We propose the Lateral Tension Profile (LTP), an extension to the Alignment Stress Map (ASM) framework [1] that extracts directional information from the alignment field surrounding the generation path. Where the ASM provides a scalar amplitude at each token and layer—how much correction the alignment delta applies—the LTP characterizes the structure of the alignment field in the plane perpendicular to the generation trajectory. The framework defines a counterfactual neighborhood at each generation step: the set of alternative tokens the model considered but did not select, represented by their unembedding difference vectors relative to the chosen token. By projecting the weight delta’s correction onto these counterfactual directions and measuring the resulting lateral tension, the LTP constructs a secondary trajectory whose divergence from the primary semantic trajectory yields an objective geometric signal. The LTP is an instrument, not a detector. It produces structured geometric measurements—an ordered scalar profile and a vector tension point at each token, rolling up to four summary statistics per monitored layer—that characterize the local alignment landscape along any generation path. These measurements carry no threshold, no label, and no assumption about what constitutes adversarial or anomalous behavior. Classification is a separate downstream problem: the same geometric dataset serves adversarial detection, alignment auditing, training diagnostics, and red-team prioritization through different analysis pipelines consuming a common signal. The framework requires no labeled adversarial data, no pre-identified behavioral directions, and adds negligible computational cost beyond the ASM architecture. [1] Ostrander, E. The Alignment Stress Map: Runtime Per-Token Sensitivity Attribution via Weight Delta Projection in Transformer Language Models. 2026.