pith. sign in

arxiv: 2606.01789 · v1 · pith:UARD3O5Lnew · submitted 2026-06-01 · 💻 cs.AI

Consistency evaluation of benchmarks used for causal discovery

Yuzhe Zhang , Chihui Chen , Lina Yao , Chen Wang This is my paper

Pith reviewed 2026-06-28 14:23 UTC · model grok-4.3

classification 💻 cs.AI
keywords causal discoverybenchmark evaluationlarge language modelsconsistency checkinggraphical causal modelsdomain knowledge
0
0 comments X

The pith

Eleven popular causal discovery benchmarks show large differences in consistency with domain literature.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper introduces an automated pipeline to assess how well benchmark causal graphs align with recent domain research. The pipeline retrieves thousands of papers from scientific databases and uses large language models to judge consistency with the graphs. It applies this to 11 real-world benchmarks, processing over 38,000 papers in total. The results reveal significant variation in how consistent these benchmarks are with current knowledge. This matters because misaligned benchmarks can mislead evaluations of causal discovery methods, especially those based on LLMs.

Core claim

The paper establishes that popular benchmarks for causal discovery vary significantly in their consistency with domain research papers, as evaluated by an LLM-based pipeline that checks 38,081 papers across 11 benchmarks.

What carries the argument

An automated pipeline that retrieves relevant research papers and prompts LLMs to evaluate consistency between benchmark causal graphs and the papers.

If this is right

  • Evaluations of causal discovery methods may be affected differently depending on the benchmark chosen.
  • LLM-based causal discovery methods are particularly sensitive to benchmark misalignment due to new discoveries.
  • Future benchmark creation should incorporate consistency checks with domain literature.
  • Researchers should select benchmarks with higher consistency for more reliable evaluations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Human verification of the LLM judgments could refine the consistency scores.
  • This approach could be extended to other AI benchmarks beyond causal discovery.
  • Periodic re-evaluation of benchmarks as new research emerges would help maintain their validity.

Load-bearing premise

The LLM prompts produce judgments of consistency that accurately reflect the true alignment between graphs and papers without systematic bias.

What would settle it

If independent human experts review a sample of the paper-graph pairs and find substantially different consistency rates than the LLM pipeline, the reported variation would not hold.

Figures

Figures reproduced from arXiv: 2606.01789 by Chen Wang, Chihui Chen, Lina Yao, Yuzhe Zhang.

Figure 1
Figure 1. Figure 1: All benchmarks by year period. before 2000 2000-2004 2005-2009 2010-2014 2015-2019 2020-2024 after 2024 Year Period 0 2 4 6 8 10 Number of Benchmarks 8 2 3 1 4 3 0 Benchmark Counts by Year Period (paper # > 2) [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Benchmarks used in at least 3 papers by year [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Number of papers/year by searching “causal [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Number of papers/year by searching “causal [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The causal graph of the Asia benchmark dataset: a semi-synthetic benchmark dataset. Vi , Vj and Z ⊆ V \ {Vi , Vj}: if Z d-separates Vi and Vj , the pair of variables are mutually in￾dependent conditioned on Z, otherwise they are correlated. Note that Z can be empty. Example 3.1 [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Benchmarks’ inconsistency rates. sachs fmri insurance ecoli cancer alzheimer diabetes alarm child asia Benchmark 0 10 20 30 40 50 60 Inconsistency Rate (%) 26.3% 28.6% 30.7% 32.8% 37.9% 39.2% 45.7% 47.2% 53.9% 57.4% Benchmark Inconsistency Rate Since Benchmark Cutoff Year [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Benchmarks’ inconsistency rates since the [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Benchmarks’ relevant paper number per year period. [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Benchmarks’ inconsistent paper number per year period. [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Benchmark numbers used in LLM-based papers. [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Top 20 benchmarks used in arXiv LLM-based causal discovery papers. [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Inconsistent paper number per year of Sachs. [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Inconsistent paper number per year of Insurance. [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Inconsistent paper number per year of Ecoli. [PITH_FULL_IMAGE:figures/full_fig_p027_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Inconsistent paper number per year of Alzheimer. [PITH_FULL_IMAGE:figures/full_fig_p027_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Inconsistent paper number per year of Alarm. [PITH_FULL_IMAGE:figures/full_fig_p028_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Inconsistent paper number per year of Cancer. [PITH_FULL_IMAGE:figures/full_fig_p028_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Inconsistent paper number per year of Arctic Sea Ice. [PITH_FULL_IMAGE:figures/full_fig_p029_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Inconsistent paper number per year of Fmri. [PITH_FULL_IMAGE:figures/full_fig_p029_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Inconsistent paper number per year of Diabetes. [PITH_FULL_IMAGE:figures/full_fig_p030_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Inconsistent paper number per year of Child. [PITH_FULL_IMAGE:figures/full_fig_p030_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Inconsistent paper number per year of Asia. [PITH_FULL_IMAGE:figures/full_fig_p031_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Retrieved papers by year. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Retrieved papers that contain relevant information by year. [PITH_FULL_IMAGE:figures/full_fig_p032_25.png] view at source ↗
read the original abstract

