Pith. sign in

REVIEW 3 major objections 2 minor 42 references

Multi-Hop Privacy Propagation for Differentially Private Federated Learning in Social Networks

T0 review · 3 major / 2 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read In social-network-connected federated learning, a client's privacy loss spreads through neighbors' choices, so this paper models privacy-budget setting as a server-client game and proves its equilibrium is stable and near-optimal in social

desk verdict Abstract is plausible, but the attached full text is a different paper, so none of the claimed proofs can actually be checked. read the letter →

arxiv 2508.07676 v1 pith:FYRVOTRH submitted 2025-08-11 cs.LG cs.DCcs.GT

Chenchen Lin , Xuehe Wang This is my paper
classification cs.LGcs.DCcs.GT
keywords federatedlearningdifferentialprivacysocialnetworkspropagationStackelberggamemean-fieldestimatorpriceofanarchybudget
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

In federated learning over a social network, a client's privacy loss is not a private matter: noise choices made by one client propagate along social edges and change the risk faced by others. This paper argues that privacy-budget setting is therefore a strategic interaction, and proposes a mechanism where a server first commits to incentive policies and each client then chooses its noise level. A multi-hop propagation model quantifies the indirect leakage, and a mean-field estimator lets clients work with an average external risk instead of full network information. The paper proves that this game has a convergent fixed point, yields closed-form equilibrium strategies, and is approximately socially optimal, with experiments indicating higher client utility and lower server cost at maintained model accuracy. If true, the result turns privacy-budget allocation in networked federated learning from a per-client choice into a globally tunable equilibrium.

What carries the argument

The argument is carried by the coupling of three components: (1) a multi-hop privacy propagation model that quantifies how a client's risk depends on others' privacy decisions along social-network paths; (2) a two-stage Stackelberg game in which the server leads with incentive policies and clients follow by choosing privacy budgets (noise magnitudes); and (3) a mean-field estimator that approximates the average external privacy risk, whose fixed point the paper proves to exist and converge. The price-of-anarchy analysis on the resulting equilibrium is what yields the approximately-optimal social welfare claim.

What would settle it

Run a small federated learning federation with three to five clients on a known social graph, fix differential-privacy noise budgets, and measure how one client's actual inference risk changes as neighbors tighten or loosen their budgets; compare that measured change with the multi-hop propagation formula. Separately, compute the exact average external risk on that graph and compare it with the fixed point of the mean-field estimator; if the error does not shrink as the graph grows or stays large on a clustered topology, the estimator's convergence guarantee is not enough to support the welfar

Watch

Extended reading notes

Core claim

Central claim: in federated learning over a social-network topology, privacy loss is contagious—a client's risk depends on the noise budgets chosen by its neighbors, and their neighbors in turn—so optimal privacy protection is not a private decision but the outcome of a strategic interaction. The paper's mechanism quantifies this indirect leakage with a multi-hop propagation model, then organizes the server-client interaction as a two-stage Stackelberg game in which the server commits to incentive policies and each client best-responds with a privacy budget (equivalently, a noise magnitude). Information asymmetry is handled by a mean-field estimator of the average external risk, for which th

Load-bearing premise

The load-bearing premise is that privacy loss composes along social-network edges in the paper's multi-hop form and that the mean-field estimator, fed with clients' equilibrium strategies, closely tracks the true average external privacy risk; if either fails, the equilibrium and welfare results hold only inside the model.

Editorial extensions

If this is right

  • Privacy-budget allocation in networked FL becomes a global game: a server can commit to incentives and clients' best responses form a stable, closed-form equilibrium.
  • Clients need not observe the whole network; the convergent mean-field estimator supplies enough information to choose noise levels.
  • The mechanism's client-centric design does not sacrifice social welfare—the price-of-anarchy bound keeps the outcome within a constant factor of the optimum.
  • Deployment requires no iterative privacy negotiation: the server commits once and clients respond, so the equilibrium is directly computable.
  • Empirically, the claimed effect is concrete: higher client utility and lower server cost while model accuracy is maintained.

