pith. sign in

arxiv: 2606.24616 · v1 · pith:I7EFM66Unew · submitted 2026-06-10 · 💻 cs.AI · cs.PF· econ.GN· q-fin.EC

AI Tokenomics: The Economics of Tokens, Computation, and Pricing in Foundation Models

Quanyan Zhu This is my paper

Pith reviewed 2026-06-27 09:57 UTC · model grok-4.3

classification 💻 cs.AI cs.PFecon.GNq-fin.EC
keywords AI tokenomicsfoundation modelstoken pricingeconomic valuecomputation costsworkflow optimizationresource allocation
0
0 comments X

The pith

Token expenditure in foundation models does not equal economic value because productivity, workflow position, and downstream effects determine worth.

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

The paper develops a framework called AI tokenomics to study how tokens function as the basic accounting unit across information processing, computation, energy use, and pricing in large AI systems. It links token-level technical costs to higher-level enterprise resource decisions and market questions. The central distinction is that spending tokens does not automatically create value; value instead arises from marginal productivity, where in the workflow the tokens are used, any hidden reasoning steps, associated risks, and how effects propagate downstream. A reader would care because current pricing and allocation practices often treat token counts as direct proxies for worth, which this view challenges. The paper ends by listing open questions in measurement and dynamic allocation that follow from treating tokens this way.

Core claim

Tokens serve as the practical accounting unit that connects information processing, computation, memory, energy, pricing, and economic value in foundation model services. The framework shows token expenditure and economic value are distinct quantities: value depends on marginal productivity, workflow position, hidden reasoning activity, risk, and downstream propagation effects rather than raw token counts alone.

What carries the argument

The AI tokenomics framework that treats tokens as the linking unit between technical costs and workflow-level production functions.

If this is right

  • Pricing mechanisms for foundation model services should incorporate marginal productivity and workflow position rather than token volume alone.
  • Enterprise resource allocation needs separate instrumentation for hidden reasoning activity and downstream propagation effects.
  • Market design for token-based AI services must account for risk and uncertainty in value realization.
  • Measurement methods must expand beyond visible token counts to capture hidden activity.
  • Optimization of token use requires dynamic allocation rules tied to productivity estimates.

Where Pith is reading between the lines

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

  • Enterprises could test the framework by running identical prompts at different workflow stages and comparing output value metrics against token spend.
  • If value depends on propagation effects, token markets might develop secondary trading for high-productivity positions in multi-agent workflows.
  • Calibration of productivity functions could lead to new contracts where payment is based on verified downstream outcomes rather than upfront token budgets.

Load-bearing premise

Tokens can serve as a sufficient accounting unit that directly connects information processing, computation, energy, pricing, and economic value across AI systems.

What would settle it

An empirical study that measures value created by different tasks and finds it correlates almost perfectly with raw token expenditure regardless of workflow position or hidden reasoning steps.

Figures

Figures reproduced from arXiv: 2606.24616 by Quanyan Zhu.

Figure 1
Figure 1. Figure 1: Representative workflow structures for the Section [PITH_FULL_IMAGE:figures/full_fig_p028_1.png] view at source ↗
read the original abstract

Tokens have become the practical accounting unit for modern foundation model services, linking information processing, computation, memory use, energy expenditure, pricing, and economic value. This paper develops a framework for AI tokenomics: the study of how tokens are generated, consumed, priced, allocated, and optimized across AI systems. We connect token-level technical costs to workflow-level production functions, enterprise resource allocation, measurement and instrumentation methods, and emerging market-design questions. The framework shows that token expenditure and economic value are distinct: value depends on marginal productivity, workflow position, hidden reasoning activity, risk, and downstream propagation effects. The paper concludes by identifying open research directions in hidden-token measurement, empirical calibration, token productivity, dynamic allocation, and token-based markets.

