Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Conversational Interfaces for Parametric Conceptual Architectural Design: Integrating Mixed Reality with LLM-driven Interaction

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read By combining speech, gestures, and a three-agent LLM pipeline, this paper claims that mixed-reality environments can let architects without programming skills generate and iterate parametric models conversationally.

desk verdict A genuinely novel integration of multi-agent LLMs with MR parametric modeling, but the evaluation is too weak to support the barrier-reduction claim; worth refereeing with major revisions. read the letter →

arxiv 2506.06066 v1 pith:6YJDSFKX submitted 2025-06-06 cs.HC

Ruochen Ji , Lyu Tiangang This is my paper
classification cs.HC
keywords mixedrealityparametricdesignconversationalinterfacelargelanguagemodelmulti-agentsystemarchitecturalgesturerecognitioncodegeneration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Parametric architectural modeling usually demands programming skills and desktop tools with visual-programming interfaces, which shuts out designers who think programmatically but do not code. This paper tries to close that gap by building a mixed-reality interface where a designer talks through a design intent, sees prompts and models in an immersive headset, and adjusts parameters by gesture. The system uses three specialized large-language-model agents—a Reasoning Agent that conducts the dialogue, a Coding Agent that writes C# code, and an Optimization Agent that checks the code for syntax and dependency errors—so that natural language ends up compiled into executable parametric models. The authors argue that this 'reasoning-to-code-generation' loop reduces the cognitive and operational barriers of parametric design and transforms mixed reality from a visualization medium into a generative design tool.

What carries the argument

The load-bearing mechanism is the 'reasoning-to-code-generation' pipeline, a multi-agent arrangement of three large-language-model roles: a Reasoning Agent that converses with the user and collects parameters until the design intent is complete, a Coding Agent that turns the final specification into C# code blocks, and an Optimization Agent that statically validates the code against a registry of precompiled methods before execution. Around this pipeline sits a two-device compiler architecture: a mixed-reality headset renders the scene and accepts speech and gesture input, while a networked desktop machine compiles any new methods into DLLs because the headset cannot compile code at runtime. Strict JSON schemas and XML-tagged prompts form the protocol that lets the agents' structured responses be parsed programmatically and routed to the interface controls.

What would settle it

A controlled experiment that feeds the same set of design prompts to the system with and without the Optimization Agent under identical task and environment conditions, measuring per-prompt compilation success and semantic fidelity to the user's stated intent; if the success gap disappears or overall success stays near 72%, the reliability claim collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that its multi-LLM conversational interface lets users generate parametric models in mixed reality through conversation and gestures, thereby reducing cognitive and operational barriers and allowing designers without programming backgrounds to complete parametric design tasks. The paper reports that in its user experiments the full pipeline compiled successfully for 31 of 43 generated code blocks (72.0%), while removing the Optimization Agent dropped success to 9 of 35 (25.7%) in a separate test condition. It also observes a trade-off between accuracy and creativity in the prompts given to the coding agent: stricter instructions produce stable, repeatable code, whereas looser instructions generate more varied and inventive approaches with more syntax or semantic errors. The paper presents the system as a proof of concept that mixed reality can host an embodied, conversational parametric design workflow, with user feedback indicating better understanding of design logic and increased creative ideation.

Load-bearing premise

The system only helps if the conversational pipeline can turn ambiguous spoken design ideas into correct, compilable parametric code, which the paper supports with a 72% compilation success rate but does not fully establish.