Reading between the lines

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

  • We infer that the mean-field approximation should improve with network size and mixing; the mechanism is most likely to deliver its welfare bound in large, dense social graphs, and least likely in small or sharply clustered networks—an empirical prediction the paper does not make.
  • The same template (server as Stackelberg leader, clients as budget setters, mean-field risk) transfers naturally to other privacy-sensitive distributed settings, such as gossip learning or device-to-device federated training, where the 'social' graph is just the communication topology.
  • A testable extension is to estimate the propagation decay parameter from real social data and measure how sensitive the equilibrium and the price-of-anarchy bound are to misestimation; the paper's guarantees likely assume this parameter is known.
  • If the mechanism is deployed, a server could tune incentives to steer the equilibrium; an open question the paper leaves implicit is whether that steering can shift welfare without violating the PoA bound.
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

3 major / 2 minor

Summary. The manuscript claims to present a socially aware privacy-preserving federated learning mechanism for networked clients, built on a multi-hop privacy propagation model, a two-stage Stackelberg game, and a mean-field estimator of external privacy risk. The abstract states that the authors prove existence and convergence of the estimator's fixed point, derive closed-form Stackelberg Nash equilibria, and establish an approximate social-welfare guarantee via a price-of-anarchy bound, with experiments showing improved client utility and reduced server cost. However, the supplied full text is not the manuscript of arXiv:2508.07676; it is an unrelated position paper on personalized conversational recommendation agents and privacy alignment (arXiv:2508.07672v1). None of the claimed definitions, theorems, proofs, or experimental details appear in the submitted file. The only assessable content is the abstract, which states results without derivations or experimental specifics. I therefore cannot verify the central technical claims.

Significance. If the claimed results were substantiated, the contribution would be significant: it would move privacy-budget selection in networked FL from isolated per-client decisions to a global game-theoretic treatment with quantified multi-hop externalities, while providing a tractable equilibrium and welfare analysis. Explicitly deriving a mean-field fixed point, closed-form Stackelberg equilibrium, and a price-of-anarchy bound would be a useful theoretical contribution with potential practical implications for incentive design in FL over social networks. However, the significance is currently prospective: the submitted manuscript does not contain the claimed content, so the contribution is not assessable. The work also appears likely to be of interest to the journal's audience if the full technical content is provided and sound.

major comments (3)
  1. [Full text (arXiv:2508.07672v1) vs abstract (arXiv:2508.07676)] The full text supplied for review is not the paper described in the abstract. It is a position paper on aligning conversational recommendation agents with users' privacy preferences, with no mention of federated learning, multi-hop privacy propagation, Stackelberg games, mean-field estimators, or social networks. None of the central claims — fixed-point existence and convergence, closed-form Nash equilibrium, price-of-anarchy bound, or the experimental results — can be checked against the submitted text. This is a load-bearing mismatch that prevents substantive review.
  2. [Abstract, theoretical claims] Even taking the abstract at face value, the core theoretical results are asserted without any supporting statements. No definitions are given for the multi-hop propagation model, the client utility or server cost functions, the equilibrium concept, or the price-of-anarchy measure. The reader cannot determine whether the theorems are correct, under what assumptions they hold, or whether the closed forms are genuinely derived rather than assumed. This is not a presentation issue; it is the absence of the paper's technical content.
  3. [Abstract, mean-field estimator and equilibrium dependence] The abstract introduces a mean-field estimator that approximates average external privacy risk, and clients' strategies depend on that estimate. The claimed price-of-anarchy bound and approximate-optimal welfare are therefore conditional on the estimator's accuracy and on the uniqueness/stability of its fixed point. The abstract provides no error analysis, convergence rate, or discussion of multiple fixed points. While this is not a demonstrated error, it is a load-bearing point that must be substantiated in the actual manuscript; it is entirely absent here.
minor comments (2)
  1. [Abstract, experiments] The experimental summary is too brief: no dataset names, number of clients, graph topology, baseline implementations, evaluation metrics, or error bars are reported. The claim that the method 'outperforms' baselines cannot be assessed. This would need to be corrected in a resubmission.
  2. [General] The abstract uses terms such as 'multi-hop propagation model' and 'mean-field estimator' without explanation. In a self-contained submission, formal definitions and notation should be provided early.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity detectable from available text; full-text mismatch prevents inspection of the claimed derivations.