In graphical causal model, causal discovery aims to construct a causal graph based on numerical data and domain knowledge in plain text. However, the evaluation of causal discovery methods remains a challenge in the area as the progress of domain researches often makes benchmark causal graphs contain mis-aligned knowledge. This problem especially affects the evaluation of large language model (LLM) based causal discovery methods as they are sensitive to the new discoveries in the literature. This work is the first to systematically study the quality of benchmark causal graphs. Specifically, we design a pipeline that automatically retrieves relevant research papers from scientific databases, and prompts LLMs to check the consistency between the benchmark causal graphs and domain research papers. We evaluate 11 popular real-world benchmarks, for which our pipeline in total proceeds 38,081 domain papers. Our results show that popular benchmarks vary significantly in their consistency with domain research, with clear implications for causal discovery research.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper introduces an automated pipeline that retrieves relevant domain papers (38,081 in total) from scientific databases and prompts LLMs to judge consistency between the causal graphs in 11 popular real-world benchmarks and the retrieved literature. It reports that the benchmarks vary significantly in their consistency with domain research and draws implications for causal discovery evaluation, especially for LLM-based methods sensitive to new findings.

Significance. If the LLM consistency judgments prove reliable, the work identifies a previously unquantified source of misalignment in standard benchmarks and supplies a scalable method for ongoing evaluation. The scale of the literature search is a clear strength. The result would directly affect how benchmark graphs are trusted when assessing causal discovery algorithms.

major comments (2)
  1. [Abstract (pipeline description)] The central quantitative results rest entirely on LLM judgments of consistency, yet the manuscript provides no validation of those judgments (human-expert agreement rates, calibration set, inter-LLM consistency, or handling of ambiguous papers). This assumption is load-bearing because every reported variation across the 11 benchmarks is derived from the LLM outputs.
  2. [Abstract (pipeline description)] No information is given on prompt engineering, sensitivity analysis, or inter-rater metrics for the LLM consistency checks. Without these, it is impossible to determine whether the observed differences among benchmarks reflect genuine literature misalignment or artifacts of the judge.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments highlighting the need for validation and methodological transparency in our LLM-based pipeline. We agree these are important and will incorporate the requested details and analyses in a major revision.

read point-by-point responses

  1. Referee: The central quantitative results rest entirely on LLM judgments of consistency, yet the manuscript provides no validation of those judgments (human-expert agreement rates, calibration set, inter-LLM consistency, or handling of ambiguous papers). This assumption is load-bearing because every reported variation across the 11 benchmarks is derived from the LLM outputs.

    Authors: We acknowledge the manuscript currently lacks explicit validation of the LLM consistency judgments. In the revision we will add a dedicated validation subsection that reports: (i) human-expert agreement rates on a stratified sample of 500 paper-graph pairs, (ii) a calibration set of 100 expert-annotated examples, (iii) inter-LLM consistency across GPT-4, Claude-3, and Llama-3, and (iv) explicit handling rules for ambiguous papers (e.g., “insufficient information” category with frequency statistics). These additions will directly quantify the reliability of the reported benchmark differences. revision: yes

  2. Referee: No information is given on prompt engineering, sensitivity analysis, or inter-rater metrics for the LLM consistency checks. Without these, it is impossible to determine whether the observed differences among benchmarks reflect genuine literature misalignment or artifacts of the judge.

