Evidence coding and quality assessment for 159 studies on internal short-circuit monitoring in lithium-ion batteries
Description
This dataset contains the evidence coding, quality assessment, sampling records, reliability data, and reproducibility materials underlying a systematic review of internal short-circuit (ISC) monitoring in lithium-ion batteries. It enables Figures 2 to 6, the numerical content underlying Figure 1, and the cross-tabulations and percentages reported in the associated article to be independently reproduced or recomputed, and individual study-level coding decisions to be examined. Records were identified by searching Scopus, ScienceDirect, and IEEE Xplore on 12-13 August 2026 for publication years 2022-2026 (Scopus 744, ScienceDirect 449, IEEE Xplore 95). The combined identification set contained 1,288 records, reduced to 859 unique records after DOI-based deduplication. Of these, 537 reports were assessed at full text and 424 met the eligibility criteria. A protocol-defined stratified selection drew 170 studies for detailed coding. Of these, 159 were coded from full text and form the analytical corpus; 11 selected studies remain eligible but uncoded because usable full text could not be verified for final extraction. Each coded study carries assignments on seven axes: ISC reproduction route, fault-reproduction fidelity, ground-truth credibility, monitoring function, analytical method family, validation domain, and deployability requirements. Study-level assignments are accompanied by full-text evidence traces. Each analytical study also carries 24 quality-assessment items across a common module and conditional experimental, model, and diagnostic modules, with applicability determined before item scoring. Coding was finalized by the first author from the primary full texts while blinded to an earlier AI-assisted preliminary pass, whose role and concordance audit are documented in the provenance records. An independent second reviewer coded a 40-study subsample on the two central evidence axes before reconciliation. Agreement was 85.0% (Cohen's kappa = 0.809) for fault-reproduction fidelity and 75.0% (kappa = 0.645) for ground-truth credibility. The 24 quality-assessment items were not independently double-rated by the second human reviewer. Monitoring function, analytical method family, validation domain, and deployability requirements are multi-label, so their category counts are non-exclusive and must not be summed to the number of studies. Quality-assessment percentages use each item's applicable denominator rather than the full corpus. Full texts of the reviewed articles are not redistributed because they remain under publisher copyright; bibliographic records retain DOI identifiers where available.
Files
Steps to reproduce
Figures 2 to 6 and the principal cross-tabulations are generated from 02_Final_Analytical_Coding_and_QA.xlsx using 06_Generate_Figures_1_to_6.py. The script also validates all numerical values underlying Figure 1. The publication layout of Figure 1 was manually redrawn in Microsoft Visio from these validated counts, so its graphical arrangement is not expected to be pixel-identical to the computational reference. Place 06_Generate_Figures_1_to_6.py and 02_Final_Analytical_Coding_and_QA.xlsx in the same directory. Install the required packages: python -m pip install pandas numpy matplotlib openpyxl Run: python 06_Generate_Figures_1_to_6.py The script creates a figures_reproduced directory. Figures 2 to 6 are exported as 600-dpi PNG and vector PDF files. Figure_1_Numeric_Reference provides the computational reference for Figure 1. The script also writes recalculated_counts.csv. Analyses use rows in Seven_Axis_Coding_170 with Full_text_read = Yes, yielding n = 159. The 11 selected but uncoded studies remain in the sheet with Full_text_read = No and are excluded from analytical and QA calculations. Protocol constants are: 1,288 records identified, 429 duplicates removed, 859 unique records screened, 537 reports assessed at full text, 424 studies eligible, and 170 selected for Layer-2 coding. Monitoring function, analytical method family, validation domain, and deployability requirements are multi-label, so their category counts are non-exclusive. Fault-reproduction fidelity and ground-truth credibility use one conservative study-level code. QA percentages use each item's applicable denominator rather than n = 159. The full coding definitions and denominator rules are provided in 03_Codebook_and_Amendment_History.xlsx and 07_README_and_Reproduction_Instructions.md. Human second-reviewer reliability data are provided in 04_Independent_Reliability_Subsample_40.xlsx, and sampling and provenance decisions are documented in 05_Sampling_and_Provenance_Audit.xlsx.