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

1 major / 0 minor

Summary. The manuscript proposes a conceptual framework for 'AI tokenomics' that positions tokens as the unifying accounting unit linking information processing, computation, memory, energy, pricing, and economic value in foundation models. It connects token-level technical costs to workflow-level production functions and enterprise allocation, and asserts that token expenditure and economic value are distinct, with value depending on marginal productivity, workflow position, hidden reasoning activity, risk, and downstream propagation effects. The paper concludes by listing open research directions in measurement, calibration, productivity, allocation, and markets.

Significance. If substantiated with formal models or data, the framework could help organize thinking about resource measurement and optimization in AI services by bridging technical costs and economic value. As presented, its primary contribution is conceptual mapping rather than new derivations, predictions, or measurements.

major comments (1)
  1. [Abstract] Abstract: The central claim that 'token expenditure and economic value are distinct' (with value depending on marginal productivity, hidden reasoning, etc.) is asserted at a definitional level but is not supported by any production function, formal model, illustrative calculation, or empirical example showing how these factors cause value to diverge from token counts.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback. The manuscript is a conceptual framework paper whose primary contribution is organizing concepts at the intersection of technical costs and economic value in foundation models. We address the major comment below and will revise accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 'token expenditure and economic value are distinct' (with value depending on marginal productivity, hidden reasoning, etc.) is asserted at a definitional level but is not supported by any production function, formal model, illustrative calculation, or empirical example showing how these factors cause value to diverge from token counts.

    Authors: We agree that the paper is conceptual rather than providing formal derivations or data. The distinction is introduced as a definitional feature of the tokenomics framework, following directly from the workflow-level production function and allocation concepts developed in the body (linking token costs to marginal productivity, hidden activity, risk, and propagation). No production function, calculation, or example is supplied. In revision we will add a short illustrative example in the abstract and main text demonstrating divergence (e.g., a workflow where hidden reasoning tokens yield higher downstream value than raw token count would predict). revision: yes

Circularity Check

0 steps flagged

No significant circularity; definitional framework only

full rationale

The paper advances a conceptual framework mapping token usage to economic ideas (production functions, marginal productivity, hidden reasoning) without any equations, fitted parameters, quantitative predictions, or theorems. The central distinction between token expenditure and economic value is asserted at the definitional level rather than derived. No self-citations, ansatzes, or uniqueness claims appear as load-bearing elements. The derivation chain is empty by inspection; the work is self-contained as an exploratory proposal with no internal reductions to its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review based solely on abstract; no specific free parameters, axioms, or invented entities are detailed in the provided text.

pith-pipeline@v0.9.1-grok · 5655 in / 975 out tokens · 14236 ms · 2026-06-27T09:57:39.220032+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

66 extracted references · 10 canonical work pages · 5 internal anchors

  1. [1]

    Acemoglu

    D. Acemoglu. Harms of AI. Working Paper 29247, National Bureau of Economic Research,

  2. [2]

    Akash network: Decentralized cloud infrastructure.https://akash

    Akash Network. Akash network: Decentralized cloud infrastructure.https://akash. network/whitepapers/, 2020. (Cited on p. 19)

    2020

  3. [3]

    M. A. Al Bari and Q. Zhu. A gestalt game-theoretic framework for designing agentic AI workflows in cyber deception. InInternational Conference on Game Theory and AI for Security, pages 228–248, Cham, 2025. Springer Nature Switzerland. (Cited on pp. 20, 21, 22)

    2025

  4. [4]

    Qwen API pricing.https://help.aliyun.com, 2026

    Alibaba Cloud. Qwen API pricing.https://help.aliyun.com, 2026. Accessed: 2026-06-

    2026

  5. [5]

    Claude API pricing.https://platform.claude.com/docs/en/ about-claude/pricing, 2026

    Anthropic. Claude API pricing.https://platform.claude.com/docs/en/ about-claude/pricing, 2026. Accessed: 2026-06-09. (Cited on pp. 2, 5, 12, 13, 14)

    2026

  6. [6]

    Context windows.https://platform.claude.com/docs/en/ build-with-claude/context-windows, 2026