full rationale

The abstract alone makes claims about a multi-hop privacy propagation model, a two-stage Stackelberg game, a mean-field estimator, fixed-point convergence, a closed-form Nash equilibrium, and a Price-of-Anarchy bound, but it contains no equations and no derived results. The submitted full text is not the manuscript of arXiv:2508.07676; it is arXiv:2508.07672v1, an unrelated position paper on personalized conversational recommendation agents. Consequently, none of the central proofs can be inspected. This is a serious evidentiary gap, but it is not circularity: the abstract does not define a quantity in terms of the result it is supposed to predict, fit a parameter and then call it a prediction, or rest a load-bearing premise on a self-citation. The structural concern that the mean-field estimator's fixed point is part of the equilibrium and then used in the welfare analysis is a potential model-internal circularity, but without the actual equations or a specific reduction (e.g., an estimator whose 'accuracy' is defined as its own fixed point, or a PoA benchmark that is identical to the equilibrium by construction) it would be speculation to flag it. The reviewing rule requires quoting the paper and exhibiting the reduction; the available text does not permit that. I therefore find no significant circularity and assign a score of 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 2 invented entities

Ledger compiled from the abstract only. Free parameters cannot be enumerated because no equations or calibration details are available: the actual full text of the paper was not supplied. The two methodological constructs (multi-hop propagation model, mean-field estimator) are recorded as invented entities because the abstract gives no external falsifiable handle for either; their fidelity is the load-bearing premise for the welfare claim. The axioms capture the structural premises of the model: propagation composition, client rationality, estimator accuracy, and background DP noise mechanics. A full audit requires the actual PDF of arXiv:2508.07676.

assumptions (4)
  • domain assumption Privacy loss propagates across social-network links and accumulates over multiple hops in the assumed compositional form.
    The abstract's core externality premise ('multi-hop propagation model'). If real leakage composes differently, the game formulation and PoA results change.
  • domain assumption Clients are rational followers in a two-stage Stackelberg game, choosing privacy budgets to maximize utility given server incentives.
    Equilibrium analysis presupposes strategic self-interested clients; stated in the abstract's game description.
  • domain assumption The mean-field estimator's fixed point converges to a value close to the true average external privacy risk.
    Abstract claims existence and convergence proofs but no accuracy guarantee; the near-optimal welfare claim inherits this assumption.
  • standard math Standard differential privacy mechanics hold: a chosen privacy budget sets the noise magnitude as in DP-FL.
    Background machinery the mechanism builds on; no composition or calibration details appear in the abstract.
invented entities (2)
  • Multi-hop privacy propagation model
    purpose: Defines how a client's privacy loss depends on other clients' privacy budgets through multi-hop social-network interactions; the foundation of the externality analysis.
    Core modeling device introduced in the problem formulation. The abstract provides no external validation that real privacy leakage follows this composition rule, so the model is only supported by the paper's own framework and experiments.
  • Mean-field estimator of external privacy risk
    purpose: Approximates the average external privacy risk under information asymmetry so clients can make budget decisions without observing others' choices.
    The abstract claims existence and convergence of its fixed point but provides no error or calibration analysis, so its accuracy against true leakage is unverified. Its role in the PoA analysis makes near-optimality dependent on its fidelity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Hop Privacy Propagation for Differentially Private Federated Learning in Social Networks." pith.science (2026). https://pith.science/paper/FYRVOTRH

@misc{pith2026250807676,
  author       = {Pith},
  title        = {Pith review of: Multi-Hop Privacy Propagation for Differentially Private Federated Learning in Social Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FYRVOTRH}},
  note         = {Machine review of arXiv:2508.07676}
}
read the original abstract

