We do NOT claim detection of a causal membrane in Spritz LDC2b. We do NOT validate a membrane. This record documents what this pipeline actually does: it survives realistic instrumental gaps and glitches where official LDC scripts fail, and reveals that current methodology may overlook valuable low-frequency secular information. What this development actually does: This is a robust signal processing pipeline for LISA Spritz LDC2b open dataset (Zenodo Open DOI 10.5281/zenodo.7436568, Published 14 Dec 2022, 1.2 GB). It reads real HDF5 files LDC2_spritz_mbhb1_training_v2.h5 (85.8 MB, 10080 NaN gaps), mbhb2 (85.8 MB, 10080 NaN), vgb (1.0 GB, 36 Verification Binaries, 6311520 samples, 146160 NaN gaps, Poisson glitches 4/day), orbits.h5. Converts X,Y,Z to A,E,T: A=(Z-X)/sqrt2, E=(X-2Y+Z)/sqrt6, T=(X+Y+Z)/sqrt3. Spritz stores Michelson, not orthogonal. Cleans real gaps: Official LDC scripts return mean nan when encountering NaN. Console from real files shows NaN counts are physical, not hardcoded. Our method nan_to_num(nan=0.0) before bandpass recovers 146k samples = 8.4 days of real instrumental data that current methodology discards. Avoids N_A*N_E cross-noise: Current pipelines compute time-domain product A*E = (S_A+N_A)(S_E+N_E) = dominant N_A N_E >60 mHz at SNR<=0.2, masking 0.092 Hz/day target with false 0.0650 peak. We use Hilbert pure-phase DeltaPhi=Phi_A-Phi_E, f_raw=gradient(DeltaPhi)/2pi/(2pi), achieving flat std ∼15 mHz (mbhb1 mean -10.7002 std 15.2706, mbhb2 -10.6569 std 15.2163, vgb -18.0460 std 15.4791) vs >60 mHz product method. Corrects 96.4 deg sky error: F_eff = sqrt(F_plus^2+F_cross^2)*(1+0.3cos(2orbit-sky_lon)) with orbital modulation, correcting true CMB dipole 171.5,-11.0 vs best LISA 315.0,-64.3. The blind spot discovered: The universe is in constant accelerated expansion now and had constant inflation early. Everything that expands generates a stretching frequency df/dt: a binary orbit generates f, space itself generates df/dt, CMB traveling through stretching space generates drift. Current LDC methodology high-pass filters all slow drifts below 0.005 Hz as laser/thermal noise, discards gaps as NaN, and searches only fast chirps with IMRPhenom LambdaCDM templates (Omega_L 0.685). If a natural secular resonance of 0.046 mHz/day existed - from universe expansion, inflation, or CMB dipole modulation f_obs=f_CMB*(1+V/c cos theta) varying with LISA orbit 1 year, modulation 1e-4 = 0.046 mHz/day in LISA band - current methodology would lose it as noise. VGB 365d control negative proves it: 36 continuous verification binaries, no MBHB chirp, ideal for secular test, measured slope -0.000284 mHz/day vs hypothetical 0.046 = zero. If real drift existed in real LISA Pathfinder thermal data ESAC (http://lpf.esac.esa.int/lpfsa/) or LISA 2035, current pipeline would also give zero and overlook it. What this is NOT: This is not a detection paper. Spritz ground truth is LambdaCDM, it does not contain any membrane. Residue offset -10 to -18 mHz is filter offset, not signal. Ratio kappa/k=5.32>4.5 overdrive, energy 3531.7x are UAT theoretical parameters, not measured in Spritz. What this IS: A demonstration that valuable low-frequency secular information may be overlooked, and a robust pipeline that survives real gaps/glitches where official scripts fail. For reproducibility: validation_script.py, field_equations.pdf, supplementary.zip with 46 PNG+CSV. References: UAT (10.5281/zenodo.17729221), UPC (10.5281/zenodo.18210808), Dataset Spritz (10.5281/zenodo.7436568). Resonant Hunter DOI intentionally not cited to keep this as independent instrumental methodology record. Keywords: LISA, Spritz, LDC2b, methodological blind spot, Hilbert phase, gaps, glitches, N_A N_E, secular drift, VGB Contact: Puán, Buenos Aires - miguel_percudani@yahoo.com.ar