    Authors: We agree that prompt details and robustness checks are missing. The revised manuscript will include: the complete system and user prompts in an appendix, a description of the iterative prompt-engineering process (including few-shot examples and chain-of-thought instructions), sensitivity results across three prompt variants and two temperature settings, and inter-rater metrics (both LLM-LLM and LLM-human) already referenced in the new validation section. These additions will allow readers to assess whether the observed benchmark variations are robust to judge artifacts. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical pipeline is self-contained

full rationale

The paper describes an empirical pipeline that retrieves domain papers from databases and applies LLM prompts to judge consistency with benchmark causal graphs. No equations, fitted parameters, predictions derived from fits, or self-referential definitions appear in the derivation. The central claim (variation in benchmark consistency) is produced by processing external literature rather than by construction from the paper's own inputs or prior self-citations. Self-citation load-bearing, ansatz smuggling, and renaming patterns are absent. The absence of validation metrics for LLM judgments is a potential reliability concern but does not create circularity under the defined criteria.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim depends on the unverified reliability of LLM consistency judgments and on the assumption that the retrieved papers form a representative sample of domain knowledge.

axioms (2)
  • domain assumption LLMs prompted with benchmark graphs and paper abstracts can produce accurate and unbiased consistency assessments.
    The pipeline is built around this capability; no human validation step is mentioned in the abstract.
  • domain assumption The 38,081 retrieved papers constitute a sufficient and unbiased sample of the relevant domain literature for each benchmark.
    The scale is reported but selection criteria and coverage are not detailed in the abstract.

pith-pipeline@v0.9.1-grok · 5681 in / 1265 out tokens · 19728 ms · 2026-06-28T14:23:34.617490+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

96 extracted references · 9 canonical work pages · 1 internal anchor

  1. [1]

    Journal of the Royal Statistical Society: Series B (Methodological), 50(2):157–194

    Local computations with probabilities on graphi- cal structures and their application to expert systems. Journal of the Royal Statistical Society: Series B (Methodological), 50(2):157–194. Ahmed Abdulaal, Adamos Hadjivasiliou, Nina Montana-Brown, Tiantian He, Ayodeji Ijishakin, Ivana Drobnjak, Daniel Castro, and Daniel Alexander

  2. [2]

    InInternational Conference on Learning Representations, volume 2024, pages 57559–57610

    Causal modelling agents: Causal graph dis- covery through synergising metadata-and data-driven reasoning. InInternational Conference on Learning Representations, volume 2024, pages 57559–57610. Bruce Abramson, John Brown, Ward Edwards, Allan Murphy, and Robert L Winkler. 1996. Hailfinder: A bayesian system for forecasting severe weather. International Jou...

    2024

  3. [3]

    Virginia Aglietti, Alan Malek, Ira Ktena, and Silvia Chiappa

    Collaborative causal discovery with atomic interventions.Advances in Neural Information Pro- cessing Systems, 34:12761–12773. Virginia Aglietti, Alan Malek, Ira Ktena, and Silvia Chiappa. 2023. Constrained causal bayesian opti- mization. InInternational Conference on Machine Learning, pages 304–321. PMLR. Sina Akbari, Ehsan Mokhtarian, AmirEmad Ghassami, ...

    work page arXiv 2023

  4. [4]

    Springer

    Proceedings, pages 247–256. Springer. Alexis Bellot, Junzhe Zhang, and Elias Bareinboim

  5. [5]

    InProceedings of the AAAI Conference on Artificial Intelligence, vol- ume 38, pages 11043–11051

    Scores for learning discrete causal graphs with unobserved confounders. InProceedings of the AAAI Conference on Artificial Intelligence, vol- ume 38, pages 11043–11051. John Binder, Daphne Koller, Stuart Russell, and Keiji Kanazawa. 1997. Adaptive probabilistic networks with hidden variables.Machine Learning, 29(2):213– 244. Philippe Brouillard, Chandler ...

    work page arXiv 1997

  6. [6]

    Haoyue Dai, Yiwen Qiu, Ignavier Ng, Xinshuai Dong, Peter Spirtes, and Kun Zhang

    Bcd nets: Scalable variational approaches for bayesian causal discovery.Advances in Neural Information Processing Systems, 34:7095–7110. Haoyue Dai, Yiwen Qiu, Ignavier Ng, Xinshuai Dong, Peter Spirtes, and Kun Zhang. 2025. Latent variable causal discovery under selection bias.arXiv preprint arXiv:2512.11219. Haoyue Dai, Peter Spirtes, and Kun Zhang. 2022...