    Anthropic. Context windows.https://platform.claude.com/docs/en/ build-with-claude/context-windows, 2026. Accessed 2026-06-09. (Cited on pp. 10, 11)

    2026

  7. [7]

    Plans and pricing.https://claude.com/pricing, 2026

    Anthropic. Plans and pricing.https://claude.com/pricing, 2026. Accessed 2026-06-09. (Cited on p. 15)

    2026

  8. [8]

    L. Bai, Z. Huang, X. Wang, J. Sun, R. Mihalcea, E. Brynjolfsson, A. Pentland, and J. Pei. How do AI agents spend your money? analyzing and predicting token consumption in agentic coding tasks.arXiv preprint arXiv:2604.22750, 2026. (Cited on pp. 10, 11, 12)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  9. [9]

    Eliciting Latent Predictions from Transformers with the Tuned Lens

    N. Belrose, Z. Furman, L. Smith, D. Halawi, I. Ostrovsky, L. McKinney, S. Biderman, and J. Steinhardt. Eliciting latent predictions from transformers with the tuned lens.arXiv preprint arXiv:2303.08112, 2023. (Cited on p. 16) 35

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  10. [10]

    LLM pricing comparison dashboard.https://benchlm.ai/llm-pricing,

    BenchLM. LLM pricing comparison dashboard.https://benchlm.ai/llm-pricing,

  11. [11]

    (Cited on p

    Accessed: 2026-06-09. (Cited on p. 13)

    2026

  12. [12]

    D. P . Bertsekas.Network Optimization: Continuous and Discrete Models. Athena Scientific,

  13. [13]

    (Cited on pp. 22, 24)

  14. [14]

    T. B. Brown, B. Mann, N. Ryder, M. Subbiah, J. Kaplan, P . Dhariwal, A. Neelakantan, P . Shyam, G. Sastry, A. Askell, et al. Language models are few-shot learners. InAdvances in Neural Information Processing Systems, volume 33, pages 1877–1901, 2020. (Cited on pp. 1, 7, 14)

    1901

  15. [15]

    Z. S. Chen and Q. Zhu. A theory of multilevel interactive equilibrium in NeuroAI.arXiv preprint arXiv:2605.10505, 2026. (Cited on p. 19)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  16. [16]

    LLM API cost comparison.https://costgoat.com/compare/llm-api, 2026

    CostGoat. LLM API cost comparison.https://costgoat.com/compare/llm-api, 2026. Accessed: 2026-06-09. (Cited on p. 13)

    2026

  17. [17]

    T. Dao, D. Y. Fu, S. Ermon, A. Rudra, and C. Ré. Flashattention: Fast and memory-efficient exact attention with io-awareness. InAdvances in Neural Information Processing Systems, volume 35, pages 16344–16359, 2022. (Cited on pp. 8, 16)

    2022

  18. [18]

    Deepseek API pricing.https://platform.deepseek.com, 2026

    DeepSeek AI. Deepseek API pricing.https://platform.deepseek.com, 2026. Accessed: 2026-06-09. (Cited on p. 13)

    2026

  19. [19]

    Elhage, N

    N. Elhage, N. Nanda, C. Olsson, T. Henighan, N. Joseph, B. Mann, A. Askell, Y. Bai, A. Chen, T. Conerly, et al. A mathematical framework for transformer circuits.https: //transformer-circuits.pub/2021/framework/index.html, 2021. Transformer Circuits Thread. (Cited on p. 16)

    2021

  20. [20]

    Ghodsi, M

    A. Ghodsi, M. Zaharia, B. Hindman, A. Konwinski, S. Shenker, and I. Stoica. Dominant re- source fairness: Fair allocation of multiple resource types. InProceedings of the 8th USENIX Symposium on Networked Systems Design and Implementation, pages 323–336, 2011. (Cited on pp. 18, 19, 22)

    2011

  21. [21]

    Gemini enterprise agent platform.https://cloud.google.com/ products/gemini-enterprise-agent-platform, 2026

