[2603.01326] Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning

Computer Science > Computation and Language arXiv:2603.01326 (cs) [Submitted on 1 Mar 2026] Title:Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning Authors:Hamed Damirchi, Ignacio Meza De la Jara, Ehsan Abbasnejad, Afshar Shamsi, Zhen Zhang, Javen Shi View a PDF of the paper titled Truth as a Trajectory: What Internal Representations Reveal About Large Language Model Reasoning, by Hamed Damirchi and Ignacio Meza De la Jara and Ehsan Abbasnejad and Afshar Shamsi and Zhen Zhang and Javen Shi View PDF HTML (experimental) Abstract:Existing explainability methods for Large Language Models (LLMs) typically treat hidden states as static points in activation space, assuming that correct and incorrect inferences can be separated using representations from an individual layer. However, these activations are saturated with polysemantic features, leading to linear probes learning surface-level lexical patterns rather than underlying reasoning structures. We introduce Truth as a Trajectory (TaT), which models the transformer inference as an unfolded trajectory of iterative refinements, shifting analysis from static activations to layer-wise geometric displacement. By analyzing displacement of representations across layers, TaT uncovers geometric invariants that distinguish valid reasoning from spurious behavior. We evaluate TaT across dense and Mixture-of-Experts (MoE) architectures on benchmarks spanning commonsense reasoning, question an...