Editorial extensions

If this is right

  • Architects and students without programming experience could generate and iterate parametric models through dialogue and gestures, making parametric exploration accessible at the conceptual stage.
  • The presence of the Optimization Agent, which checks code against precompiled method DLLs, is asserted to be the difference between a usable success rate of 72.0% and a near-unusable 25.7% in the paper's conditions.
  • The accuracy-versus-creativity trade-off in prompt design suggests that how tightly a design task is specified determines whether code generation is reliable or explorative, a knob future interfaces could expose to users.
  • Because the system separates reusable methods from task logic and compiles methods into DLLs, the architecture points toward an extensible library of parametric operations that conversational interfaces could grow over time.
  • Within mixed reality, the system frames parametric modeling as embodied exploration, letting users compare design variations in situ by adjusting parameters rather than returning to a desktop application.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's comparison of the 72.0% and 25.7% compilation rates is not a clean controlled experiment: the two runs differ in tasks and environment, so the drop may reflect task difficulty changes rather than the Optimization Agent alone; a matched comparison would be needed to isolate its effect.
  • Even when generated code fails to compile, the conversational Reasoning Agent may still deliver value by helping users articulate their design logic, meaning the system's reported benefits might come largely from dialogue rather than from working code.
  • The 72% compilation ceiling implies that roughly one in four code generations fails, so for the interface to leave the lab, error recovery, not just error avoidance, deserves to be a central design target.
  • The accuracy-creativity trade-off hints at a testable extension: exposing prompt strictness as a user-controlled dial would let designers choose between stable output and explorative variation during a session.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This paper presents a conversational mixed-reality interface for parametric architectural design, combining speech, gesture, and a three-agent LLM pipeline (Reasoning, Coding, Optimization) that translates user intent into compilable C# parametric code executed on a Meta Quest 3. The authors report a user study with 27 architecture-related participants and a separate code-generation experiment, and they argue that this approach reduces cognitive and operational barriers to parametric modeling and enables designers without programming backgrounds to explore design spaces. The central evidence is a compilation success rate of 72% (31/43) with the Optimization Agent versus 25.7% (9/35) without it, plus qualitative participant feedback.

Significance. If the central claim were rigorously supported, this would be a useful contribution to HCI and computational design, offering a concrete architecture for combining conversational LLM agents with MR-based parametric modeling. The paper is transparent about several limitations and provides a detailed account of the system modules, agent prompts, and compilation workflow. However, the current quantitative evidence does not establish the claimed barrier reduction: the compilation metric is not tied to task-level correctness, the ablation comparison is confounded, and the qualitative benefits are not attributed to the full pipeline. The topic is timely and the system is interesting, but the evaluation needs substantial strengthening before the conclusions can be accepted.

major comments (4)
  1. [Section 5.3.1 and 5.3.3] The reported 72% (31/43) compilation success rate is per generated code block, not per completed design task. For a task that requires k code generations, the chance that all blocks compile is 0.72^k, yet the paper does not report how many of the 27 user sessions ended with a correct, usable model. The conclusion in Section 5.3.3 that users 'were able to complete modeling tasks with relatively few compilation failures' therefore lacks task-level quantitative support. Please report per-task completion rates and explicitly define what constitutes task completion.
  2. [Section 5.1] Phase 1 outcomes were recorded 'irrespective of correctness,' and the independent expert evaluation mentioned in Phase 2 is never reported with results. A compiled model can implement different geometry from what the user requested, so compilation success alone does not measure semantic fidelity. The central claim that the system reliably maps design intentions to executable parametric code requires a semantic correctness check, such as expert ratings of whether the generated model matches the user's stated intent.
  3. [Section 5.3.1] The comparison between 72.0% (31/43) with the Optimization Agent and 25.7% (9/35) without it conflates the presence of the OA with changes in task set, environment, and LLM session. No confidence intervals or significance tests are provided. This comparison therefore cannot isolate the OA's contribution and does not establish the reliability of the full pipeline. A controlled ablation with matched tasks and repeated trials is needed.
  4. [Section 5.2.2 and 5.3.3] The qualitative benefits quoted in Section 5.2.2, such as 'RA guided me through the parametric design process,' may be produced by the Reasoning Agent's conversational scaffolding alone, independent of whether the Coding and Optimization Agents generate correct code. Because no condition separates conversational guidance from code generation, the paper's attribution of reduced barriers to the full multi-LLM system is not established. A control condition—for example, an RA-only mode with a stub code generator—would help test this attribution.
minor comments (5)
  1. [Section 5.1] The phrase 'Coding Agent and Output Agent workflows' appears to refer to the Coding and Optimization Agents; please align the terminology across the paper.
  2. [Figure 6] Figure 6 is referenced in Section 5.2.1 but its content is not described in the text; please add an explanatory caption or in-text description.
  3. [Section 5.3.2] The claimed 'accuracy versus creativity' trade-off is illustrated with a single example; consider reporting quantitative variability across multiple independent runs or prompt variants to support the claim.
  4. [Figure 7 and Section 5.2.2] The comparison with Grasshopper in Figure 7 is purely qualitative, with no task completion times, error rates, or workload measures; please soften claims about relative barriers or add a quantitative baseline.
  5. [Section 4.2.1] The citation [34] is given for 'flipped interaction prompting,' but the cited prompt-pattern catalog may not be the original source; please verify and cite the primary reference if appropriate.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the evaluation is empirical and the cited reasoning-to-code pipeline is external prior work.

