Abstract
Large language models frequently commit unrecoverable reasoning errors mid-generation: once a wrong step is taken, subsequent tokens compound the mistake rather than correct it. We introduce Latent Phase-Shift Rollback (LPSR): at each generation step, we monitor the residual stream at a critical layer lcrit, detect abrupt directional reversals (phase shifts) via a cosine-similarity + entropy dual gate, and respond by rolling back the KV-cache and injecting a pre-computed steering vector. No fine-tuning, gradient computation, or additional forward passes are required. LPSR achieves 44.0% on MATH-500 with an 8B model versus 28.8% for standard AR (+15.2 pp; McNemar χ² = 66.96, p < 10⁻¹⁵). Critically, prompted self-correction, the most natural inference-time baseline, scores only 19.8%, below standard AR; LPSR exceeds it by +24.2 pp (χ² = 89.4, p ≈ 0). LPSR also outperforms Best-of-16 (+7.8 pp) at 5.4× lower token cost, and surpasses a standard 70B model (35.2%) with 8.75× fewer parameters at ∼3× the token budget. A 32-layer sweep reveals a novel \textbf{detection-correction dissociation}: error-detection AUC peaks at layer~14 (0.718) but task accuracy peaks at layer~16 (44.0% vs.\ 29.2%), demonstrating that optimal monitoring depth differs for detection and correction.