    work page arXiv 2025

  7. [7]

    Anish Dhir, Ruby Sedgwick, Avinash Kori, Ben Glocker, and Mark Van Der Wilk

    Bivariate causal discovery using bayesian model selection.arXiv preprint arXiv:2306.02931. Anish Dhir, Ruby Sedgwick, Avinash Kori, Ben Glocker, and Mark Van Der Wilk. 2024. Contin- uous bayesian model selection for multivariate causal discovery.arXiv preprint arXiv:2411.10154. Shuyu Dong, Michele Sebag, Kento Uemura, Akito Fujii, Shuang Chang, Yusuke Koy...

    work page arXiv 2024

  8. [8]

    Hwang, Y

    Multimodal pooled perturb-cite-seq screens in patient models define mechanisms of cancer immune evasion.Nature genetics, 53(3):332–341. Jensen FV . 1996.An introduction to Bayesian networks. London, UK: UCL Press. Amanda Gentzel, Dan Garant, and David Jensen. 2019. The case for evaluating causal models using inter- ventional measures and empirical data.Ad...

    work page arXiv 1996

  9. [9]

    Efficient Causal Graph Discovery Using Large Language Models

    An expert system for control of waste water treatment—a pilot project. Technical report, Techni- cal report, Judex Datasystemer A/S, Aalborg, 1989. In Danish. Thomas Jiralerspong, Xiaoyin Chen, Yash More, Vedant Shah, and Yoshua Bengio. 2024. Efficient causal graph discovery using large language models. Preprint, arXiv:2402.01207. Diviyan Kalainathan and ...

    work page internal anchor Pith review Pith/arXiv arXiv 1989

  10. [10]

    Peter JF Lucas, Linda C Van der Gaag, and Ameen Abu- Hanna

    Can large language models build causal graphs? InNeurIPS 2022 Workshop on Causality for Real-world Impact. Peter JF Lucas, Linda C Van der Gaag, and Ameen Abu- Hanna. 2004. Bayesian networks in biomedicine and health-care. Alessandro Magrini, Stefano Di Blasi, Federico Mattia Stefanini, and 1 others. 2017. A conditional linear gaussian network to assess t...

    2022

  11. [11]

    InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 8975–8982

    Discovering fully oriented causal networks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 8975–8982. Ehsan Mokhtarian, Mohmmadsadegh Khorasani, Jalal Etesami, and Negar Kiyavash. 2023. Novel ordering- based approaches for causal structure learning in the presence of unobserved variables. InProceedings of the AAAI Confer...

    work page doi:10.24432/c5fs30 2023

  12. [12]

    Tim Van den Bulcke, Koenraad Van Leemput, Bart Naudts, Piet van Remortel, Hongwu Ma, Alain Ver- schoren, Bart De Moor, and Kathleen Marchal

    International Conference on Learning Repre- sentations, ICLR. Tim Van den Bulcke, Koenraad Van Leemput, Bart Naudts, Piet van Remortel, Hongwu Ma, Alain Ver- schoren, Bart De Moor, and Kathleen Marchal. 2006. Syntren: a generator of synthetic gene expression data for design and analysis of structure learning al- gorithms.BMC bioinformatics, 7(1):43. Anike...

    2006

  13. [13]

    InInternational Conference on Ma- chine Learning, pages 50650–50668

    Optimal kernel choice for score function-based causal discovery. InInternational Conference on Ma- chine Learning, pages 50650–50668. PMLR. 14 X Wang, Y Du, S Zhu, L Ke, Z Chen, J Hao, and J Wang

  14. [14]

    InProceedings of the Thirtieth International Joint Conference on Artificial Intelli- gence (IJCAI-21), pages 3566–3573

    Ordering-based causal discovery with rein- forcement learning. InProceedings of the Thirtieth International Joint Conference on Artificial Intelli- gence (IJCAI-21), pages 3566–3573. IJCAI Interna- tional Joint Conferences on Artificial Intelligence Organization. Yunxia Wang, Fuyuan Cao, Kui Yu, and Jiye Liang

  15. [15]

    InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 8584–8593

    Efficient causal structure learning from mul- tiple interventional datasets with unknown targets. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 8584–8593. Yunxia Wang, CAO Fuyuan, Kui Yu, and Jiye Liang. 2025a. Federated causal structure learning with non- identical variable sets. InForty-second International Conference...

