Pith. sign in

REVIEW 1 major objections 6 minor 42 references

Evaluation of Combination Therapy amid Patient-Level Heterogeneity

T0 review · 1 major / 6 minor · reviewed 2026-07-10 · glm-5.2

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

Pith's one-line read Combination drug benefit may be patient matching, not synergy

desk verdict Solid causal-inference framework for combination therapy; main soft spot is the homoscedasticity assumption that underpins the matching, not just Bliss independence. read the letter →

arxiv 2607.08058 v1 pith:RNHTXDAR submitted 2026-07-09 stat.ME

Lingjie Wei , Xinzhou Guo , Shuoxun Xu This is my paper
classification stat.ME
keywords combinationpatient-leveltherapyheterogeneitymonotherapiesamidbenefitsclassical
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

The paper argues that the apparent superiority of many combination therapies over single drugs can be explained without invoking pharmacological synergy between the combined agents. If different patients respond best to different monotherapies, then giving both drugs together looks better on average simply because each patient gets at least one drug that works for them — a phenomenon pharmacologists call independent action. The classical statistical test, which compares the combination's average outcome to each monotherapy's average, cannot distinguish this heterogeneity-driven benefit from a genuine drug-drug interaction. The authors propose a new target parameter: the difference between the combination's mean outcome and the mean of each patient's best monotherapy response. Testing whether this difference is positive isolates true pharmacological interaction from the ceiling effect of patient-level heterogeneity.

What carries the argument

The estimation strategy is an outcome-based optimal matching scheme. For each patient who received monotherapy A, the algorithm finds a patient in the monotherapy B arm whose estimated conditional expected outcome under B is closest, then imputes the missing Y(B) using the matched patient's observed outcome. This mimics an independent draw from the conditional distribution of Y(B) given that patient's covariates, allowing direct computation of max{Y_i(A), Y_ei(B)} as a transparent proxy for the unobservable max{Y_i(A), Y_i(B)}. The key theoretical innovation is a piecewise inverse-weighting argument that bounds the approximation error between the matched quantity and a pseudo-oracle quantity

What would settle it

A clinical trial where the combination therapy's mean outcome exceeds the heterogeneity-adjusted benchmark E[max{Y(A), Y(B)}] estimated under conditional Bliss independence, but where independent cell-level or biomarker data subsequently reveals strong positive cross-world dependence between Y(A) and Y(B) — showing that the apparent interaction was an artifact of the independence assumption rather than a genuine pharmacological synergy.

Watch

Extended reading notes

Core claim

The central object is the cross-world parameter τ = E[Y(A+B)] − E[max{Y(A), Y(B)}], where Y(A), Y(B), and Y(A+B) are potential outcomes under monotherapy A, monotherapy B, and combination A+B. Because max{Y(A), Y(B)} depends on the joint distribution of two outcomes never observed simultaneously in the same patient, τ is not identifiable by standard clinical-trial methods. The paper shows that under a conditional Bliss independence assumption — requiring that, given baseline covariates, the indicator events of responding to drug A and responding to drug B are uncorrelated at every threshold — the quantity E[max{Y(A), Y(B)}] reduces to a functional of identifiable marginal distributions. The

Load-bearing premise

The identification result requires that, conditional on observed covariates, whether a patient responds to drug A is uncorrelated with whether they respond to drug B at every response threshold. This cross-world independence condition cannot be tested from trial data because both outcomes are never observed in the same patient, and it may fail when two drugs share biological resistance pathways.