Federated learning (FL) enables collaborative model training across decentralized clients without sharing local data, thereby enhancing privacy and facilitating collaboration among clients connected via social networks. However, these social connections introduce privacy externalities: a client's privacy loss depends not only on its privacy protection strategy but also on the privacy decisions of others, propagated through the network via multi-hop interactions. In this work, we propose a socially-aware privacy-preserving FL mechanism that systematically quantifies indirect privacy leakage through a multi-hop propagation model. We formulate the server-client interaction as a two-stage Stackelberg game, where the server, as the leader, optimizes incentive policies, and clients, as followers, strategically select their privacy budgets, which determine their privacy-preserving levels by controlling the magnitude of added noise. To mitigate information asymmetry in networked privacy estimation, we introduce a mean-field estimator to approximate the average external privacy risk. We theoretically prove the existence and convergence of the fixed point of the mean-field estimator and derive closed-form expressions for the Stackelberg Nash Equilibrium. Despite being designed from a client-centric incentive perspective, our mechanism achieves approximately-optimal social welfare, as revealed by Price of Anarchy (PoA) analysis. Experiments on diverse datasets demonstrate that our approach significantly improves client utilities and reduces server costs while maintaining model performance, outperforming both Social-Agnostic (SA) baselines and methods that account for social externalities.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 32 canonical work pages

  1. [1]

    D. A. E. Acar, Y. Zhao, R. M. Navarro, M. Mattina, P. N. Whatmough, and V. Saligrama. Federated learning based on dynamic regularization. arXiv preprint arXiv:2111.04263, 2021

    arXiv 2021

  2. [2]

    C. A. Arevalo, S. L. Noorbakhsh, Y. Dong, Y. Hong, and B. Wang. Task-agnostic privacy-preserving representation learning for federated learning against attribute inference attacks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 10909--10917, 2024

    work page 2024

  3. [3]

    Bun and T

    M. Bun and T. Steinke. Concentrated differential privacy: Simplifications, extensions, and lower bounds. In Theory of cryptography conference, pages 635--658. Springer, 2016

    2016

  4. [4]

    Y. Chen, W. Xu, X. Wu, M. Zhang, and B. Luo. Personalized local differentially private federated learning with adaptive client sampling. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 6600--6604. IEEE, 2024

    work page 2024

  5. [5]

    L. N. Darlow, E. J. Crowley, A. Antoniou, and A. J. Storkey. Cinic-10 is not imagenet or cifar-10. arXiv preprint arXiv:1810.03505, 2018

    arXiv 2018

  6. [6]

    N. Ding, Z. Sun, E. Wei, and R. Berry. Incentive mechanism design for federated learning and unlearning. In Proceedings of the Twenty-fourth International Symposium on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing, pages 11--20, 2023

    work page 2023

  7. [7]

    Dwork, A

    C. Dwork, A. Roth, et al. The algorithmic foundations of differential privacy. Foundations and Trends in Theoretical Computer Science , 9 0 (3--4): 0 211--407, 2014

    2014

  8. [8]

    Geiping, H

    J. Geiping, H. Bauermeister, H. Dr \"o ge, and M. Moeller. Inverting gradients-how easy is it to break privacy in federated learning? Advances in neural information processing systems, 33: 0 16937--16947, 2020

    work page 2020

Show all 42 references
  1. [9]

    P. Guo, S. Zeng, W. Chen, X. Zhang, W. Ren, Y. Zhou, and L. Qu. A new federated learning framework against gradient inversion attacks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 16969--16977, 2025

    2025

  2. [10]

    C. He, C. Tan, H. Tang, S. Qiu, and J. Liu. Central server free federated learning over single-sided trust social networks. arXiv preprint arXiv:1910.04956, 2019

    1910 arXiv

  3. [11]

    H. Hu, Z. Salcic, L. Sun, G. Dobbie, and X. Zhang. Source inference attacks in federated learning. In 2021 IEEE International Conference on Data Mining (ICDM), pages 1102--1107. IEEE, 2021

