Latent Manifold Fragmentation in Wearable sEMG: An Empirical Audit of Cross-Subject Generalization

Surface electromyography (sEMG) gesture recognition literature frequently reports high offline accuracy ($>80\%$), yet these claims often mask methodological pitfalls such as temporal window leakage and rest-class metric inflation. In this paper, we introduce sEMG-Bench, a standardized, leakage-free benchmark framework, and conduct an empirical audit of four representative deep architectures alongside feature-engineered baselines on NinaPro DB5. Our empirical findings reveal three key insights: (1) Under strict Leave-One-Subject-Out cross-validation (LOSOCV), zero-shot active gesture recognition collapses across all deep backbones ($\sim$20–25% Active F1), showing a pronounced disparity with reported offline claims. (2) Signal rectification ($|x(t)|$) provides a vital physical energy prior, recovering $+6\%$ to $+20\%$ Active Macro F1 over raw signed inputs. (3) Paired t-tests ($p>0.05$) demonstrate statistical parity between lightweight 2D CNNs and complex Transformers, indicating that architectural over-engineering yields no significant generalization advantage under inter-subject distribution shifts. Linear probing reveals the underlying mechanism: frozen latent representations encode biometric subject identity ($>81\%$ accuracy) rather than gesture-invariant topologies, proving that cross-subject adaptation is geometrically essential for consumer-grade wearable interfaces.