Editorial extensions

If this is right

  • Regulatory agencies could require the heterogeneity-adjusted test alongside classical superiority comparisons before approving combination therapies, especially when the combination adds toxicity risk.
  • Drug developers designing combination trials could use the matching framework to pre-specify whether the trial is powered to detect interaction beyond independent action, rather than merely marginal superiority.
  • The sensitivity analysis parameterizing cross-world correlation provides a concrete language for regulators and sponsors to discuss how much unmeasured dependence between drug responses would need to exist before the interaction conclusion flips.
  • The piecewise inverse-weighting proof technique for nonlinear matching estimands could be adapted to other cross-world or joint-potential-outcome parameters in causal inference.
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

1 major / 6 minor

Summary. This paper proposes a new statistical inference framework for evaluating combination therapy that adjusts for patient-level heterogeneity in monotherapy responses. The key idea is that classical comparisons of combination therapy versus the best monotherapy (by marginal mean) can be overly optimistic: when patients respond differentially to different monotherapies, the relevant benchmark is E[max{Y(A), Y(B)}] rather than max{E[Y(A)], E[Y(B)]}. The authors define the target parameter tau = E[Y(A+B)] - E[max{Y(A), Y(B)}], establish (partial) identifiability under a conditional Bliss independence assumption, and propose an outcome-based optimal matching estimator with sqrt(N)-rate asymptotic normality. The method is applied to the ACTG 175 trial. The core idea is well-motivated by pharmacological literature, and the technical contributions address genuine nonlinearity and non-identifiability challenges.

Significance. The paper addresses an important and practically relevant problem at the intersection of causal inference and pharmacology. The distinction between patient-level heterogeneity and true pharmacological interaction is well-grounded in the pharmacological literature (Palmer and Sorger, 2017), and translating this into a formal potential-outcome framework with identifiability results is a valuable contribution. The outcome-based matching scheme for a nonlinear estimand, the piecewise inverse-weighting argument for asymptotic theory, and the partial identification and sensitivity analysis framework are all substantive methodological contributions. The ACTG 175 reanalysis is a concrete demonstration. Reproducible simulation code is provided as supplementary material.

major comments (1)
  1. Assumption 2 (Regularity on noises, Section 3.3) requires that the noise epsilon(T) is independent of X (not merely mean-independent). This condition is load-bearing for the estimation strategy, not just the asymptotic theory: the matching estimator in Algorithm 1 matches subjects on f_B(X) alone, and for the imputed outcome Y_{e_i}(B) to approximate a draw from F_{Y(B)|X_i}, the conditional distribution F_{Y(B)|X} must depend on X only through f_B(X). Under Assumption 2 this holds because Y(B) = f_B(X) + epsilon(B) with epsilon(B) independent of X. However, if the noise is heteroscedastic (Var(epsilon(B)|X) depends on X), two subjects with the same f_B(X) but different X have different F_{Y(B)|X}, and the matching imputation is invalid. The paper does not discuss this critical role of the independence (versus mean-independence) condition, nor does it provide heteroscedasticity DiC in or
minor comments (6)
  1. Section 4.2, Theorem 6: The Gaussian working model (Assumption 6) assumes constant correlation rho and constant variances sigma_A, sigma_B across X. The paper mentions a model check in the Appendix supporting plausibility for ACTG 175, but the main text does not describe what was checked or report results. A brief summary of the model check would strengthen the sensitivity analysis, given that the sensitivity parameters depend on this parametric structure.
  2. Section 2.1, Eq. (1): The classical evaluation framework is presented as testing whether E[Y(A+B)] exceeds max{E[Y(A)], E[Y(B)]}. The paper could more clearly acknowledge that some classical frameworks do account for subgroup heterogeneity, to better position the contribution.
  3. Algorithm 1, Step 1: The matching is described as optimal 1-to-1 matching minimizing sum of absolute differences in f_B_hat(X). The constraint that matching is without replacement is mentioned in Section 5 but not in Algorithm 1 itself. This should be specified in the algorithm for reproducibility.
  4. Table 1: The 'Classical' column reports coverage for tau, but the classical method targets eta, not tau. The caption could clarify that the classical interval is being evaluated as a (mis-targeted) interval for tau to illustrate the inadequacy.
  5. Section 6: The outcome model f_B_hat is estimated using a second-order power series on the didanosine arm without external data. Remark 1 notes that internal estimation is empirically robust based on Appendix simulations, but a brief note on how Assumption 3 (uniform convergence rate) is expected to hold in this application would be helpful.
  6. Figure 2: The three-region diagram is conceptually clear but could benefit from axis labels or a brief caption explaining what the horizontal axis represents (e.g., increasing combination effect).

