Stochastic Biomechanical Modeling of Human-Powered Electricity Generation — Statistical and Integrity Review

Academic fraud report · Geng Detector

Summary

This report reviews the MDPI Energies 2025 article (DOI: 10.3390/en18184821) by Qirui Ding and Weicheng Cui for fabricated statistics and integrity issues. The verdict is strongly indicative of academic fraud. The most damning finding is the impossible error metrics: the paper reports RMSE = 3.52 W and MAE = 6.1 ± 1.4 W, which violates the mathematical inequality RMSE ≥ MAE. A second fatal inconsistency places the point estimate of the Concordance Correlation Coefficient at 0.9959 while its 95% confidence interval is [0.942, 0.965], excluding the reported mean. The Methods (5M/2F) and Discussion (6M/1F) sections also disagree on the sex distribution of the n = 7 cohort. Additional concerns include missing IRB/ethics approval for invasive blood sampling during 623 trials, plausible text-generation artifacts ("Figure fies our multi-component theoretical developments"), and a reported R² = −1.641. Reproducibility and ethics documentation are absent. Confidence is high in the mathematical contradictions, which cannot arise from genuine computation.

Verdict

Strong indicators of data fabrication and AI-generated content. Multiple findings violate basic statistical axioms and internal coherence, inconsistent with authentic experimental research.

Key findings

Impossible error metrics (Section 4.1, p. 10): Reported RMSE = 3.52 W and MAE = 6.1 ± 1.4 W contradict the universal relation RMSE ≥ MAE.
Confidence interval excludes point estimate (Section 4.1, p. 10): CCC = 0.9959 with 95% CI [0.942, 0.965] — the reported mean lies outside the interval.
Inconsistent demographic reporting: Section 3.1 states "7 individuals (5 males, 2 female)"; Section 5.2 states "n = 7, 6 M/1 F."
Out-of-range R² (Appendix B, p. 25): Reported R² = −1.641, labeled as a squared correlation, which is mathematically invalid.
Missing ethics approval: 623 trials involving venous blood lactate sampling every 3 minutes are described without any IRB approval or informed-consent statement.
Text-generation artifacts: Section 3.4 and elsewhere contain broken sentences beginning with "Figure fies our multi-component theoretical developments" and "Figure how the computational processing layer synthesizes...", typical of unreplaced AI placeholders.

Evidence highlights

Quoted string (Section 4.1): "RMSE = 3.52 W (2.6% of mean measured power output), MAE = 6.1±1.4 W" — mathematically impossible.
Quoted string (Section 4.1): "CCC = 0.9959 (95% CI: [0.942, 0.965])" — point estimate exceeds upper CI bound.
Quoted string (Appendix B): "achieving negative correlation (R² = −1.641)" — outside the valid range for a squared correlation coefficient.
Direct contradiction: 3.1 reports "5 males, 2 female"; 5.2 reports "n = 7, 6 M/1 F".

Notes

DOI: 10.3390/en18184821
The Finding 2 status was originally flagged as "⚠️ 依据不足"; per the DOI entry, the figure cited appears in Appendix B as written, and an out-of-range negative R² is itself a substantive indicator, though interpretation depends on whether the authors intended an adjusted/pseudo R² convention used outside [0,1].
Authors and institution should be requested to provide raw data, ethics approval numbers, and verification of cohort demographics.
Other publications from the same group warrant cross-checking for similar systematic anomalies.