    2021

  4. [12]

    H. X. Hu, C. Cao, Q. Hu, and Y. Zhang. Federated learning enabled graph convolutional autoencoder and factorization machine for potential friendship prediction in social networks. Information Fusion, 102: 0 102042, 2024

    2024

  5. [13]

    S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh. Scaffold: Stochastic controlled averaging for federated learning. In International conference on machine learning, pages 5132--5143. PMLR, 2020

    2020

  6. [14]

    L. U. Khan, Z. Han, D. Niyato, and C. S. Hong. Socially-aware-clustering-enabled federated learning for edge networks. IEEE Transactions on Network and Service Management, 18 0 (3): 0 2641--2658, 2021

    2021

  7. [15]

    Krizhevsky, G

    A. Krizhevsky, G. Hinton, et al. Learning multiple layers of features from tiny images, 2009

    2009

  8. [16]

    T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith. Federated optimization in heterogeneous networks. Proceedings of Machine learning and systems, 2: 0 429--450, 2020

    2020

  9. [17]

    X. Lin, J. Wu, J. Li, X. Zheng, and G. Li. Friend-as-learner: Socially-driven trustworthy and efficient wireless federated edge learning. IEEE Transactions on Mobile Computing, 22 0 (1): 0 269--283, 2021

    2021

  10. [18]

    G. Liu, Q. Yang, H. Wang, X. Lin, and M. P. Wittie. Assessment of multi-hop interpersonal trust in social networks by three-valued subjective logic. In IEEE INFOCOM 2014-IEEE Conference on Computer Communications, pages 1698--1706. IEEE, 2014

    2014

  11. [19]

    W. Mao, Q. Ma, G. Liao, and X. Chen. Game analysis and incentive mechanism design for differentially private cross-silo federated learning. IEEE Transactions on Mobile Computing, 23 0 (10): 0 9337--9351, 2024

    2024

  12. [20]

    McMahan, E

    B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas. Communication-efficient learning of deep networks from decentralized data. In Artificial intelligence and statistics, pages 1273--1282. PMLR, 2017

    2017

  13. [21]

    H. B. McMahan, D. Ramage, K. Talwar, and L. Zhang. Learning differentially private recurrent language models. In International Conference on Learning Representations, 2018

    2018

  14. [22]

    Netzer, T

    Y. Netzer, T. Wang, A. Coates, A. Bissacco, B. Wu, A. Y. Ng, et al. Reading digits in natural images with unsupervised feature learning. In NIPS workshop on deep learning and unsupervised feature learning, volume 2011, page 4. Granada, 2011

    2011

  15. [23]

    J. S. Ng, W. Y. B. Lim, H.-N. Dai, Z. Xiong, J. Huang, D. Niyato, X.-S. Hua, C. Leung, and C. Miao. Joint auction-coalition formation framework for communication-efficient federated learning in uav-enabled internet of vehicles. IEEE Transactions on Intelligent Transportation S...

    2020

  16. [24]

    Nguyen, T.-H

    D.-L. Nguyen, T.-H. Nguyen, T.-H. Do, and M. Yoo. Probability-based multi-hop diffusion method for influence maximization in social networks. Wireless Personal Communications, 93 0 (4): 0 903--916, 2017

    2017

  17. [25]

    Papadimitriou

    C. Papadimitriou. Algorithms, games, and the internet. In Proceedings of the thirty-third annual ACM symposium on Theory of computing, pages 749--753, 2001

    2001

  18. [26]

    Rakhlin, O

    A. Rakhlin, O. Shamir, and K. Sridharan. Making gradient descent optimal for strongly convex stochastic optimization. arXiv preprint arXiv:1109.5647, 2011

    2011 arXiv

  19. [27]

    K. Ren, G. Liao, Q. Ma, and X. Chen. Differentially private auction design for federated learning with non-iid data. IEEE Transactions on Services Computing, 17 0 (5): 0 2236--2247, 2023

    2023

  20. [28]

