LSP-Orbital v4.1: Lo-Shu Geometric Orbit Propagation with Polynomial Feature Expansion

LSP-Orbital v4.1 extends v1.0 with polynomial feature expansion (21-dim) and large-scale training (410 satellites, 3,280 samples, 7.8x v1.0). Key results: 49.6% mean error reduction over pure Keplerian propagation (dt 30-120min), win rate 80%. Iridium LEO error: 12.60km to 6.80km (-46%). Theorem LSP-O: Lo-Shu row-imbalance predicts tangential orbital residuals with R^2=0.333 (LEO, N=2000, robust), R^2=0.641 (MEO, N=680). Full training dataset included (3280x21 features, 3280x3 targets, all orbit-type coefficients). Supersedes: 10.5281/zenodo.20082651.