  16. [16]

    InProceedings of the AAAI Con- ference on Artificial Intelligence, volume 40, pages 36757–36765

    Robust causal discovery under imperfect struc- tural constraints. InProceedings of the AAAI Con- ference on Artificial Intelligence, volume 40, pages 36757–36765. Zidong Wang, Fei Liu, Qi Feng, Qingfu Zhang, and Xi- aoguang Gao. 2025b. Llm-enhanced score function evolution for causal structure learning. InProceed- ings of the Thirty-Fourth International J...

    work page arXiv 1994

  17. [17]

    original paper

    Causal-driven skill prerequisite structure dis- covery. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 20604– 20612. Yan Zeng, Shohei Shimizu, Ruichu Cai, Feng Xie, Mi- chio Yamamoto, and Zhifeng Hao. 2021. Causal discovery with multi-domain lingam for latent fac- tors. InCausal Analysis Workshop Series, pages 1–4. PMLR....

    work page arXiv 2021

  18. [18]

    Sachs Original paper (Sachs et al., 2005) Papers using it: (Eulig et al., 2025; Shen et al., 2025; Roy et al., 2025; Shahverdikondori et al., 2024; Kang et al., 2026; Aglietti et al., 2023; Olko et al., 2023; Annadani et al., 2023; Perry et al., 2022; Dai et al., 2022; Addanki and Kasiviswanathan, 2021; Cundy et al., 2021; Wang et al., 2025a; Li et al., 2...

    2005

  19. [19]

    Child Original paper (Spiegelhalter, 1992) Papers using it: (Shen et al., 2025; Peyrard and West, 2020; Olko et al., 2023; Wang et al., 2025a, 2024; Vashishtha et al., 2025; Duong et al., 2025; Ke et al., 2022; Lippe et al., 2021; Guo et al., 2024a; Zhang et al., 2023b, 2022; Wang et al., 2022; Ling et al., 2025b; Guo et al., 2024b; Cui et al., 2022)

    1992

  20. [20]

    Alarm Original paper (Beinlich et al., 1989) Papers using it: (Akbari et al., 2021; Roy et al., 2025; Xie et al., 2024; Li et al., 2022; Olko et al., 2023; Wang et al., 2025a; Duong et al., 2025; Lippe et al., 2021; Guo et al., 2024a; Zhang et al., 2023b, 2022; Wang et al., 2022; Zhang et al., 2021; Ling et al., 2025b; Guo et al., 2024b; Cui et al., 2022)

    1989

  21. [21]

    Asia Original paper (lau, 1988) Papers using it: (Shen et al., 2025; Roy et al., 2025; Shahverdikondori et al., 2024; Olko et al., 2023; Kocaoglu, 2023; Addanki and Kasiviswanathan, 2021; Vashishtha et al., 2025; Duong et al., 2025; Ke et al., 2022; Lippe et al., 2021; Bellot et al., 2024; Zhang et al., 2023b, 2022)

    1988

  22. [22]

    Insurance Original paper (Binder et al., 1997) Papers using it: (Mokhtarian et al., 2023; Akbari et al., 2021; Feng et al., 2025a,b; Peyrard and West, 2020; Wang et al., 2025a; Guo et al., 2024a; Zhang et al., 2022; Wang et al., 2022; Zhang et al., 2021; Guo et al., 2024b; Cui et al., 2022)

    1997

  23. [23]

    Cancer Original paper (Korb and Nicholson, 2010) Papers using it: (Feng et al., 2025a,b; Shahverdikondori et al., 2024; Peyrard and West, 2020; Olko et al., 2023; Vashishtha et al., 2025; Duong et al., 2025; Lippe et al., 2021; Zhang et al., 2023b, 2022)

    2010

  24. [24]

    Barley Original paper (Kristensen and Rasmussen, 2002) Papers using it: (Akbari et al., 2021; Wang et al., 2025a; Ling et al., 2025c; Zhang et al., 2022, 2021; Ling et al., 2025b)

