Re-evaluation Report (Corrected): Green Synthesis of Nitrogen-Doped Carbon Dots from Fresh Tea Leaves for Selective Fe³⁺ Ions Detection and Cellular Imaging (DOI: 10.3390/nano12060986)

Academic fraud report · Geng Detector

Summary

This corrected report re-examines a 2022 paper in Nanomaterials (MDPI) describing tea-leaf-derived N-doped carbon dots for Fe³⁺ sensing and cellular imaging. The overall verdict remains "highly suspicious" (orange), but most forensic flags were downgraded after benign explanations were validated against the original text. One finding survives as confirmed at the textual level: the selectivity bar chart in Figure 5D shows an implausibly narrow F/F₀ distribution (≈0.85–1.05) for all non-Fe³⁺ ions while Fe³⁺ drops to ≈0.2, with near-invisible error bars. Benford/leading-digit tests, copy-move block matches, PRNU similarities, noise-variance CV, AI-style phrasing, LOD proximity to literature, CCK-8 monotonicity, and TEM scale plausibility were each rebutted by documented benign causes: instrument-precision rounding, MDPI's unified typesetting pipeline, multi-panel layout, normalized F/F₀ readings, standardized CCK-8 kits, and low-cytotoxicity dose-response. Bayesian synthesis (prior 5%, n≈11 calibration) yields a posterior of 96.3% (95% CI 77.3%–99.4%, BF ≈487.9), but this likely overstates joint strength because assumptions of conditional independence are violated.

Verdict

🟠 Highly suspicious. One text-level finding (Figure 5D selectivity pattern) is confirmed; seven other forensic signals were downgraded because benign explanations tied to instrument precision, MDPI typesetting, and the paper's own methodological statements are well-supported in the original text.

Key findings

Figure 5D selectivity pattern (CONFIRMED): All non-Fe³⁺ ions yield F/F₀ clustered in 0.85–1.05 with very small error bars; Fe³⁺ alone drops to ≈0.2. Although normalized F/F₀ and weak probe–ion interaction are physically expected, the near-zero dispersion is atypical.
Benford + last-digit joint test (DOWNGRADED): MAD 0.0407–0.1511 across digits 1–4; χ²=37.2–1012.5; last-digit χ²=612.5; odd/even ratio 25/159; neighbor-digit deviation 18.0σ. Rebutted: data are constrained-precision measurements (TEM ≈2.32 nm, zeta −13.80 mV, XPS 285.08/399.08/532.08 eV, LOD 0.079 µM) where Benford assumptions are partially violated.
Copy-move block matches at offsets (32,0), (0,32), (40,0) across many panels (DOWNGRADED): Rebutted by multi-panel layouts (Figures 1A/B, 2A–D, 3A–D, 5A–D, 7A–F) and MDPI's uniform rendering pipeline.
PRNU near-identity (NCC=1.000) across multiple image pairs (DOWNGRADED): Rebutted by MDPI's single rendering pipeline embedding all figures into one PDF, and by the report's own visual assessment finding no evident reuse/compositing in Figure 7.
Noise-variance CV=1.310 in img-019/img-021 (DOWNGRADED): Rebutted by Figure 7's bright-field + 405 nm + 488 nm multi-channel design, where channel brightness differs intrinsically.
AI-style phrasing, AI score 0.35 (DOWNGRADED): Spelling errors ("specctrum," "specifc," "teh") and score in the marginal band argue against heavy AI polishing; detector accuracy on mixed technical text is limited.
LOD 0.079 µM vs cited 0.075 µM (DOWNGRADED): ≈5% difference; Table 1 spans 0.021–1.68 µM and uses the same 3σ/K formula.
CCK-8 monotonic 100%→87% across 0–400 µg/mL (DOWNGRADED): Standardized kit + microplate reader (450 nm) and the authors' claim of "very low cytotoxicity" support a shallow, monotonic curve.
Figure 1A scale bar 20 nm vs 2.32 nm particles (DOWNGRADED): Routine in TEM practice; authors state "monodispersed … without apparent aggregation."

Evidence highlights

Figure 5D bar pattern independently verifiable from the published figure.
Numerical anchors preserved: TEM 2.32 nm (range 2.0–2.5 nm); zeta −13.80 mV; XPS 285.08/399.08/532.08 eV; LOD 0.079 µM; literature comparator 0.075 µM.
Automated Bayesian synthesis (geng_lr_v3, prior 5%, n=11 calibration) reports posterior 96.3%, 95% CI 77.3%–99.4%, BF ≈487.9; 52 cleared cues excluded.
Top contributors are PRNU comparisons (img-000 vs img-001/002/003) and a column-difference statistic; these are precisely the cues weakened by the MDPI pipeline benign explanation.

Notes

Posterior is likely over-confident because the model treats correlated image-pipeline cues as conditionally independent; treat the 96.3% as an upper bound, not a calibrated probability.
The only text-level concern not rebutted is the Figure 5D pattern; recommend requesting the underlying .csv of F/F₀ measurements with replicate values.
Suggested actions: request raw TEM, selectivity, and CCK-8 data; consider a neutral PubPeer comment; allow ~30 days for author response before any escalation to MDPI Nanomaterials editorial office or Central South University.
Report is AI-assisted; final判定 requires institutional investigation. Confidence in the downgrades is moderate-to-high; confidence in the upgraded Bayesian posterior is low.