Provenance-Constrained Reconstruction of Multi-Generation Lithium-Ion Battery Aging Experiments: From Intervention Outcomes to Collapse-Precursor Hypotheses

This study presents a provenance-constrained reconstruction of a multi-generation series of computational lithium-ion battery-aging experiments performed using publicly available NASA battery-aging data. The reconstructed research history contains distinct experimental generations, including battery-specific computational intervention experiments, End-of-Life (EOL) analyses, 26-cell aggregate experiments, and later multidimensional Risk-Dose, Principal Component Analysis (PCA), and finite-state risk analyses. Recovered case-level computational EOL results include 13→14 (+7.69%), 30→38 (+26.67%), 91→132 (+45.05%), and a zero-gain case. These values are preserved as historical case-level results and are not interpreted as representative population-level effects or physical battery-life extension. A key provenance finding is that multiple recovered 26-cell runs share the same baseline mean EOL of 55.58 while reporting different controlled mean EOL values of 50.23, 58.42, 61.15, and 63.58. Because the exact version lineage remains unresolved, these results are not pooled and no single aggregate intervention effect is selected. Later historical analyses contain Risk-Dose visualization, PCA-based multidimensional representations, finite-state risk mapping, and an explicit COLLAPSE_PRECURSOR state. However, the currently recovered evidence does not establish that these later methods generated the earlier intervention results. Exact historical Risk-Dose equations and some finite-state thresholds also remain unresolved and are not retrospectively reconstructed. The principal contribution of this work is methodological and empirical: it demonstrates how restoring computational provenance can reveal incompatible historical results, prevent favorable-result selection, and distinguish recovered observations from unresolved methodology and prospective hypotheses. The study does not claim experimentally validated physical battery-life extension, a universal intervention effect, or superiority over conventional State-of-Health (SOH) or Remaining-Useful-Life (RUL) methods. The accompanying reproducibility package contains the manuscript, standalone reconstruction code, an evidence ledger, provenance status information, and SHA-256 integrity records. The reproduction code intentionally reproduces only claims and calculations supported by recovered evidence and does not fabricate unrecovered historical algorithms. Version: v0.3 Author: Kim, M.-G. Independent Researcher Date: 2026-09-21