    2002

  25. [25]

    Hailfinder Original paper (Abramson et al., 1996) Papers using it: (Mokhtarian et al., 2023; Akbari et al., 2021; Li et al., 2022; Zhang et al., 2021; Ling et al., 2025b; Cui et al., 2022)

    1996

  26. [26]

    Fmri hippocampus Original paper (Poldrack et al., 2015) Papers using it: (Li et al., 2024b; Chen et al., 2024; Zeng et al., 2021)

    2015

  27. [27]

    Alzheimer Original paper (Petersen et al., 2010; Shen et al., 2020) Papers using it: (Abdulaal et al., 2024; Feng et al., 2025a; Vashishtha et al., 2025)

    2010

  28. [28]

    Arctic sea ice Original paper (Huang et al., 2021) Papers using it: (Abdulaal et al., 2024; Feng et al., 2025b; Kiciman et al., 2023) 17

    2021

  29. [29]

    Diabetes Original paper (Long et al., 2022) Papers using it: (Feng et al., 2025a,b; Lippe et al., 2021)

    2022

  30. [30]

    Ecoli70(100) Original paper (Schäfer and Strimmer, 2005) Papers using it: (Mokhtarian et al., 2023; Akbari et al., 2021; Kang et al., 2026; Chen et al., 2021)

    2005

  31. [31]

    Gene expression Original paper (Sethuraman et al., 2023) Papers using it: (Guruswamy Sethuraman and Fekri, 2026; Xie et al., 2024; Li et al., 2025b; Guo et al., 2024a; Ling et al., 2025b)

    2023

  32. [32]

    Hepar2 Original paper (Onisko, 2003) Papers using it: (Mokhtarian et al., 2023; Roy et al., 2025; Li et al., 2022)

    2003

  33. [33]

    Pigs Original paper (FV, 1996) Papers using it: (Lippe et al., 2021; Guo et al., 2024a; Ling et al., 2025b)

    1996

  34. [34]

    Reged Original paper (Statnikov et al., 2015) Papers using it: (Mian et al., 2021; Kaltenpoth and Vreeken, 2023; Guo et al., 2024b)

    2015

  35. [35]

    Andes Original paper (Conati et al., 1997) Papers using it: (Xie et al., 2024; Li et al., 2022)

    1997

  36. [36]

    Arth150 Original paper (Opgen-Rhein and Strimmer, 2007) Papers using it: (Mokhtarian et al., 2023; Akbari et al., 2021; Kang et al., 2026)

    2007

  37. [37]

    Auto mpg Original paper (Quinlan, 1993) Papers using it: (Eulig et al., 2025; Shen et al., 2025)

    1993

  38. [38]

    Carpo Original paper Link Papers using it: (Mokhtarian et al., 2023; Akbari et al., 2021)

    2023

  39. [39]

    HK stock Original paper (Huang et al., 2020) Papers using it: (Li et al., 2024b; Cai et al., 2023)

    2020

  40. [40]

    Link Original paper (Jensen and Kong, 1999) Papers using it: (Wang et al., 2022; Ling et al., 2025b)

    1999

  41. [41]

    Mildew Original paper (Jensen and Jensen, 1996) Papers using it: (Xie et al., 2024; Li et al., 2022)

    1996

  42. [42]

    Neuropain Original paper (Tu et al., 2019) Papers using it: (Liu et al., 2024a; Feng et al., 2025a; Kiciman et al., 2023; Vashishtha et al., 2025) 18

    2019

  43. [43]

    Obesity Original paper (Long et al., 2022) Papers using it: (Feng et al., 2025a,b)

    2022

  44. [44]

    Syntren Original paper (Van den Bulcke et al., 2006) Papers using it: (Dhir et al., 2024; Ling et al., 2025b)

    2006

  45. [45]

    WIN95PTS Original paper (Heckerman et al., 1995) Papers using it: (Xie et al., 2024; Wang et al., 2022; Zhang et al., 2021)

    1995

  46. [46]

    CE-Tueb Original paper (Mooij et al., 2016) Papers using it: (Dhir et al., 2023)

    2016

  47. [47]

    CE-cha Original paper (Guyon et al., 2019) Papers using it: (Dhir et al., 2023)