    Google Cloud. Gemini enterprise agent platform.https://cloud.google.com/ products/gemini-enterprise-agent-platform, 2026. Accessed 2026-06-09. (Cited on p. 15)

    2026

  22. [22]

    Gemini API pricing.https://ai.google.dev/gemini-api/docs/ pricing, 2026

    Google DeepMind. Gemini API pricing.https://ai.google.dev/gemini-api/docs/ pricing, 2026. Accessed: 2026-06-09. (Cited on pp. 2, 5, 12, 13, 14) 36

    2026

  23. [23]

    Hoffmann, S

    J. Hoffmann, S. Borgeaud, A. Mensch, E. Buchatskaya, T. Cai, E. Rutherford, D. de Las Casas, L. A. Hendricks, J. Welbl, A. Clark, et al. Training compute-optimal large language models. InAdvances in Neural Information Processing Systems, volume 35, pages 30016–30030, 2022. (Cited on pp. 4, 5, 9, 16, 23)

    2022

  24. [24]

    Scaling Laws for Neural Language Models

    J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Rad- ford, J. Wu, and D. Amodei. Scaling laws for neural language models.arXiv preprint arXiv:2001.08361, 2020. (Cited on pp. 4, 5, 9, 16, 23)

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  25. [25]

    F. P . Kelly, A. K. Maulloo, and D. K. H. Tan. Rate control for communication networks: Shadow prices, proportional fairness and stability.Journal of the Operational Research Soci- ety, 49(3):237–252, 1998. (Cited on pp. 18, 19, 22, 24)

    1998

  26. [26]

    Kudo and J

    T. Kudo and J. Richardson. Sentencepiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66–71,

    2018

  27. [27]

    3, 5, 7)

    (Cited on pp. 3, 5, 7)

  28. [28]

    ChatOpenAI integration.https://docs.langchain.com/oss/python/ integrations/chat/openai, 2026

    LangChain. ChatOpenAI integration.https://docs.langchain.com/oss/python/ integrations/chat/openai, 2026. Accessed 2026-06-09. (Cited on pp. 10, 11)

    2026

  29. [29]

    Lewis, E

    P . Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. Küttler, M. Lewis, W.-t. Yih, T. Rocktäschel, and S. Riedel. Retrieval-augmented generation for knowledge- intensive NLP tasks. InAdvances in Neural Information Processing Systems, volume 33, pages 9459–9474, 2020. (Cited on pp. 7, 23)

    2020

  30. [30]

    Merizzi, T

    N. Merizzi, T. Smith, D. Kearns-Manolatos, N. Mittal, and G. Churiwala. The pivot to tokenomics: Navigating AI’s new spend dynamics. Industry report, Deloitte, Jan. 2026. Accessed January 2026. (Cited on p. 2)

    2026

  31. [31]

    Llama models.https://ai.meta.com/llama, 2026

    Meta AI. Llama models.https://ai.meta.com/llama, 2026. Accessed: 2026-06-09. (Cited on p. 13)

    2026

  32. [32]

    Mistral API pricing.https://docs.mistral.ai, 2026

    Mistral AI. Mistral API pricing.https://docs.mistral.ai, 2026. Accessed: 2026-06-09. (Cited on p. 13)

    2026

  33. [33]

    R. B. Myerson. Optimal auction design.Mathematics of Operations Research, 6(1):58–73,

  34. [34]

    17, 18, 20) 37

    (Cited on pp. 17, 18, 20) 37

  35. [35]

    Nisan, T

    N. Nisan, T. Roughgarden, E. Tardos, and V . V . Vazirani, editors.Algorithmic Game Theory. Cambridge University Press, 2007. (Cited on pp. 17, 18, 20)

    2007

  36. [36]

    Interpreting GPT: The logit lens.https://www.alignmentforum.org/ posts/AcKRB8wDpdaN6v6ru/interpreting-gpt-the-logit-lens, 2020

