pith. sign in

arxiv: 2605.18630 · v1 · pith:KZS2V7M3new · submitted 2026-05-18 · 💻 cs.AI · physics.comp-ph

SCICONVBENCH: Benchmarking LLMs on Multi-Turn Clarification for Task Formulation in Computational Science

Nithin Somasekharan , Youssef Hassan , Shiyao Lin , Gihan Panapitiya , Patrick Emami , Anurag Acharya , Sameera Horawalavithana , Shaowu Pan This is my paper

Pith reviewed 2026-05-20 10:31 UTC · model grok-4.3

classification 💻 cs.AI physics.comp-ph
keywords LLM benchmarkingmulti-turn clarificationscientific task formulationdisambiguationinconsistency resolutioncomputational scienceconversational grounding
0
0 comments X

The pith

Frontier LLMs resolve only 52.7 percent of disambiguation cases when clarifying ill-posed scientific task requests in fluid mechanics.

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

The paper introduces SCICONVBENCH to test how large language models handle multi-turn clarification when users give vague or internally contradictory requests in computational science. The benchmark covers four domains: fluid mechanics, solid mechanics, materials science, and partial differential equations. It measures two main skills: drawing out missing information and spotting and correcting contradictions. A structured task ontology combined with rubric scoring tracks clarification behavior, how well the model stays grounded in the conversation, and whether the final specification matches the original intent. Sympathetic readers would care because real scientific assistance begins with imprecise problems that must be refined through dialogue before any computation or analysis can proceed reliably.

Core claim

SCICONVBENCH pairs a structured task ontology with a rubric-based evaluation framework to measure LLM performance on eliciting missing information and resolving inconsistencies during scientific task formulation. Current frontier models perform relatively well on inconsistency resolution, but even the best model resolves only 52.7 percent of the disambiguation cases in fluid mechanics. Frontier LLMs frequently make silent assumptions and perform implicit specification repairs that are not grounded in the conversation with users.

What carries the argument

SCICONVBENCH benchmark that uses a structured task ontology paired with rubric-based scoring to evaluate clarification behavior, conversational grounding, and final-specification fidelity across multi-turn scientific dialogues.

If this is right

  • Frontier LLMs handle inconsistency resolution better than they handle disambiguation of missing information.
  • Even the strongest model reaches only 52.7 percent success on disambiguation tasks within fluid mechanics.
  • Models commonly insert silent assumptions and ungrounded repairs instead of staying within the user conversation.
  • Reliable computational science assistants require explicit evaluation of upstream conversational reasoning before any computation begins.

Where Pith is reading between the lines

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

  • Training pipelines for scientific assistants could add targeted examples of iterative clarification to reduce reliance on unstated assumptions.
  • Comparable benchmarks may be useful in adjacent domains such as experimental biology or chemistry where initial requests are also often ill-posed.
  • Developers might prioritize datasets that reward explicit grounding over implicit repair when building next-generation scientific dialogue systems.

Load-bearing premise

The structured task ontology paired with the rubric-based evaluation framework accurately and comprehensively captures real-world multi-turn clarification needs in computational science task formulation.

What would settle it

A side-by-side test in which the benchmark cases are replaced by live multi-turn dialogues between the model and actual domain experts, then measuring whether the model's final specification matches the expert's intended task at a rate significantly above or below the reported 52.7 percent.

Figures

Figures reproduced from arXiv: 2605.18630 by Anurag Acharya, Gihan Panapitiya, Nithin Somasekharan, Patrick Emami, Sameera Horawalavithana, Shaowu Pan, Shiyao Lin, Youssef Hassan.

Figure 1
Figure 1. Figure 1: Flow over a cylinder showing how skipped clarification leads to a wrong flow regime. Large language models (LLMs) are increasingly used as con￾versational interfaces for computational science, supporting scientific question answering [58], code generation [60], and agentic execution of scientific simulation workflows [70, 52]. Yet most scientific benchmarks for LLMs assess these ca￾pabilities given complet… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of SCICONVBENCH. The benchmark spans four computational science domains and two task types. For each instance, a model interacts with a simulated user to resolve missing or conflicting information and then produces a final specification. Evaluation compares the final specification against the reference specification while using the full conversation as context to assess whether the model resolved … view at source ↗
Figure 3
Figure 3. Figure 3: Case distribu￾tion across the four SCICON￾VBENCH domains. Following recent conversational benchmark design [7, 17, 69], we separate final output success from conversation-grounded success, since a model may guess or silently repair missing scientific details without resolving them through dialogue. Each instance is evaluated as a structured judgment problem using the conversation transcript, the final spec… view at source ↗
Figure 4
Figure 4. Figure 4: Case level resolution rate (Section 3.6) comparison among different models for the different [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Component level resolution rate (Section 3.6) comparison among different models for the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Pareto analysis across Capability, Robustness, and Usability. Top row: disambiguation. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Outcomes on general numeric prompts (textbook-style problems without a fixed tool stack). Each bar decomposes outcomes into Conversation-Grounded Resolution Rate (CGRR, colored), Silent Resolution Rate (SRR, grey), and unresolved cases; the bar top is the Final Resolution Rate (FRR). Three domains are available in this split (fluid mechanics, solid mechanics, materials science). Two qualitative patterns ar… view at source ↗
Figure 8
Figure 8. Figure 8: Outcomes on tool-use prompts (OpenFOAM, FEA, materials-science tools, and PDE solver setup). Bars use the same decomposition as [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Per-domain breakdown (FRR(d) and CGRR(d)). Denominator: total missing entities or planted inconsistencies per (domain, model, task). E.3 Full domain-level results Tables 3 and 4 report the full per-domain breakdown of all outcome and diagnostic metrics used in the paper [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Unguided vs. guided agent for GEMINI 2.5 PRO across all four domains. Bar top is FRR (%); the colored portion is CGRR (conversation-grounded) and the hatched portion is SRR (silent resolution). Same filtering, case pool, judge and SRR correction as the main-text figures. On inconsistency, the guided agent substantially improves CGRR in fluid mechanics (+18pp) and materials science (+11pp), with smaller ga… view at source ↗
read the original abstract