full rationale

This paper does not contain a circular derivation. Its central claim—that a conversational MR interface with a multi-agent LLM pipeline lowers barriers to parametric design—is supported by an empirical evaluation: a 27-participant user study (Section 5.2) and a code-generation experiment reporting compilation success rates (Section 5.3.1). No parameter is fitted to the outcome it later predicts: the 72.0% compilation rate with the Optimization Agent and 25.7% without it are measured outcomes, not inputs. The 'reasoning-to-code-generation' pipeline is borrowed from prior work (reference [26], Code as Policies), which is external and not authored by the present paper's authors, so it is not a self-citation chain. No uniqueness theorem is invoked, no ansatz is smuggled in via citation, and no known result is renamed as a contribution. The main weakness is evidential—compilation success is not the same as semantic fidelity, and the with-/without-OA comparison confounds task set and environment—but that is a correctness or validity concern, not circularity. Accordingly, the circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper does not propose new theoretical constructs; it makes empirical claims about a system. The central assumptions are about the reliability of LLMs and the appropriateness of chosen tasks.

assumptions (3)
  • domain assumption LLMs can generate syntactically valid C# code for parametric shapes from natural language instructions.
    The whole pipeline depends on this; compilation success is the only proxy and it is only 72% in the main condition.
  • domain assumption Speech and gesture are adequate input modalities for parametric design in MR.
    The interface relies on these modalities; no validation that they are better than alternatives.
  • ad hoc to paper The two design tasks (skyscraper and planar lot generation) represent typical architectural conceptual design.
    Tasks were selected by researchers, not justified against a broader taxonomy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Conversational Interfaces for Parametric Conceptual Architectural Design: Integrating Mixed Reality with LLM-driven Interaction." pith.science (2026). https://pith.science/paper/6YJDSFKX

@misc{pith2026250606066,
  author       = {Pith},
  title        = {Pith review of: Conversational Interfaces for Parametric Conceptual Architectural Design: Integrating Mixed Reality with LLM-driven Interaction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6YJDSFKX}},
  note         = {Machine review of arXiv:2506.06066}
}
read the original abstract

Mixed reality (MR) environments offer embodied spatial interaction, providing intuitive 3D manipulation capabilities that enhance the conceptual design process. Parametric modeling, a powerful and advanced architectural design method, enables the generation of complex, optimized geometries. However, its integration into MR environments remains limited due to precision constraints and unsuitable input modalities. Existing MR tools prioritize spatial interaction but lack the control and expressiveness required for parametric workflows, particularly for designers without formal programming backgrounds. We address this gap by introducing a novel conversational MR interface that combines speech input, gesture recognition, and a multi-agent large language model (LLM) system to support intuitive parametric modeling. Our system dynamically manages parameter states, resolves ambiguous commands through conversation and contextual prompting, and enables real-time model manipulation within immersive environments. We demonstrate how this approach reduces cognitive and operational barriers in early-stage design tasks, allowing users to refine and explore their design space. This work expands the role of MR to a generative design platform, supporting programmatic thinking in design tasks through natural, embodied interaction.

Figures

Figures reproduced from arXiv: 2506.06066 by the authors.