    P. Sun, G. Liao, X. Chen, and J. Huang. A socially optimal data marketplace with differentially private federated learning. IEEE/ACM Transactions on Networking, 32 0 (3): 0 2221--2236, 2024

    2024

  21. [29]

    Truex, L

    S. Truex, L. Liu, K.-H. Chow, M. E. Gursoy, and W. Wei. Ldp-fed: Federated learning with local differential privacy. In Proceedings of the third ACM international workshop on edge systems, analytics and networking, pages 61--66, 2020

    2020

  22. [30]

    Y. Wang, Z. Su, Y. Pan, A. Benslimane, Y. Liu, T. H. Luan, and R. Li. Trade privacy for utility: A learning-based privacy pricing game in federated learning. In ICC 2023-IEEE International Conference on Communications, pages 6307--6311. IEEE, 2023

    2023

  23. [31]

    K. Wei, J. Li, M. Ding, C. Ma, H. Su, B. Zhang, and H. V. Poor. User-level privacy-preserving federated learning: Analysis and performance optimization. IEEE Transactions on Mobile Computing, 21 0 (9): 0 3388--3401, 2021

    2021

  24. [32]

    M. Wu, D. Ye, J. Ding, Y. Guo, R. Yu, and M. Pan. Incentivizing differentially private federated learning: A multidimensional contract approach. IEEE Internet of Things Journal, 8 0 (13): 0 10639--10651, 2021

    2021

  25. [33]

    H. Xiao, K. Rasul, and R. Vollgraf. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747, 2017

    2017 arXiv

  26. [34]

    Y. Xu, M. Xiao, H. Tan, A. Liu, G. Gao, and Z. Yan. Incentive mechanism for differentially private federated learning in industrial internet of things. IEEE Transactions on Industrial Informatics, 18 0 (10): 0 6927--6939, 2021

    2021

  27. [35]

    Y. Xu, M. Xiao, Y. Zhu, J. Wu, S. Zhang, and J. Zhou. Aoi-guaranteed incentive mechanism for mobile crowdsensing with freshness concerns. IEEE Transactions on Mobile Computing, 23 0 (5): 0 4107--4125, 2023

    2023

  28. [36]

    Y. Xu, M. Yin, M. Fang, and N. Z. Gong. Robust federated learning mitigates client-side training data distribution inference attacks. In Companion Proceedings of the ACM Web Conference 2024, pages 798--801, 2024

    2024

  29. [37]

    G. Yang, Z. Shi, S. He, and J. Zhang. Socially privacy-preserving data collection for crowdsensing. IEEE Transactions on Vehicular Technology, 69 0 (1): 0 851--861, 2019

    2019

  30. [38]

    D. Yu, K. Zhang, Y. Tao, W. Xu, Y. Zou, and X. Cheng. Correlation-aware and personalized privacy-preserving data collection. In 2024 International Conference on Computing, Networking and Communications (ICNC), pages 724--729. IEEE, 2024

    2024

  31. [39]

    H. Yu, Z. Liu, Y. Liu, T. Chen, M. Cong, X. Weng, D. Niyato, and Q. Yang. A fairness-aware incentive scheme for federated learning. In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, pages 393--399, 2020

    2020

  32. [40]

    Yuan and X

    W. Yuan and X. Wang. A game-theoretic framework for privacy-aware client sampling in federated learning. IEEE Transactions on Networking, 2025

    2025

  33. [41]

    Zhan and J

    Y. Zhan and J. Zhang. An incentive mechanism design for efficient edge learning by deep reinforcement learning approach. In IEEE INFOCOM 2020-IEEE conference on computer communications, pages 2489--2498. IEEE, 2020

    2020

  34. [42]

    Y. Zhan, C. H. Liu, Y. Zhao, J. Zhang, and J. Tang. Free market of multi-leader multi-follower mobile crowdsensing: An incentive mechanism design by deep reinforcement learning. IEEE Transactions on Mobile Computing, 19 0 (10): 0 2316--2329, 2019

    2019

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.