Large Language Models (LLMs) are increasingly deployed as scientific AI as- sistants, and a growing body of benchmarks evaluates their capabilities across knowledge retrieval, reasoning, code generation, and tool use. These evaluations, however, typically assume the scientific problem is already well-posed, whereas practical scientific assistance often begins with an ill-posed user request that must be refined through dialogue before any computation, analysis, or experiment can be carried out reliably. We introduce SCICONVBENCH, a benchmark for multi- turn clarification in scientific task formulation across four computational science problem domains: fluid mechanics, solid mechanics, materials science, and par- tial differential equations (PDEs). SCICONVBENCH targets two complementary capabilities: eliciting missing information (disambiguation) and detecting and correcting erroneous requests containing internally contradictory information (in- consistency resolution). Our benchmark pairs a structured task ontology with a rubric-based evaluation framework, enabling systematic measurement of LLM per- formance across three dimensions: clarification behavior, conversational grounding, and final-specification fidelity. Current frontier models perform relatively well on inconsistency resolution, but even the best model resolves only 52.7% of the disambiguation cases in fluid mechanics. We further find that frontier LLMs fre- quently make silent assumptions and perform implicit specification repairs that are not grounded in the conversation with users. SCICONVBENCH establishes a foundation for evaluating the upstream conversational reasoning that a reliable computational science assistant requires. The code and data can be found at https://github.com/csml-rpi/SciConvBench.

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 / 2 minor

Summary. The paper introduces SCICONVBENCH, a benchmark for multi-turn clarification in scientific task formulation across fluid mechanics, solid mechanics, materials science, and PDEs. It targets disambiguation of missing information and inconsistency resolution using a structured task ontology paired with a rubric-based evaluation framework that scores clarification behavior, conversational grounding, and final-specification fidelity. Key empirical results include frontier models resolving only 52.7% of disambiguation cases in fluid mechanics, with frequent silent assumptions and ungrounded implicit repairs observed across models.

Significance. If the benchmark's ontology and rubric prove faithful to real usage, the work is significant for highlighting upstream conversational limitations in LLMs deployed as scientific assistants. The open release of code and data at the provided GitHub link enables reproducibility and community extension; the concrete performance gaps (e.g., 52.7%) and qualitative observations about implicit specification repairs supply falsifiable targets for improving scientific AI reliability.

major comments (2)
  1. [Benchmark Construction] Benchmark construction (methods section on dataset generation): the central claims about model performance gaps and silent assumptions rest on the assumption that ontology-derived disambiguation and inconsistency instances faithfully proxy real-world scientist-LLM interactions. The paper generates cases via structured ontology rather than sampling logged queries or expert-elicited scenarios; without a validation study (e.g., expert rating of realism or comparison to actual clarification dialogues), the 52.7% fluid-mechanics figure and the qualitative finding risk being benchmark artifacts rather than model properties.
  2. [Evaluation Framework] Evaluation framework (rubric and scoring section): the headline disambiguation rate and inconsistency-resolution results depend on the rubric accurately capturing grounding and fidelity. The manuscript should report inter-rater reliability, rubric development process, and any statistical tests for the reported percentages; absent these, the quantitative claims lack the robustness needed to support the paper's conclusions about frontier-model limitations.
minor comments (2)
  1. [Abstract] Abstract: the 52.7% figure is reported without naming the best-performing model; adding this detail would improve immediate interpretability of the main result.
  2. [Discussion] The paper would benefit from an explicit limitations subsection discussing potential mismatches between the four chosen domains and broader computational science workflows.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive feedback on SCICONVBENCH. We address the major comments point-by-point below, agreeing to strengthen the manuscript with additional details and validation where appropriate.

read point-by-point responses

  1. Referee: Benchmark construction (methods section on dataset generation): the central claims about model performance gaps and silent assumptions rest on the assumption that ontology-derived disambiguation and inconsistency instances faithfully proxy real-world scientist-LLM interactions. The paper generates cases via structured ontology rather than sampling logged queries or expert-elicited scenarios; without a validation study (e.g., expert rating of realism or comparison to actual clarification dialogues), the 52.7% fluid-mechanics figure and the qualitative finding risk being benchmark artifacts rather than model properties.

    Authors: We recognize the value of validating the benchmark instances against real-world data. Our structured task ontology enables comprehensive and reproducible coverage of clarification needs in computational science domains, which would be challenging with sparse logged interactions. Nevertheless, we agree that empirical validation would bolster confidence in the results. In the revised manuscript, we will add a dedicated subsection describing the ontology development process in greater detail and report on a pilot study in which domain experts assess the realism of generated cases. We will also update the limitations section to discuss this aspect transparently. revision: yes

  2. Referee: Evaluation framework (rubric and scoring section): the headline disambiguation rate and inconsistency-resolution results depend on the rubric accurately capturing grounding and fidelity. The manuscript should report inter-rater reliability, rubric development process, and any statistical tests for the reported percentages; absent these, the quantitative claims lack the robustness needed to support the paper's conclusions about frontier-model limitations.