Figure 1
Figure 1. Overview of system design objectives. This framework establishes a comprehensive conversational parametric design [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System Data Flow Fang & Kao et al. [17] evaluated differences in user experience be￾tween parametric CAD and free-form VR, concluding that VR en￾hances creative exploration while traditional CAD remains superior for precision-oriented tasks. Studies by Coppens et al. [15], Salim et al. [29], Alhazzaa et al. [3], and Buyruk [9] reveal the potential of immersive environments for intuitive spatial reasoning and ma￾nipu… view at source ↗
Figure 3
Figure 3. Overview of system modules and user interface [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Result of user study questionnaire [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Result of parametric design generation result [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Comparison between A: Our proposed UI, and B: WIMP [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 21 canonical work pages

  1. [1]

    https://www.grasshopper3d.com/. 1

    work page

  2. [2]

    M. Ahn, A. Brohan, N. Brown, Y . Chebotar, O. Cortes, B. David, C. Finn, C. Fu, K. Gopalakrishnan, K. Hausman, A. Herzog, D. Ho, J. Hsu, J. Ibarz, B. Ichter, A. Irpan, E. Jang, R. J. Ruano, K. Jeffrey, S. Jesmonth, N. J. Joshi, R. Julian, D. Kalashnikov, Y . Kuang, K.- H. Lee, S. Levine, Y . Lu, L. Luu, C. Parada, P. Pastor, J. Quiambao, K. Rao, J. Rettin...

    arXiv doi:10.48550/arxiv.2204.01691

  3. [3]

    Alhazzaa and W

    K. Alhazzaa and W. Yan. Integrating parametric modeling, BIM, and building performance analysis into augmented reality for architectural design and education. In Proceedings of the 2023 7th International Conference on Virtual and Augmented Reality Simulations , pp. 68–

    work page 2023

  4. [4]

    Alibay, M

    F. Alibay, M. Kavakli, J.-R. Chardonnet, and M. Z. Baig. The usabil- ity of speech and/or gestures in multi-modal interface systems. In Proceedings of the 9th International Conference on Computer and Automation Engineering , pp. 73–77. ACM. doi: 10.1145/3057039. 3057089 2

    work page doi:10.1145/3057039

  5. [5]

    Alrashedy, P

    K. Alrashedy, P. Tambwekar, Z. Zaidi, M. Langwasser, W. Xu, and M. Gombolay. Generating CAD code with vision-language models for 3d designs. Version Number: 1. doi: 10.48550/ARXIV.2410.05340 2

    arXiv doi:10.48550/arxiv.2410.05340

  6. [6]

    Baig and M

    M. Baig and M. Kavakli. Analyzing novice and expert user’s cognitive load in using a multi-modal interface system. doi: 10.1109/ICSENG. 2018.8638206 2

    work page arXiv 2018

  7. [7]

    Bhooshan

    S. Bhooshan. Parametric design thinking: A case-study of practice- embedded architectural research. 52:115–143. doi: 10.1016/j.destud. 2017.05.003 2

    doi:10.1016/j.destud 2017

  8. [8]

    put-that-there

    R. A. Bolt. “put-that-there”: V oice and gesture at the graphics inter- face. 14(3):262–270. doi: 10.1145/965105.807503 2

    arXiv

Show all 42 references
  1. [9]

    Buyruk and G

    Y . Buyruk and G. C ¸ a˘gdas ¸. Interactive parametric design and robotic fabrication within mixed reality environment. 12(24):12797. doi: 10. 3390/app122412797 2, 3

  2. [10]

    Caetano, L

    I. Caetano, L. Santos, and A. Leit ˜ao. Computational design in ar- chitecture: Defining parametric, generative, and algorithmic design. 9(2):287–300. doi: 10.1016/j.foar.2019.12.008 1, 2, 3

    2019 doi

  3. [11]

    Z. Cai. 3description: An intuitive human-AI collaborative 3d model- ing approach. In Proceedings of the 11th International Conference on Digital and Interactive Arts , ARTECH ’23. Association for Comput- ing Machinery. event-place: Faro, Portugal. doi: 10.1145/3632776. 3632785 2

    doi

  4. [12]

