Testing the Representational Sufficiency of Capacity-Based State of Health in Lithium-Ion Batteries

This preprint examines whether capacity-based state of health (SOH) preserves the information required for independently selected present and future battery target records. Using the public Luh–Blank, NASA Ames, and Stroebl lithium-ion battery aging datasets, the study combines matched-state analyses with held-out prediction. Closely matched capacity preserves a selected terminal-voltage record comparatively tightly but leaves greater ambiguity in reference-impedance and pulse-resistance records. In a longitudinal Luh–Blank analysis, capacity-matched present states within the same aging condition retain measurable differences in the next standardized 10-s pulse-resistance record. Adding contemporaneous electrical information to a strong capacity-history and aging-context baseline reduces held-out future-resistance prediction error by 56.25%, and by 45.38% relative to a naïve resistance-persistence forecast. An independent continuous-capacity analysis in the Stroebl dataset shows that recorded aging history also improves prediction of a selected 10-s DCIR target. By contrast, the same four-timescale pulse-resistance block does not establish improved prediction of the next standardized capacity record in a separate Luh–Blank experiment. The results support a target-relative interpretation of battery-state representation adequacy: information omitted by a capacity-based representation can remain consequential for one target while adding no established value for another. This record contains: the preprint manuscript; and the accompanying Supplementary Material. This version has not been peer reviewed. A separate reproducibility archive containing analysis code, processed result tables, verification outputs, figure-source material, and provenance records will be deposited and linked to this record.