    Authors: We agree that providing more details on the evaluation framework will improve the paper's rigor. The rubric was iteratively developed by the author team, drawing on examples from each domain to define criteria for clarification behavior, conversational grounding, and final-specification fidelity. In the revision, we will include a full account of this development process. Furthermore, we will perform and report an inter-rater reliability assessment on a subset of evaluated conversations and include appropriate statistical measures, such as confidence intervals, for the key performance percentages. revision: yes

Circularity Check

0 steps flagged

No significant circularity in benchmark construction or performance reporting

full rationale

The paper introduces SCICONVBENCH as a new benchmark consisting of a structured task ontology and rubric-based evaluation for multi-turn clarification tasks in computational science domains. Reported metrics such as the 52.7% disambiguation resolution rate in fluid mechanics are obtained by directly applying frontier LLMs to the generated test cases and scoring their responses against the rubric. These are empirical measurements on independently constructed instances rather than quantities derived from parameters fitted inside the paper or reduced by definitional loops. No self-citation chains, uniqueness theorems, or ansatzes are invoked to justify the central claims, and the ontology serves as an explicit methodological choice for case generation rather than a self-referential input that forces the outcomes.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper contributes a new evaluation framework rather than a mathematical derivation; it rests on the constructed task ontology and rubric, which are domain-specific design choices without independent empirical validation outside this work.

axioms (1)
  • domain assumption Scientific problems in computational domains frequently begin as ill-posed requests that require multi-turn dialogue to become well-specified.
    This premise is stated directly in the abstract as the motivation for the benchmark.

pith-pipeline@v0.9.0 · 5858 in / 1325 out tokens · 51353 ms · 2026-05-20T10:31:46.481904+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

119 extracted references · 119 canonical work pages · 17 internal anchors

  1. [1]

    Bruce Croft

    Mohammad Aliannejadi, Hamed Zamani, Fabio Crestani, and W. Bruce Croft. Asking clarifying questions in open-domain information-seeking conversations. InProceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 475–484. ACM, 2019. doi: 10.1145/3331184.3331265

    work page doi:10.1145/3331184.3331265 2019

  2. [2]

    Analysing mixed initiatives and search strategies during conversational search

    Mohammad Aliannejadi, Julia Kiseleva, Aleksandr Chuklin, Jeff Dalton, and Mikhail Burtsev. Analysing mixed initiatives and search strategies during conversational search. InProceedings of the 29th ACM International Conference on Information & Knowledge Management. ACM, 2020. doi: 10.1145/3459637. 3482231. Also: ConvAI3 / ClariQ shared task at EMNLP 2020 workshop

    work page doi:10.1145/3459637 2020

  3. [3]

    Claude sonnet 4.6 system card

    Anthropic. Claude sonnet 4.6 system card. https://www.anthropic.com/ claude-sonnet-4-6-system-card, February 2026. System card, February 17, 2026

    work page 2026

  4. [4]

    Argyle, Ethan C

    Lisa P. Argyle, Ethan C. Busby, Nancy Fulda, Joshua R. Gubler, Christopher Rytting, and David Wingate. Out of one, many: Using language models to simulate human samples.Political Analysis, 31(3):337–351,

    work page

  5. [5]

    doi: 10.1017/pan.2023.2

    work page doi:10.1017/pan.2023.2 2023

  6. [6]

    Fluid intelligence: A forward look on ai foundation models in computational fluid dynamics, 2025

    Neil Ashton, Johannes Brandstetter, and Siddhartha Mishra. Fluid intelligence: A forward look on ai foundation models in computational fluid dynamics, 2025. URL https://arxiv.org/abs/2511. 20455. 10

    work page 2025

  7. [7]

    Askeland, Benjamin Wheatley, and Wendelin J

    Donald R. Askeland, Benjamin Wheatley, and Wendelin J. Wright.The Science and Engineering of Materials. Cengage, 8 edition, 2025

    work page 2025

  8. [8]

    MT-Bench-101: A fine-grained benchmark for evaluating large language models in multi-turn dialogues

    Ge Bai, Jie Liu, Xingyuan Bu, Yancheng He, Jiaheng Liu, Zhanhui Zhou, Zhuoran Lin, Wenbo Su, Tiezheng Ge, Bo Zheng, and Wanli Ouyang. MT-Bench-101: A fine-grained benchmark for evaluating large language models in multi-turn dialogues. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7...

    work page doi:10.18653/v1/2024.acl-long.401 2024

  9. [9]

    $\tau^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment

    Victor Barrès, Nicolai Dorka, Uros Damnjanovic, Alon Perelstein, Michael Huang, Michael Kuhmuench, Victor Chevrier, Abraham Park, Roger Schraner, Karthik Nair, Sidd Nair, Akash Garg, Drew Lingen- felter, Ashwin Frett, Ramesh Shanmugam, Clay Davey, Rob Subramaniam, Douglas Burdick, Caitlin Dwyer, et al. τ 2-bench: Evaluating conversational agents in a dual...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2506.07982 2025

  10. [10]

    Ferdinand P. Beer, E. Russell Johnston, John T. DeWolf, and David F. Mazurek.Mechanics of Materials. McGraw-Hill Education, 8 edition, 2020

    work page 2020

  11. [11]

    Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D

    Andres M. Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D. White, and Philippe Schwaller. ChemCrow: Augmenting large-language models with chemistry tools.Nature Machine Intelligence, 6: 525–535, 2024. doi: 10.1038/s42256-024-00832-8

    work page doi:10.1038/s42256-024-00832-8 2024

  12. [12]

    MultiWOZ—a large-scale multi-domain wizard-of-oz dataset for task-oriented dialogue modelling

    Paweł Budzianowski, Tsung-Hsien Wen, Bo-Hsiang Tseng, Iñigo Casanueva, Stefan Ultes, Osman Ra- madan, and Milica Gaši´c. MultiWOZ—a large-scale multi-domain wizard-of-oz dataset for task-oriented dialogue modelling. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Pro- cessing, pages 5016–5026. Association for Computational Li...

