Geng Integrity Report: Towards Efficient Random-Order Enumeration for Join Queries

Academic fraud report · Geng Detector

Summary

Verdict: No indicators of academic fraud detected. This VLDB 2026 paper (DOI: 10.14778/3796195.3796203) by Harbin Institute of Technology authors presents an algorithm for random-order join enumeration. Key findings: all figures are algorithm diagrams or runtime/memory benchmark plots from TPC-DS+ and Twitter datasets with no signs of image reuse or splicing; performance curves include honest OOR (Out of Resources) annotations in Figures 4–5, atypical of data fabrication; theorems use worst-case complexity analysis under the k-clique assumption rather than p-values; publication timeline (2025 technical report, 2026 PVLDB) is consistent; the authors provide open-source code at https://github.com/Chen-Py/JoinREnum and document hardware (Intel Xeon E7-8860, 384GB RAM) and baselines (PGMJoin, NWCO). Limitations: detection tools designed for biomedical image fraud are not well-suited to theory-heavy CS papers, so confidence is moderate rather than absolute. No follow-up action recommended.

Verdict

Cleared (✅). No evidence of academic misconduct was identified. The investigation found no indicators of image duplication, data fabrication, image splicing, statistical anomalies, or irregular citation patterns.

Key findings

All figures (Figure 1–7) are algorithm structure diagrams (e.g., RRATree, Division) or system performance line/bar charts for TPC-DS+ and Twitter datasets; no biological imagery is present.
Performance curves in Figures 4–5 include explicit OOR (Out of Range/Out of Resources) annotations, indicating transparent reporting of unfavorable results.
The theoretical contributions (Theorems 1–3, Lemmas 1–11) rely on worst-case complexity bounds under the combinatorial k-clique assumption, with no p-values or statistical testing where none would be expected.
Methodology section (Section 6.1) documents hardware (Intel Xeon E7-8860, 384GB RAM) and clearly distinguishes baselines: PGMJoin (public code) vs. NWCO (re-implemented from paper).
Authors provide an open-source repository: https://github.com/Chen-Py/JoinREnum.
Publication timeline (technical report cited as [11] in 2025; PVLDB Vol. 19, No. 5, pp 889–901, May 2026) is internally consistent with normal CS publishing practice.

Evidence highlights

DOI: 10.14778/3796195.3796203
Figures 4–5: Contain OOR markings where the baseline algorithm underperformed, a behavior inconsistent with fabricated performance data.
Reference [11]: 2025 Technical Report, predating the 2026 VLDB publication by roughly one year.
Section 6.1: Explicit hardware specification (Intel Xeon E7-8860, 384GB RAM).
Open-source link: https://github.com/Chen-Py/JoinREnum

Notes

The detection toolkit is calibrated primarily for biomedical image-fraud and Western blot/immunofluorescence manipulation. Applied to a theory-heavy database systems paper, its discriminative power is limited; this report therefore expresses moderate confidence rather than absolute certainty.
No irregularities were detected, but absence of evidence is not proof of integrity in every dimension beyond the tools' scope.
No author contact, PubPeer post, or institutional inquiry is recommended on the basis of this report.