    Castelo-Branco and A

    R. Castelo-Branco and A. Leit ˜ao. Algorithmic design in virtual reality. 2(1):31–52. doi: 10.3390/architecture2010003 2

    doi

  5. [13]

    Chatterjee

    S. Chatterjee. Free-form shape modeling in XR: A systematic review. Version Number: 1. doi: 10.48550/ARXIV.2401.00924 2

    doi arXiv

  6. [14]

    Cheng, C

    G. Cheng, C. Zhang, W. Cai, L. Zhao, C. Sun, and J. Bian. Empow- ering large language models on robotic manipulation with affordance prompting. Version Number: 1. doi: 10.48550/ARXIV.2404.11027 2

    doi arXiv

  7. [15]

    Coppens, T

    A. Coppens, T. Mens, and M.-A. Gallas. Parametric modelling within immersive environments - building a bridge between existing tools and virtual reality headsets. pp. 711–716. doi: 10.52842/conf.ecaade. 2018.2.711 3

    2018 doi

  8. [16]

    Desolda, A

    G. Desolda, A. Esposito, F. M ¨uller, and S. Feger. Digital modeling for everyone: Exploring how novices approach voice-based 3d mod- eling. vol. 14145 LNCS, pp. 133–155. doi: 10.1007/978-3-031-42293 -5 11 2

    doi

  9. [17]

    Fang and T.-L

    Y .-M. Fang and T.-L. Kao. Comparisons of emotional responses, flow experiences, and operational performances in traditional parametric computer-aided design modeling and virtual-reality free-form model- ing. 13(11):6568. doi: 10.3390/app13116568 3

    doi

  10. [18]

    Friedrich, S

    M. Friedrich, S. Langer, and F. Frey. Combining gesture and voice control for mid-air manipulation of CAD models in VR environments. vol. 2, pp. 119–127. 2

  11. [19]

    L. Gao, A. Madaan, S. Zhou, U. Alon, P. Liu, Y . Yang, J. Callan, and G. Neubig. PAL: Program-aided language models. Version Number:

  12. [20]

    Giunchi, A

    D. Giunchi, A. Sztrajman, S. James, and A. Steed. Mixing modalities of 3d sketching and speech for interactive model retrieval in virtual reality. In Proceedings of the 2021 ACM International Conference on Interactive Media Experiences , IMX ’21, pp. 144–155. Association for C...

    2021

  13. [21]

    doi: 10.48550/ARXIV.2211.10435 2

    doi arXiv

  14. [22]

    Y . Hong, H. Zhen, P. Chen, S. Zheng, Y . Du, Z. Chen, and C. Gan. 3d-LLM: Injecting the 3d world into large language models. Version Number: 1. doi: 10.48550/ARXIV.2307.12981 2

    doi arXiv

  15. [23]

    G ¨urel and B

    A. G ¨urel and B. S ¸enyapılı Ozcan. Cognitive comparison of design methods in the conceptual phase. 21(4):581–601. doi: 10.1177/ 14780771231188474 1, 2

  16. [24]

    H ¨ollein, A

    L. H ¨ollein, A. Cao, A. Owens, J. Johnson, and M. Nießner. Text2room: Extracting textured 3d meshes from 2d text-to-image models. Version Number: 2. doi: 10.48550/ARXIV.2303.11989 2

    doi arXiv

  17. [25]

    Huang, Z

    S. Huang, Z. Jiang, H. Dong, Y . Qiao, P. Gao, and H. Li. Instruct2act: Mapping multi-modality instructions to robotic actions with large lan- guage model. Version Number: 3. doi: 10.48550/ARXIV.2305.11176 2

    doi arXiv

  18. [26]

    Liang, W

    J. Liang, W. Huang, F. Xia, P. Xu, K. Hausman, B. Ichter, P. Florence, and A. Zeng. Code as policies: Language model programs for embod- ied control. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pp. 9493–9500. IEEE. doi: 10.1109/ICRA48891 .2023.1016...