    work page doi:10.18653/v1/d18-1547 2018

  13. [13]

    MetaOpenFOAM: An LLM-based multi-agent framework for CFD.arXiv preprint arXiv:2407.21320, 2024

    Yuxuan Chen, Xu Zhu, Hua Zhou, and Zhuyin Ren. MetaOpenFOAM: An LLM-based multi-agent framework for CFD.arXiv preprint arXiv:2407.21320, 2024. doi: 10.48550/arxiv.2407.21320. URL https://arxiv.org/abs/2407.21320

    work page doi:10.48550/arxiv.2407.21320 2024

  14. [14]

    arXiv preprint arXiv:2410.05080 , year=

    Ziru Chen, Shijie Chen, Yuting Ning, Qianheng Zhang, Boshi Wang, Botao Yu, Yifei Li, Zeyi Liao, Chen Wei, Ziru Lu, Vishal Arber, Anthony Gitter, Liang Dong, and Heng Ji. ScienceAgentBench: Toward rigorous assessment of language agents for data-driven scientific discovery. InInternational Conference on Learning Representations, 2025. doi: 10.48550/arxiv.24...

    work page doi:10.48550/arxiv.2410.05080 2025

  15. [15]

    Jordan, Joseph E

    Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Hao Zhang, Banghua Zhu, Michael I. Jordan, Joseph E. Gonzalez, and Ion Stoica. Chatbot arena: An open platform for evaluating LLMs by human preference. InInternational Conference on Machine Learning,

    work page

  16. [16]

    Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference

    doi: 10.48550/arxiv.2403.04132. URLhttps://arxiv.org/abs/2403.04132

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2403.04132

  17. [17]

    User simulation with large language models for evaluating task-oriented dialogue.arXiv preprint arXiv:2309.13233, 2023

    Sam Davidson, Salvatore Hwang, Danbi Lee, Justin Cherian, Minhwa Lee, and Zhou Li. User simulation with large language models for evaluating task-oriented dialogue.arXiv preprint arXiv:2309.13233, 2023. doi: 10.48550/arxiv.2309.13233. URLhttps://arxiv.org/abs/2309.13233

    work page doi:10.48550/arxiv.2309.13233 2023

  18. [18]

    Srinivasan, Mahmoud Golestanian, Yuan Tian, Tianyi Zhang, P

    Rushikesh Deotale, A. Srinivasan, Mahmoud Golestanian, Yuan Tian, Tianyi Zhang, P. Vlachos, and Hector Gomez. ALL-FEM: Agentic LLMs fine-tuned for finite element methods.Computer Methods in Applied Mechanics and Engineering, 2026. doi: 10.1016/j.cma.2026.118985

    work page doi:10.1016/j.cma.2026.118985 2026

  19. [19]

    Primack, Summer Yue, and Chen Xing

    Kaustubh Deshpande, Ved Sirdeshmukh, Johannes Baptist Mols, Lifeng Jin, Ed-Yeremai Hernandez- Cardona, Dean Lee, Jeremy Kritz, Willow E. Primack, Summer Yue, and Chen Xing. Multichallenge: A realistic multi-turn conversation evaluation benchmark challenging to frontier llms. InFindings of the Association for Computational Linguistics: ACL 2025, pages 1863...

    work page doi:10.18653/v1/2025.findings-acl.958 2025

  20. [20]

    CalculiX: A three-dimensional structural finite element program, 1998

    Guido Dhondt and Klaus Wittig. CalculiX: A three-dimensional structural finite element program, 1998. URLhttps://www.calculix.de/. Software, accessed 2026-04-12

    work page 1998

  21. [21]

    Fine-tuning a large language model for automating computational fluid dynamics simulations.Theoretical and Applied Mechanics Letters, 2025

    Zhehao Dong, Zhen Lu, and Yue Yang. Fine-tuning a large language model for automating computational fluid dynamics simulations.Theoretical and Applied Mechanics Letters, 2025. doi: 10.1016/j.taml.2025. 100594. URLhttps://arxiv.org/abs/2504.09602. 11

    work page doi:10.1016/j.taml.2025 2025

  22. [22]

    Yao Dou, Michel Galley, Baolin Peng, Chris Kedzie, Weixin Cai, Alan Ritter, Chris Quirk, Wei Xu, and Jianfeng Gao. Simulatorarena: Are user simulators reliable proxies for multi-turn evaluation of AI assistants? InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 35212–35290. Association for Computational Lingui...

    work page doi:10.18653/v1/2025.emnlp-main 2025

  23. [23]

    URLhttps://aclanthology.org/2025.emnlp-main.1786/

    work page 2025

  24. [25]

    Fu, Freda Shi, Kinjal Basu, Raghuveer Lagudu, Aditya Saxena, Aditya Grover, Can Bollücke, Noah A

    Belinda Z. Fu, Freda Shi, Kinjal Basu, Raghuveer Lagudu, Aditya Saxena, Aditya Grover, Can Bollücke, Noah A. Smith, and Amit Dhurandhar. QuestBench: Evaluating information-gathering abilities of large language models. InInternational Conference on Learning Representations (ICLR), 2025. URL https://openreview.net/forum?id=BwGeIhGPgn

    work page 2025

  25. [26]

    doi:10.48550/arXiv.2409.06097 , abstract =

    Yujian Gan, Changling Zhang, Jinxia Fu, and Matthew Purver. ClarQ-LLM: A benchmark for models clarifying and requesting information in task-oriented dialog.arXiv preprint arXiv:2409.06097, 2024. doi: 10.48550/arxiv.2409.06097. URLhttps://arxiv.org/abs/2409.06097

    work page doi:10.48550/arxiv.2409.06097 2024

  26. [27]

    Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities

    Gemini Team, Google DeepMind. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities.arXiv preprint arXiv:2507.06261, 2025. URL https://arxiv.org/abs/2507.06261

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [28]

    Gerhart, John I

    Andrew L. Gerhart, John I. Hochstein, and Philip M. Gerhart.Munson, Young and Okiishi’s Fundamentals of Fluid Mechanics. Wiley, 9 edition, 2020

    work page 2020

  28. [29]

    Goodno and James M

    Barry J. Goodno and James M. Gere.Mechanics of Materials. Cengage, 9 edition, 2018

    work page 2018

  29. [30]

    LLM-RUBRIC: A multidimensional, calibrated approach to automated evaluation of natural language texts

    Helia Hashemi, Jason Eisner, Corby Rosset, Benjamin Van Durme, and Chris Kedzie. LLM-RUBRIC: A multidimensional, calibrated approach to automated evaluation of natural language texts. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL), 2024. doi: 10.18653/ v1/2024.acl-long.745. URLhttps://aclanthology.org/2024....

    work page 2024

  30. [31]

    MirrorBench: A Benchmark to Evaluate Conversational User-Proxy Agents for Human-Likeness

    Ashutosh Hathidara, Julien Yu, Vaishali Senthil, Sebastian Schreiber, and Anil Babu Ankisettipalli. MirrorBench: A benchmark to evaluate conversational user-proxy agents for human-likeness.arXiv preprint arXiv:2601.08118, 2026. doi: 10.48550/arxiv.2601.08118. URL https://arxiv.org/abs/ 2601.08118

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2601.08118 2026

  31. [32]

    AutoFEA: Enhancing AI copilot by integrating finite element analysis using large language models with graph neural networks

    Shifu Hou, Rick Johnson, Ramandeep Makhija, Lingwei Chen, and Yanfang Ye. AutoFEA: Enhancing AI copilot by integrating finite element analysis using large language models with graph neural networks. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 24078–24085, 2025. doi: 10.1609/AAAI.V39I22.34582. URL https://ojs.aaai.org/...

    work page doi:10.1609/aaai.v39i22.34582 2025

  32. [33]

    Teaching language models to gather information proactively

    Tenghao Huang, Sihao Chen, Muhao Chen, Jonathan May, Longqi Yang, Mengting Wan, and Pei Zhou. Teaching language models to gather information proactively. InFindings of the Association for Com- putational Linguistics: EMNLP 2025, pages 15588–15599. Association for Computational Linguis- tics, 2025. doi: 10.18653/v1/2025.findings-emnlp.843. URL https://acla...

    work page doi:10.18653/v1/2025.findings-emnlp.843 2025

  33. [34]

    SWE-bench: Can Language Models Resolve Real-World GitHub Issues?

    Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. SWE-bench: Can language models resolve real-world GitHub issues? InInternational Conference on Learning Representations, 2024. doi: 10.48550/arxiv.2310.06770. URL https:// openreview.net/forum?id=VTF8yNQM66

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.06770 2024

  34. [35]

    Aligning language models to explicitly handle ambiguity

    Hyuhng Joon Kim, Youna Kim, Cheonbok Park, Junyeob Kim, Choonghyun Park, Kang Min Yoo, Sang- goo Lee, and Taeuk Kim. Aligning language models to explicitly handle ambiguity. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2024. doi: 10.48550/arXiv.2404.11972

    work page doi:10.48550/arxiv.2404.11972 2024

  35. [36]

    Clam: Selective clarification for ambiguous questions with generative language models

    Lorenz Kuhn, Yarin Gal, and Sebastian Farquhar. Clam: Selective clarification for ambiguous questions with generative language models. InProceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2023

    work page 2023

  36. [37]

    Vaibhav Kumar and Alan W. Black. Clarq: A large-scale and diverse dataset for clarification question generation. InProceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7296–7301. Association for Computational Linguistics, 2020. doi: 10.18653/v1/2020.acl-main.651. 12

    work page doi:10.18653/v1/2020.acl-main.651 2020

  37. [38]

    MT-Eval: A multi-turn capabilities evaluation benchmark for large language models

    Wai-Chung Kwan, Xingshan Zeng, Yufei Wang, Yusen Sun, Liangyou Li, Lifeng Shang, Qun Liu, and Kam-Fai Wong. MT-Eval: A multi-turn capabilities evaluation benchmark for large language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2024. doi: 10.48550/arxiv.2401.16745

    work page doi:10.48550/arxiv.2401.16745 2024

  38. [39]

    LLMs Get Lost In Multi-Turn Conversation

    Philippe Laban, Hiroaki Hayashi, Yingbo Zhou, and Jennifer Neville. Llms get lost in multi-turn conversa- tion. InInternational Conference on Learning Representations, 2026. doi: 10.48550/arXiv.2505.06120. URLhttps://openreview.net/forum?id=VKGTGGcwl6

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.06120 2026

  39. [40]

    Asking clarification questions to handle ambiguity in open-domain qa

    Dongryeol Lee, Segwang Kim, Minwoo Lee, Hwanhee Lee, Joonsuk Park, Sang-Woo Lee, and Kyomin Jung. Asking clarification questions to handle ambiguity in open-domain qa. InFindings of the Association for Computational Linguistics: EMNLP 2023, pages 11526–11544. Association for Computational Lin- guistics, 2023. doi: 10.18653/v1/2023.findings-emnlp.772. URL ...

    work page doi:10.18653/v1/2023.findings-emnlp.772 2023

  40. [41]

    CONTRADOC: Understanding self-contradictions in documents with large language models

