Limitations of Capacity-Based State-of-Health Metrics in Lithium-Ion Batteries: Evidence from Matched-State Analysis

Lithium-ion battery health is commonly assessed using scalar metrics such as capacity fade, implicitly assuming that operational validity can be determined from a single degradation coordinate. We evaluate this assumption using empirical degradation data from two independent publicly available datasets: the NASA Ames battery aging dataset and the Luh and Blank 2024 NMC battery aging dataset. System state is represented using both normalized capacity and a resistance metric, and operational validity is defined through simultaneous threshold conditions on these variables. In the NASA dataset, 20 of 115 matched-capacity pairs exhibited different validity status despite nearly identical capacity values, with the higher-resistance state corresponding to the invalid condition in all cases (empirical p = 0.028). In the Luh and Blank dataset, analyzing all 76 processed cells, 136,075 of 1,487,592 matched-capacity pairs exhibited different validity status with directional agreement of 0.894 (empirical p < 0.001, Z = 5.97). A third dataset using a cruder resistance proxy was directionally consistent but statistically inconclusive. These results demonstrate that capacity alone is insufficient to determine operational validity and that failure behavior depends on multi-parameter state conditions across independent datasets and experimental protocols. Code and processed data supporting this work are available at:https://doi.org/10.5281/zenodo.20044598