    2023

  19. [27]

    J. H. Lee, N. Gu, and M. J. Ostwald. Creativity and parametric de- sign? comparing designer’s cognitive approaches with assessed levels of creativity. 3(2):78–94. doi: 10.1080/21650349.2014.931826 1, 2

    2014

  20. [28]

    B. Liu, Y . Jiang, X. Zhang, Q. Liu, S. Zhang, J. Biswas, and P. Stone. LLM+p: Empowering large language models with optimal planning proficiency. Version Number: 3. doi: 10.48550/ARXIV.2304.11477 2

    doi arXiv

  21. [29]

    K. Lin, C. Agia, T. Migimatsu, M. Pavone, and J. Bohg. Text2motion: from natural language instructions to feasible plans. 47(8):1345–1365. doi: 10.1007/s10514-023-10131-7 2

    doi

  22. [30]

    C. Sun, J. Han, W. Deng, X. Wang, Z. Qin, and S. Gould. 3d-GPT: Procedural 3d modeling with large language models. Version Number:

  23. [31]

    Salim, H

    F. Salim, H. Mulder, and J. Burry. A system for form fostering: Para- metric modeling of responsive forms in mixed reality. pp. 531–540. doi: 10.52842/conf.caadria.2010.531 3

    2010 doi

  24. [32]

    Z. Wang, S. Cai, G. Chen, A. Liu, X. Ma, and Y . Liang. Describe, ex- plain, plan and select: Interactive planning with large language mod- els enables open-world multi-task agents. Version Number: 3. doi: 10 .48550/ARXIV.2302.01560 2

    arXiv

  25. [33]

    doi: 10.48550/ARXIV.2310.12945 2

    doi arXiv

  26. [34]

    S. H. Vemprala, R. Bonatti, A. Bucker, and A. Kapoor. ChatGPT for robotics: Design principles and model abilities. 12:55682–55696. doi: 10.1109/ACCESS.2024.3387941 2

    2024

  27. [35]

    Williams, J

    A. Williams, J. Garcia, and F. Ortega. Understanding multimodal user gesture and speech behavior for object manipulation in augmented re- ality using elicitation. 26(12):3479–3489. doi: 10.1109/TVCG.2020. 3023566 2

    2020 doi

  28. [36]

    Weng, Y .-M

    S.-C. Weng, Y .-M. Chiou, and E.-L. Do. Dream mesh: A speech-to- 3d model generative pipeline in mixed reality. pp. 345–349. doi: 10. 1109/AIxVR59861.2024.00059 2

    2024

  29. [37]

    White, Q

    J. White, Q. Fu, S. Hays, M. Sandborn, C. Olea, H. Gilbert, A. El- nashar, J. Spencer-Smith, and D. C. Schmidt. A prompt pattern catalog to enhance prompt engineering with ChatGPT. doi: 10.48550/arXiv. 2302.11382 5

    doi arXiv

  30. [38]

    X. Zhou, A. Williams, and F. Ortega. Eliciting multimodal ges- ture+speech interactions in a multi-object augmented reality environ- ment. doi: 10.1145/3562939.3565637 2 Figure 5: Result of user study questionnaire Figure 6: Result of parametric design generation result Figure...

  31. [39]

    Wortmann and B

    T. Wortmann and B. Tunc ¸er. Differentiating parametric design: Digi- tal workflows in contemporary architecture and construction. 52:173–

  32. [41]

    Z. Yin, Y . Wang, T. Papatheodorou, and P. Hui. Text2vrscene: Explor- ing the framework of automated text-driven generation system for VR experience. In 2024 IEEE Conference Virtual Reality and 3D User Interfaces (VR), pp. 701–711. doi: 10.1109/VR58804.2024.00090 2

    2024

  33. [76]

    doi: 10.1145/3603421.3603431 2, 3

    ACM. doi: 10.1145/3603421.3603431 2, 3

  34. [197]

    doi: 10.1016/j.destud.2017.05.004 1

    2017 doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.