    Jierui Li, Vipul Raheja, and Dhruv Kumar. CONTRADOC: Understanding self-contradictions in documents with large language models. InProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics. Association for Computational Linguistics, 2024. doi: 10.48550/arXiv.2311.09182

    work page doi:10.48550/arxiv.2311.09182 2024

  41. [42]

    From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline

    Tianle Li, Wei-Lin Chiang, Evan Frick, Lisa Dunlap, Tianhao Wu, Banghua Zhu, Joseph E. Gonzalez, and Ion Stoica. From crowdsourced data to high-quality benchmarks: Arena-hard and benchbuilder pipeline. InInternational Conference on Machine Learning, 2024. doi: 10.48550/arxiv.2406.11939. URL https://arxiv.org/abs/2406.11939

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2406.11939 2024

  42. [43]

    Zongxi Li, Yang Li, Haoran Xie, and S. Joe Qin. Condambigqa: A benchmark and dataset for conditional ambiguous question answering. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2025. doi: 10.18653/v1/2025.emnlp-main.115. URLhttps://aclanthology.org/2025.emnlp-main.115/

    work page doi:10.18653/v1/2025.emnlp-main.115 2025

  43. [44]

    Srolovitz, and Tongqi Wen

    Siyu Liu, Jiamin Xu, Beilin Ye, Bo Hu, David J. Srolovitz, and Tongqi Wen. Mattools: Benchmarking large language models for materials science tools.arXiv preprint arXiv:2505.10852, 2025. doi: 10.48550/ arxiv.2505.10852. URLhttps://arxiv.org/abs/2505.10852

    work page arXiv 2025

  44. [45]

    AgentBench: Evaluating LLMs as Agents

    Xiao Liu, Hao Yu, Hanchen Zhang, Yifan Xu, Xuanyu Lei, Hanyu Lai, Yu Gu, Hangliang Ding, Kaiwen Men, Kejuan Yang, Shudan Zhang, Xiang Deng, Aohan Zeng, Zhengxiao Du, Chenhui Zhang, Sheng Shen, Tianjun Zhang, Yu Su, Huan Sun, Minlie Huang, Yuxiao Dong, and Jie Tang. AgentBench: Evaluating LLMs as agents. InInternational Conference on Learning Representatio...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2308.03688 2024

  45. [46]

    Wells, editors.Automated Solution of Differential Equations by the Finite Element Method: The FEniCS Book, volume 84 ofLecture Notes in Computational Science and Engineering

    Anders Logg, Kent-Andre Mardal, and Garth N. Wells, editors.Automated Solution of Differential Equations by the Finite Element Method: The FEniCS Book, volume 84 ofLecture Notes in Computational Science and Engineering. Springer, 2012. doi: 10.1007/978-3-642-23099-8

    work page doi:10.1007/978-3-642-23099-8 2012

  46. [47]

    SciAgent: Tool-augmented language models for scientific reasoning.arXiv preprint arXiv:2402.11451,

    Yubo Ma, Zhibin Gou, Junheng Hao, Ruochen Xu, Shuohang Wang, Liangming Pan, Yujiu Yang, Yixin Cao, Aixin Sun, Hany Awadalla, and Weizhu Chen. SciAgent: Tool-augmented language models for scientific reasoning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 2024. doi: 10.4...

    work page doi:10.48550/arxiv.2402.11451 2024

  47. [48]

    GAIA: a benchmark for General AI Assistants

    Grégoire Mialon, Clémentine Fourrier, Craig Swift, Thomas Wolf, Yann LeCun, and Thomas Scialom. GAIA: A benchmark for general AI assistants.arXiv preprint arXiv:2311.12983, 2024. doi: 10.48550/ arxiv.2311.12983. URLhttps://arxiv.org/abs/2311.12983

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  48. [49]

    , author Michael, J

    Sewon Min, Julian Michael, Hannaneh Hajishirzi, and Luke Zettlemoyer. AmbigQA: Answering am- biguous open-domain questions. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 5783–5797. Association for Computational Linguistics, 2020. doi: 10.18653/v1/2020.emnlp-main.466

    work page doi:10.18653/v1/2020.emnlp-main.466 2020

  49. [50]

    Brenner, and Peter Norgaard

    Nayantara Mudur, Hao Cui, Subhashini Venugopalan, Paul Raccuglia, Michael P. Brenner, and Peter Norgaard. FEABench: Evaluating language models on multiphysics reasoning ability.arXiv preprint,

    work page

  50. [51]

    URL https://arxiv.org/abs/2504.06260v1

    doi: 10.48550/arxiv.2504.06260. URL https://arxiv.org/abs/2504.06260v1. Presented at NeurIPS 2024 workshops

    work page doi:10.48550/arxiv.2504.06260 2024

  51. [52]

    Bo Ni and Markus J. Buehler. MechAgents: Large language model multi-agent collaborations can solve mechanics problems.Extreme Mechanics Letters, 2024. doi: 10.48550/arxiv.2311.08166. 13

    work page doi:10.48550/arxiv.2311.08166 2024

  52. [53]

    A Survey on LLM-based Conversational User Simulation

    Bo Ni, Yu Wang, Leyao Wang, Branislav Kveton, Franck Dernoncourt, et al. A survey on LLM-based conversational user simulation. InProceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics. Association for Computational Linguistics, 2026. doi: 10.18653/v1/2026.eacl-long.200. URLhttps://arxiv.org/abs/2604.24977

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.18653/v1/2026.eacl-long.200 2026

  53. [54]

    Update to gpt-5 system card: Gpt-5.2

    OpenAI. Update to gpt-5 system card: Gpt-5.2. https://openai.com/index/ gpt-5-system-card-update-gpt-5-2/ , December 2025. System card update, December 11, 2025

    work page 2025

  54. [55]

    gpt-oss-120b & gpt-oss-20b Model Card

    OpenAI. gpt-oss-120b & gpt-oss-20b model card.arXiv preprint arXiv:2508.10925, 2025. URL https://arxiv.org/abs/2508.10925