    2019

  48. [48]

    Cognition and Aging in the Chronic Fatigue Syndrome Original paper (Heins et al., 2013) Papers using it: (Qiao et al., 2024a)

    2013

  49. [49]

    DWDClimate Original paper (Mooij et al., 2016) Papers using it: (Shen et al., 2025)

    2016

  50. [50]

    MAGIC-IRRI Original paper (Scutari, 2016) Papers using it: (Kang et al., 2026)

    2016

  51. [51]

    MAGIC-NIAB Original paper (Scutari et al., 2014) Papers using it: (Kang et al., 2026)

    2014

  52. [52]

    New York Times Original paper Link Papers using it: (Liu et al., 2024a)

  53. [53]

    Pittsburgh Bridges Original paper (Reich and Fenves, 1989) Papers using it: (Ni, 2022)

    1989

  54. [54]

    Categorical Cause-Effect Pairs Original paper (Ni, 2022) Papers using it: (Ni, 2022)

    2022

  55. [55]

    Abalone Original paper (Warwick et al., 1994) Papers using it: (Ni, 2022)

    1994

  56. [56]

    Alcohol Original paper (Long et al., 2022) Papers using it: (Feng et al., 2025b) 19

    2022

  57. [57]

    Algibra I Original paper Link Papers using it: (Yu et al., 2024)

    2024

  58. [58]

    APM Original paper Link Papers using it: (Eulig et al., 2025)

    2025

  59. [59]

    Apple gastronome Original paper (Liu et al., 2024a) Papers using it: (Feng et al., 2025a)

  60. [60]

    Big five Original paper Link Papers using it: (Dai et al., 2025; Dong et al., 2024)

    2025

  61. [61]

    Brain tumor Original paper Link Papers using it: (Liu et al., 2024a)

  62. [62]

    Chemical Original paper (Ke et al., 2021) Papers using it: (Zhao et al., 2025)

    2021

  63. [63]

    (no?)Chemistry image Original paper (Schäfer and Strimmer, 2005) Papers using it: tba

    2005

  64. [64]

    climatic analysis Original paper (Compo et al., 2011) Papers using it: (Liu et al., 2024a)

    2011

  65. [65]

    COVID-19 Original paper Link Papers using it: (Wang et al., 2025b; Vashishtha et al., 2025)

    2025

  66. [66]

    Credit Original paper (Quinlan, 1987) Papers using it: (Li et al., 2022)

    1987

  67. [67]

    dream Original paper (Kalainathan and Goudet, 2019) Papers using it: (Roy et al., 2025)

    2019

  68. [68]

    football Original paper Link Papers using it: (Qiao et al., 2024b)

  69. [69]

    G7 Original paper (Demirer et al., 2018) Papers using it: (Jalaldoust et al., 2022)

    2018

  70. [70]

    General social survey Original paper Link Papers using it: (Li et al., 2025b) 20

  71. [71]

    IHDP Original paper (Hill, 2011) Papers using it: (Ashman et al., 2023)

    2011

  72. [72]

    Lucas Original paper (Lucas et al., 2004) Papers using it: (Roy et al., 2025)

    2004

  73. [73]

    Magnetic Original paper (Hwang et al., 2024) Papers using it: (Zhao et al., 2025)

    2024

  74. [74]

    (tba)Micro24 Original paper (Schäfer and Strimmer, 2005) Papers using it: tba

    2005

  75. [75]

    (tba)Micro25 Original paper (Schäfer and Strimmer, 2005) Papers using it: tba

    2005

  76. [76]

    Munin Original paper (Andreassen et al., 1989) Papers using it: (Dong et al., 2025; Wang et al., 2022)

    1989

  77. [77]

    Pacific walker circulation Original paper (Runge et al., 2019) Papers using it: (Liu and Kuang, 2023)

    2019

  78. [78]

    Pathfinder Original paper (Heckerman et al., 1992) Papers using it: (Zhang et al., 2021)

    1992

  79. [79]

    Pharmacokinetics Original paper (Grzegorzewski et al., 2021) Papers using it: (Li et al., 2024a)

    2021

  80. [80]

    Physics Original paper (Lee et al., 2024) Papers using it: (Kang et al., 2026)

    2024

Showing first 80 references.