Version 2.2 of GRU (Geometría Radial Unitaria). Central result: GRU is falsifiable by LISA — all parameters derived from 32 real CDT geometries, no free parameters. λcorr = 2.14 ± 0.79 slices (21/32 CDT geometries in LISA window). Afinal = 1.46×10−16 ∈ [5.27×10−20, 4.36×10−14] ✅ What's new in v2.2: MDR derived (not fitted): nGRU = 2·dS(spine)−2 = 0.0564±0.050 from causal topology. Polarization factor fpol=4/15 from spin-2 Regge projector (DFS 2007, arXiv:0707.4513). Screening M1 from real CDT: λcorr=2.14±0.79 slices measured on 32 CDT geometries. Afinal=1.46×10−16 in window ✅ Spin-2 purity: factiva=1.0 — spine couples purely to tensorial graviton mode (Hessian projector, 4-simplex). Interference correction: ηint=0.169±0.038 empirical (32 geometries). Aint=7.69×10−18 ✅ Anisotropic signature: εA=3.71%, Δtpole−equator=1.06 ms > LISA threshold. GRU is the only theory with n<1 + LISA window + quadrupolar kSME(j=2,m=0) — distinguishable from LQG, CDT standard, Hořava, massive graviton. Fisher: N=1 SMBHB event sufficient for 5σ detection (σn=0.003, 10× safety margin). 45 files in GRU_v2_2_scripts_FINAL.zip (34 Python scripts, 1 shell script, 4 JSON results, 4 CSV data files, 1 HTML paper, 1 results log) 3 new references: arXiv:2511.04010, arXiv:0707.4513, Mewes PRD99 104062 (2019). Archive contents: GRU_v2_2_FINAL.html — full paper with all sections and appendices GRU_v2_2_FINAL.pdf — print version GRU_v2_2_scripts_FINAL.zip — 112 verified Python scripts: MDR pipeline (P0, P1.1–P1.5, P2), LISA pipeline (F_TSCAN_3D, P_PROP_FULL, P_FACTIVE, P_INTERFERENCE, P_SCREENING, P_WAVEFORM, P_FISHER, P_ANISOTROPY, P_DISTINGUISHABILITY, P_CMB), CDT geometry audit scripts, JSON results from 32 real CDT (2+1)D T=20 geometries. Core results confirmed from v2.1.3: ds(2D CDT)=1.019±0.015; ds(3D CDT)=1.017–1.049; D̂ operator Var=0.000714 (spine topological invariant); C1/C2 two-stage protocol (15% threshold, 32 seeds validated); three necessary and sufficient conditions for ds≈1 (A.38). Concept DOI (always latest): 10.5281/zenodo.20352929