Integrity review of "Environmental Assessment of Closed-Loop Regeneration of Spent LFP Batteries Based on Factory-Level Inventory Data" (Energies, DOI: 10.3390/en19163749)

Academic fraud report · Geng Detector

Summary

Verdict: Suspect (yellow). The paper presents a factory-level life-cycle assessment of closed-loop regeneration of spent LFP batteries with internally consistent numerical work and an appropriate timeline. However, multiple last-digit statistical indicators (chi-squared = 118.3, p ≈ 0; even/odd ratio = 0.39, p = 0.0028; adjacent-digit deviation 5.7σ; Benford MAD rising from 0.0378 at the 2nd digit to 0.1546 at the 4th digit) deviate jointly, raising questions about the naturalness of the inventory data. Important caveats apply: Benford-style tests assume unconstrained natural data, but Tables 2–3 are normalized to 1 kg of regenerated LFP CAM and rounded, which weakens the test's premise. Image forensics (copy-move on five figures, PRNU co-sourcing between Figure 1 and Figure 2, heterogeneous noise variance on Figure 3) are all attributable to known false-positive mechanisms on software-rendered statistical charts. Math, methodology, references, authorship, and timeline checks pass. The reviewer downgrades the case from "highly suspect" to "suspect" and recommends requesting raw factory records from the authors.

Verdict

🟡 Suspect. Numeric internal consistency and methodology are intact, but the joint deviation of four independent last-digit statistics constitutes the only unresolved substantive concern. Image-related signals are explained by known false-positive mechanisms on software-rendered statistical charts.

Key findings

Last-digit distribution in Tables 2–5 deviates sharply (χ² = 118.3, p = 0.0000); even/odd ratio = 70/181 = 0.39 (p = 0.0028); adjacent-digit correlation 5.7σ above expectation.
Benford joint test shows strong deviation at the 2nd, 3rd, and 4th digits (max χ² = 1203.7, p = 0.0000); MAD rises 0.0378 → 0.1546 from the 2nd to the 4th digit.
Caveat: Tables 2–3 data are factory-level inputs normalized to 1 kg of regenerated LFP CAM and rounded to 2–3 decimals; ratios and process stoichiometry induce co-variation. Benford's "natural, scale-free, unrounded" premise is only partially satisfied, so statistical significance does not equal proof of fabrication.
Copy-move detection flagged 51–207 matching pairs across five figures (img-000 to img-004), with peaks at img-002 (offset (0, 32), 207 pairs) and img-001 (offset (32, 0), 130 pairs). All offsets align with a 32 px module width characteristic of chart panel layouts, consistent with rule-based repeated chart elements rather than content reuse.
PRNU co-sourcing flagged between Figure 1 (process flow) and Figure 2 (bar chart) with NCC = 0.230 and PCE = 30.2; both are software-rendered charts processed by the same MDPI production pipeline, so the sensor-origin premise of PRNU analysis is not met.
Noise variance heterogeneity on Figure 3 (CV = 1.21 vs 0.57–0.94 elsewhere) is consistent with the multi-panel composite (absolute bar charts + percentage contribution + activity-level contributors), in which content density differs across sub-panels.
AI-generated text score = 0.35/1.0, within the detector's ambiguous band (0.3–0.5) with no suspicious passage identified.
Mathematical self-consistency checks pass: GWP100 reduction (6.59 → 6.08 = 7.73% claimed, 7.74% computed), FFP, SOP, HTPc percentages, and S2 grid-decarbonization contribution (1.071 × 0.953 = 1.021 kg CO2-eq) all match within ≤0.05 percentage-point rounding tolerance.
Authorship and funding (Guangzhou Customs Technology Center; Chinese Academy of Sciences; Grant 2025HK243 from the General Administration of Customs) align with customs oversight of solid-waste and recycled-material flows.
Timeline (Received 17 June 2026; Accepted 20 July 2026; Published 10 August 2026; 33-day review) is plausible for MDPI Energies. Several 2026 references (e.g., [7], [11], [19], [22]) are used; the in-press status of reference [22] could not be independently verified.

Evidence highlights

Last-digit χ² = 118.3, p = 0.0000; even/odd ratio = 0.39, p = 0.0028; adjacent-digit deviation = 5.7σ.
Benford MAD progression: 0.0378 (2nd digit) → 0.1546 (4th digit).
Copy-move: img-002 (0, 32) = 207 pairs; img-001 (32, 0) = 130 pairs; module-aligned 32 px offsets.
PRNU: NCC = 0.230, PCE = 30.2 (Figure 1 vs Figure 2).
Noise CV: img-002 = 1.21 vs 0.57–0.94 for other figures.
Internal-consistency checks all within rounding tolerance: GWP100 7.73% (claimed) vs 7.74% (computed); grid-decarbonization S2 electricity contribution 0.967 kg CO2-eq vs paper-claimed reduction consistent to 0.003 kg.
DOI: 10.3390/en19163749.

Notes

The reviewers' machine-tool caveats (Benford premise violated for normalized/rounded LCA inventory data; copy-move/PRNU specificity low on rendered statistical charts) are explicitly acknowledged in the source report and used to downgrade findings rather than dismiss them.
The reviewer recommends requesting the original factory-level ledgers from the authors (currently withheld on commercial-confidentiality grounds) as the decisive next step.
Image-forensics conclusions are not independent confirmations of misconduct because the test premises (sensor-origin differentiation for PRNU; content reuse for copy-move) are not satisfied by homogeneous MDPI-style statistical charts.
All verdict changes in the source (from 🟠 to 🟡) are reproduced here for transparency.