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  55. [56]

    Openfoamgpt: A retrieval-augmented large language model (llm) agent for openfoam-based computational fluid dynamics.Physics of Fluids, 37(3), 2025

    Sandeep Pandey, Ran Xu, Wenkang Wang, and Xu Chu. Openfoamgpt: A retrieval-augmented large language model (llm) agent for openfoam-based computational fluid dynamics.Physics of Fluids, 37(3), 2025

    work page 2025

  56. [57]

    Interpretation of natural language rules in conversational machine reading

    Marzieh Saeidi, Max Bartolo, Patrick Lewis, Sameer Singh, Tim Rocktäschel, Mike Sheldon, Guillaume Bouchard, and Sebastian Riedel. Interpretation of natural language rules in conversational machine reading. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 2166–2176. Association for Computational Linguistics, ...

    work page doi:10.18653/v1/d18-1233 2018

  57. [58]

    Reliable LLM-based user simulator for task-oriented dialogue systems.arXiv preprint arXiv:2402.13374, 2024

    Ivan Sekulic, Silvia Terragni, Victor Guimarães, Nghia Khau, Bruna Guedes, Modestas Filipavicius, André Ferreira Manso, and Roland Mathis. Reliable LLM-based user simulator for task-oriented dialogue systems.arXiv preprint arXiv:2402.13374, 2024. doi: 10.48550/arxiv.2402.13374. URL https://arxiv. org/abs/2402.13374

    work page doi:10.48550/arxiv.2402.13374 2024

  58. [59]

    Shackelford.Introduction to Materials Science for Engineers

    James F. Shackelford.Introduction to Materials Science for Engineers. Pearson, 9 edition, 2021

    work page 2021

  59. [60]

    Non-collaborative user simulators for tool agents

    Jeonghoon Shim, Woojung Song, Cheyon Jin, Seungwon Kook, and Yohan Jo. Non-collaborative user simulators for tool agents. InInternational Conference on Learning Representations, 2026. doi: 10.48550/ arxiv.2509.23124. URLhttps://openreview.net/forum?id=UAUimofy3W

    work page arXiv 2026

  60. [61]

    CFDLLMBench: A Benchmark Suite for Evaluating Large Language Models in Computational Fluid Dynamics

    Nithin Somasekharan, Ling Yue, Yadi Cao, Weichao Li, Patrick Emami, Pochinapeddi Sai Bhargav, Anurag Acharya, Xingyu Xie, and Shaowu Pan. CFDLLMBench: A benchmark suite for evaluating large language models in computational fluid dynamics.arXiv preprint arXiv:2509.20374, 2025. doi: 10.48550/arXiv.2509.20374. URLhttps://arxiv.org/abs/2509.20374

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2509.20374 2025

  61. [62]

    Cfdllmbench: A benchmark suite for evaluating large language models in computational fluid dynamics.Journal of Data-centric Machine Learning Research, 13:1–40, 2026

    Nithin Somasekharan, Ling Yue, Yadi Cao, Weichao Li, Patrick Emami, Pochinapeddi Sai Bhargav, Anurag Acharya, Xingyu Xie, and Shaowu Pan. Cfdllmbench: A benchmark suite for evaluating large language models in computational fluid dynamics.Journal of Data-centric Machine Learning Research, 13:1–40, 2026

    work page 2026

  62. [63]

    SciEval: A multi-level large language model evaluation benchmark for scientific research

    Liangtai Sun, Yang Han, Zihan Zhao, Da Ma, Zhennan Shen, Baocai Chen, Lu Chen, and Kai Yu. SciEval: A multi-level large language model evaluation benchmark for scientific research. InProceedings of the AAAI Conference on Artificial Intelligence, 2024. doi: 10.48550/arxiv.2308.13149. URL https: //ojs.aaai.org/index.php/AAAI/article/view/29872

    work page doi:10.48550/arxiv.2308.13149 2024

  63. [64]

    Scicode: A research coding benchmark curated by scientists

    Minyang Tian, Luyu Gao, Shizhuo Dylan Zhang, Xinan Chen, Cunwei Fan, Xuefei Guo, Roland Haas, Pan Ji, Kittithat Krongchon, Yao Li, Shengyan Liu, Di Luo, Yutao Ma, Hao Tong, Kha Trinh, Chenyu Tian, Zihan Wang, Bohao Wu, Yanyu Xiong, Shengzhu Yin, Minhui Zhu, Kilian Lieret, Yanxin Lu, Genglin Liu, Yufeng Du, Tianhua Tao, Ofir Press, Jamie Callan, Eliu Huert...

    work page doi:10.48550/arxiv.2407.13168 2024

  64. [65]

    Axelrod, R

    Gladys Tyen, Hassan Mansoor, Victor Carbune, Peter Chen, and Tony Mak. LLMs cannot find reasoning errors, but can correct them given the error location. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar, editors,Findings of the Association for Computational Linguistics: ACL 2024, pages 13894–13908, Bangkok, Thailand, August 2024. Association for Computatio...

    work page doi:10.18653/v1/2024 2024

  65. [66]

    Ugural and Saul K

    Ansel C. Ugural and Saul K. Fenster.Advanced Mechanics of Materials and Applied Elasticity. Pearson, 6 edition, 2021

    work page 2021

  66. [67]

    SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models

    Xiaoxuan Wang, Ziniu Hu, Pan Lu, Yanqiao Zhu, Jieyu Zhang, Satyen Subramaniam, Arjun Rajan Loomba, Shichang Zhang, Yizhou Sun, and Wei Wang. SciBench: Evaluating college-level scientific problem- solving abilities of large language models. InProceedings of the 41st International Conference on Machine Learning (ICML), 2024. doi: 10.48550/arxiv.2307.10635. ...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2307.10635 2024

