Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability

Xinyan Jiang, Ninghao Liu, Di Wang, Lijie Hu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:52776-52802, 2026.

Abstract

Evaluating LLM reliability via scalar probabilities often fails to capture the structural dynamics of reasoning. We introduce TRACED, a framework that assesses reasoning quality through theoretically grounded geometric kinematics. By decomposing reasoning traces into Progress (displacement) and Stability (curvature), we reveal a distinct topological divergence: correct reasoning manifests as high-progress, stable trajectories, whereas hallucinations are characterized by low-progress, unstable patterns (stalled displacement with high curvature fluctuations). Leveraging these signatures, our probabilistic framework achieves competitive performance and superior robustness across diverse benchmarks. Crucially, TRACED bridges geometry and cognition by mapping high curvature to "Hesitation Loops” and displacement to ”Certainty Accumulation”, offering a physical lens to decode the internal dynamics of machine thought.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-jiang26af, title = {Beyond Scalars: Evaluating and Understanding {LLM} Reasoning via Geometric Progress and Stability}, author = {Jiang, Xinyan and Liu, Ninghao and Wang, Di and Hu, Lijie}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {52776--52802}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/jiang26af/jiang26af.pdf}, url = {https://proceedings.mlr.press/v306/jiang26af.html}, abstract = {Evaluating LLM reliability via scalar probabilities often fails to capture the structural dynamics of reasoning. We introduce TRACED, a framework that assesses reasoning quality through theoretically grounded geometric kinematics. By decomposing reasoning traces into Progress (displacement) and Stability (curvature), we reveal a distinct topological divergence: correct reasoning manifests as high-progress, stable trajectories, whereas hallucinations are characterized by low-progress, unstable patterns (stalled displacement with high curvature fluctuations). Leveraging these signatures, our probabilistic framework achieves competitive performance and superior robustness across diverse benchmarks. Crucially, TRACED bridges geometry and cognition by mapping high curvature to "Hesitation Loops” and displacement to ”Certainty Accumulation”, offering a physical lens to decode the internal dynamics of machine thought.} }
Endnote
%0 Conference Paper %T Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability %A Xinyan Jiang %A Ninghao Liu %A Di Wang %A Lijie Hu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-jiang26af %I PMLR %P 52776--52802 %U https://proceedings.mlr.press/v306/jiang26af.html %V 306 %X Evaluating LLM reliability via scalar probabilities often fails to capture the structural dynamics of reasoning. We introduce TRACED, a framework that assesses reasoning quality through theoretically grounded geometric kinematics. By decomposing reasoning traces into Progress (displacement) and Stability (curvature), we reveal a distinct topological divergence: correct reasoning manifests as high-progress, stable trajectories, whereas hallucinations are characterized by low-progress, unstable patterns (stalled displacement with high curvature fluctuations). Leveraging these signatures, our probabilistic framework achieves competitive performance and superior robustness across diverse benchmarks. Crucially, TRACED bridges geometry and cognition by mapping high curvature to "Hesitation Loops” and displacement to ”Certainty Accumulation”, offering a physical lens to decode the internal dynamics of machine thought.
APA
Jiang, X., Liu, N., Wang, D. & Hu, L.. (2026). Beyond Scalars: Evaluating and Understanding LLM Reasoning via Geometric Progress and Stability. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:52776-52802 Available from https://proceedings.mlr.press/v306/jiang26af.html.

Related Material