6) is not falsified. The TL conjecture is INCONCLUSIVE pending the definitive MultiNest analysis specified in Section VIII and Appendix A. Key Words: MTN-G / Non-Hermitian Algebraic Gravity · Temporal Layering · Exceptional Point coalescence · Type III₁ von Neumann algebra · emergent spacetime · primordial power spectrum · k²-suppression · CMB low-ℓ anomalies · quadrupole suppression · E/B parity asymmetry · directional power spectrum · spherical-cap patch scan · Planck 2018 PR3 · Bayesian model comparison · MultiNest · look-elsewhere correction · Poisson channel density · SRSM network · T-E cross-correlation · E-mode angular skewness · pre-registration · falsifiability