  67. [68]

    MINT: Evaluating LLMs in multi-turn interaction with tools and language feedback.arXiv preprint arXiv:2309.10691, 2023

    Xingyao Wang, Zihan Wang, Jiateng Liu, Yangyi Chen, Lifan Yuan, Hao Peng, and Heng Ji. MINT: Evaluating LLMs in multi-turn interaction with tools and language feedback. InInternational Conference on Learning Representations, 2024. doi: 10.48550/arxiv.2309.10691. URL https://openreview.net/ forum?id=jp3gWrMuIZ

    work page doi:10.48550/arxiv.2309.10691 2024

  68. [69]

    ProfBench: Multi-Domain Rubrics requiring Professional Knowledge to Answer and Judge

    Zhilin Wang, Jaehun Jung, Ximing Lu, Shizhe Diao, Ellie Evans, Jiaqi Zeng, Pavlo Molchanov, Yejin Choi, Jan Kautz, and Yi Dong. ProfBench: Multi-domain rubrics requiring professional knowledge to answer and judge.arXiv preprint arXiv:2510.18941, 2025. doi: 10.48550/arxiv.2510.18941. URL https://arxiv.org/abs/2510.18941

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2510.18941 2025

  69. [70]

    White.Fluid Mechanics

    Frank M. White.Fluid Mechanics. McGraw-Hill Education, 9 edition, 2021

    work page 2021

  70. [71]

    William D

    Jr. William D. Callister and David G. Rethwisch.Materials Science and Engineering: An Introduction. Wiley, 10 edition, 2018

    work page 2018

  71. [72]

    Rmtbench: Benchmarking llms through multi-turn user-centric role-playing

    Hao Xiang, Tianyi Tang, Yang Su, Bowen Yu, An Yang, Fei Huang, Yichang Zhang, Yaojie Lu, Hongyu Lin, Xianpei Han, Jingren Zhou, Junyang Lin, and Le Sun. Rmtbench: Benchmarking llms through multi-turn user-centric role-playing. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025. doi: 10.48550/arxiv.2507.20352. UR...

    work page doi:10.48550/arxiv.2507.20352 2025

  72. [73]

    Narasimhan

    Shunyu Yao, Noah Shinn, Pedram Razavi, and Karthik R. Narasimhan. τ-bench: A benchmark for tool- agent-user interaction in real-world domains. InInternational Conference on Learning Representations,

    work page

  73. [74]

    $\tau$-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains

    doi: 10.48550/arxiv.2406.12045. URLhttps://openreview.net/forum?id=roNSXZpUDN

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2406.12045

  74. [75]

    Foam-agent: A multi-agent framework for automating openfoam-based cfd simulation

    Ling Yue, Nithin Somasekharan, Yadi Cao, and Shaowu Pan. Foam-agent: A multi-agent framework for automating openfoam-based cfd simulation. InNeurIPS 2025 Workshop ML4PS, 2025

    work page 2025

  75. [76]

    Mohd Zaki, Jayadeva, Mausam, and N. M. Anoop Krishnan. MaScQA: Investigating materials science knowledge of large language models.Digital Discovery, 3(2):313–327, 2024. doi: 10.1039/D3DD00188A. URLhttps://doi.org/10.1039/D3DD00188A

    work page doi:10.1039/d3dd00188a 2024

  76. [77]

    HoneyComb: A flexible LLM-based agent system for materials science

    Huan Zhang, Yu Song, Ziyu Hou, Santiago Miret, and Bang Liu. HoneyComb: A flexible LLM-based agent system for materials science. InFindings of the Association for Computational Linguistics: EMNLP

    work page

  77. [78]

    doi: 10.48550/arxiv.2409.00135

    Association for Computational Linguistics, 2024. doi: 10.48550/arxiv.2409.00135. URL https: //arxiv.org/abs/2409.00135v1

    work page doi:10.48550/arxiv.2409.00135 2024

  78. [79]

    MatSciBench: Benchmarking the reasoning ability of large language models in materials science.arXiv preprint arXiv:2510.12171, 2025

    Junkai Zhang, Jingru Gan, Xiaoxuan Wang, Zian Jia, Changquan Gu, Jianpeng Chen, Yanqiao Zhu, Mingyu Derek Ma, Dawei Zhou, Ling Li, and Wei Wang. MatSciBench: Benchmarking the reasoning ability of large language models in materials science.arXiv preprint arXiv:2510.12171, 2025. doi: 10.48550/arXiv.2510.12171. URLhttps://arxiv.org/abs/2510.12171

    work page doi:10.48550/arxiv.2510.12171 2025

  79. [80]

    Zhang, W

    Michael J.Q. Zhang, W. Bradley Knox, and Eunsol Choi. Modeling future conversation turns to teach LLMs to ask clarifying questions. InInternational Conference on Learning Representations, 2025. doi: 10.48550/arXiv.2410.13788. URLhttps://openreview.net/forum?id=futureCQs

    work page doi:10.48550/arxiv.2410.13788 2025

  80. [81]

    CLAMBER: A benchmark of identifying and clarifying ambiguous information needs in large language models

    Tong Zhang, Peixin Qin, Yang Deng, Chen Huang, Wenqiang Lei, Junhong Liu, Dingnan Jin, Hongru Liang, and Tat-Seng Chua. CLAMBER: A benchmark of identifying and clarifying ambiguous information needs in large language models. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 10746–10766...

    work page doi:10.18653/v1/2024.acl-long.578 2024

Showing first 80 references.