    Nostalgebraist. Interpreting GPT: The logit lens.https://www.alignmentforum.org/ posts/AcKRB8wDpdaN6v6ru/interpreting-gpt-the-logit-lens, 2020. Alignment Fo- rum. (Cited on p. 16)

    2020

  37. [37]

    API pricing.https://openai.com/api/pricing/, 2026

    OpenAI. API pricing.https://openai.com/api/pricing/, 2026. Accessed: 2026-06-09. (Cited on pp. 2, 5, 12, 13, 14)

    2026

  38. [38]

    Prompt engineering.https://developers.openai.com/api/docs/guides/ prompt-engineering, 2026

    OpenAI. Prompt engineering.https://developers.openai.com/api/docs/guides/ prompt-engineering, 2026. Accessed 2026-06-09. (Cited on p. 10)

    2026

  39. [39]

    J. S. Park, J. C. O’Brien, C. J. Cai, M. R. Morris, P . Liang, and M. S. Bernstein. Generative agents: Interactive simulacra of human behavior. InProceedings of the 36th Annual ACM Symposium on User Interface Software and T echnology, pages 1–22, 2023. (Cited on p. 19)

    2023

  40. [40]

    Filecoin: A decentralized storage network.https://filecoin.io/ filecoin.pdf, 2017

    Protocol Labs. Filecoin: A decentralized storage network.https://filecoin.io/ filecoin.pdf, 2017. (Cited on p. 19)

    2017

  41. [41]

    Radford, J

    A. Radford, J. Wu, R. Child, D. Luan, D. Amodei, and I. Sutskever. Language models are unsupervised multitask learners.OpenAI T echnical Report, 2019. (Cited on p. 1)

    2019

  42. [42]

    V . J. Reddi, C. Cheng, D. Kanter, P . Mattson, G. Schmuelling, C.-J. Wu, B. Anderson, M. Breughe, M. Charlebois, W. Chou, et al. MLPerf inference benchmark. In2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture, pages 446–459,

  43. [43]

    Rochet and J

    J.-C. Rochet and J. Tirole. Platform competition in two-sided markets.Journal of the Euro- pean Economic Association, 1(4):990–1029, 2003. (Cited on p. 20)

    2003

  44. [44]

    Schick, J

    T. Schick, J. Dwivedi-Yu, R. Dessì, R. Raileanu, M. Lomeli, E. Hambro, L. Zettlemoyer, N. Cancedda, and T. Scialom. Toolformer: Language models can teach themselves to use tools. InAdvances in Neural Information Processing Systems, volume 36, pages 68539–68551,

  45. [45]

    4, 10, 15, 19)

    (Cited on pp. 4, 10, 15, 19)

  46. [46]

    Sennrich, B

    R. Sennrich, B. Haddow, and A. Birch. Neural machine translation of rare words with subword units. InProceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1715–1725, 2016. (Cited on pp. 3, 5, 7) 38

    2016

  47. [47]

    Shoham and K

    Y. Shoham and K. Leyton-Brown.Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations. Cambridge University Press, 2008. (Cited on p. 19)

    2008

  48. [48]

    Strubell, A

    E. Strubell, A. Ganesh, and A. McCallum. Energy and policy considerations for deep learning in NLP. InProceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3645–3650, 2019. (Cited on p. 17)

    2019

  49. [49]

    Topcuoglu, S

    H. Topcuoglu, S. Hariri, and M.-Y. Wu. Performance-effective and low-complexity task scheduling for heterogeneous computing.IEEE T ransactions on Parallel and Distributed Sys- tems, 13(3):260–274, 2002. (Cited on p. 22)

    2002

  50. [50]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is all you need. InAdvances in Neural Information Processing Sys- tems, volume 30, 2017. (Cited on pp. 1, 8, 14, 16)

    2017

  51. [51]

    X. Wang, J. Wei, D. Schuurmans, Q. Le, E. Chi, S. Narang, A. Chowdhery, and D. Zhou. Self-consistency improves chain of thought reasoning in language models. InInternational Conference on Learning Representations, 2023. (Cited on pp. 8, 10, 16)

    2023

  52. [52]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. H. Chi, Q. V . Le, and D. Zhou. Chain-of-thought prompting elicits reasoning in large language models. InAd- vances in Neural Information Processing Systems, volume 35, pages 24824–24837, 2022. (Cited on pp. 8, 10, 16)

    2022

  53. [53]

    Wooldridge.An Introduction to MultiAgent Systems

    M. Wooldridge.An Introduction to MultiAgent Systems. Wiley, 2 edition, 2009. (Cited on p. 19)

    2009

  54. [54]

    C.-J. Wu, R. Raghavendra, U. Gupta, B. Acun, N. Ardalani, K. Maeng, G. Chang, F. A. Behram, J. Huang, C. Bai, et al. Sustainable AI: Environmental implications, challenges and opportunities. InProceedings of Machine Learning and Systems, volume 4, pages 795– 813, 2022. (Cited on p. 17)

    2022

  55. [55]

    Grok API pricing.https://docs.x.ai, 2026

    xAI. Grok API pricing.https://docs.x.ai, 2026. Accessed: 2026-06-09. (Cited on pp. 2, 5, 12, 13)

    2026

  56. [56]

    Yang and Q

    Y.-T. Yang and Q. Zhu. PACT: A contract-theoretic framework for pricing agentic AI ser- vices powered by large language models. InProceedings of the 2025 IEEE Global Communi- cations Conference (GLOBECOM), Taipei, Taiwan, 2025. IEEE. (Cited on pp. 15, 34) 39

    2025

  57. [57]

    Yang and Q

    Y.-T. Yang and Q. Zhu. Internet of agentic AI: Incentive-compatible distributed teaming and workflow.arXiv preprint arXiv:2602.03145, 2026. (Cited on pp. 19, 22)

    work page arXiv 2026

  58. [58]

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y. Cao. ReAct: Synergizing reasoning and acting in language models. InInternational Conference on Learning Represen- tations, 2023. (Cited on pp. 4, 10, 15, 19)

    2023

  59. [59]

    Glm API pricing.https://open.bigmodel.cn, 2026

    Zhipu AI. Glm API pricing.https://open.bigmodel.cn, 2026. Accessed: 2026-06-09. (Cited on p. 13)

    2026

  60. [60]

    Q. Zhu. Foundations of cyber resilience: The confluence of game, control, and learning theories.arXiv preprint arXiv:2404.01205, 2024. (Cited on pp. 19, 20)

    work page arXiv 2024

  61. [61]

    Q. Zhu. Game theory meets LLM and agentic AI: Reimagining cybersecurity for the age of intelligent threats.arXiv preprint arXiv:2507.10621, 2025. (Cited on pp. 20, 21)

    work page arXiv 2025

  62. [62]

    Q. Zhu. Insurance of agentic AI.arXiv preprint arXiv:2606.05449, 2026. (Cited on p. 25)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  63. [63]

    Zhu and M

    Q. Zhu and M. A. Al Bari. Agentic AI for cyber deception: A gestalt game-theoretic ap- proach to defending against botnet DDoS attacks. InProceedings of the 59th Hawaii Interna- tional Conference on System Sciences, 2026. (Cited on pp. 20, 21, 22)

    2026

  64. [64]

    Zhu and T

    Q. Zhu and T. Ba¸ sar. Revisiting game-theoretic control in socio-technical networks: Emerg- ing design frameworks and contemporary applications.arXiv preprint arXiv:2411.01794,

    work page arXiv

  65. [65]

    19, 20, 22)

    (Cited on pp. 19, 20, 22)

  66. [66]

    Zhu and Z

    Q. Zhu and Z. Han. Learning, misspecification, and cognitive arbitrage in linear-quadratic network games.arXiv preprint arXiv:2603.17157, 2026. (Cited on pp. 19, 22) 40

    work page arXiv 2026