Circularity Check

0 steps flagged · score 1.0 of 10

No circularity found; derivation chain is self-contained with explicitly stated assumptions

full rationale

The paper's derivation chain is genuinely self-contained. (1) The identification result (Theorem 1) follows from Assumption 1 (conditional Bliss independence) via standard probability: if Y(A) and Y(B) are conditionally independent given X, the joint distribution factors into the product of marginals, so E[max{Y(A),Y(B)}] = E_H[max{Y(A),Y(B)}]. This is a direct mathematical consequence, not a definitional equivalence. (2) The matching estimator (Algorithm 1) is a novel procedure whose asymptotic theory (Theorems 2-3) relies on a 'piecewise inverse-weighting argument' developed in the paper, not imported from self-citation. (3) Partial identification (Theorem 4) follows from the standard fact that positive dependence increases E[max{·,·}] relative to independence. (4) The Gaussian sensitivity analysis (Theorem 6) uses the well-known closed-form for the maximum of a bivariate normal. No load-bearing step reduces to its inputs by construction. The assumptions (Bliss independence, noise regularity, estimation rate, matching bias) are stated explicitly and are domain-motivated (pharmacology), not derived from the paper's own results. The ACTG 175 analysis applies the method to external public data without fitting-then-predicting the same quantity. The skeptic's concern about Assumption 2 (homoscedasticity) is a correctness/robustness issue, not circularity. Score 1 reflects minor self-citations to methodological literature that are not load-bearing for the central claims.

Assumptions & free parameters 1 free parameters · 6 assumptions · 1 invented entities

The axiom ledger is lean. The framework is model-free by design, with no free parameters fitted to data in the main estimation procedure. The key identification axiom (Assumption 1) is a domain assumption adapted from pharmacology. The sensitivity analysis introduces a Gaussian working model (Assumption 6) that is ad hoc to the paper but clearly labeled as such and not required for the main results.

free parameters (1)
  • None (model-free framework) = N/A
    The framework is explicitly model-free: no parametric form is imposed on f_T(X) or the joint distribution of potential outcomes. The outcome model f̂_B is a nonparametric estimator satisfying a rate condition (Assumption 3). The Gaussian working model in Section 4.2 is for sensitivity analysis only, not for the main estimator.
assumptions (6)
  • domain assumption Assumption 1 (Conditional Bliss Independence): Cov[I{Y(A)>t}, I{Y(B)>t} | X] = 0 for all t a.s.
    Invoked in Section 3.1 for point identification of τ (Theorem 1). Adapted from pharmacological Bliss independence. Satisfied under conditional cross-world independence of noise terms ϵ(A) ⊥ ϵ(B).
  • domain assumption Assumption 2 (Regularity on noises): Y_i(T_i) = f_{T_i}(X_i) + ϵ_i(T_i), where X ⊥ ϵ(T) and ϵ(T) is absolutely continuous with bounded density.
    Invoked in Section 3.3 for asymptotic normality (Theorems 2–3). Requires homoscedastic, independent noise structure. Stated as mild and satisfied by Gaussian/Laplace noise.
  • standard math Assumption 3 (Uniform convergence): ‖f̂_B − f_B‖_∞ = o_P(N_A^{-1/4})
    Invoked in Section 3.3. A nonparametric rate condition on the outcome model estimator. Stated as satisfiable by series estimators (Newey's estimator).
  • standard math Assumption 4 (Matching bias): E[|f̂_B(X_i) − f̂_B(X_{e_i})|^4] = o_P(N_A^{-1})
    Invoked in Section 3.3. Controls matching discrepancy. Stated as weaker than comparable conditions in Abadie and Spiess (2022). Sufficient conditions referenced from Abadie and Imbens (2012) Proposition 1.
  • domain assumption Assumption 5 (Non-negative dependence): Cov[I{Y(A)>t}, I{Y(B)>t} | X] ≥ 0 for all t a.s.
    Invoked in Section 4.1 for partial identification (Theorem 4). Relaxes Assumption 1 to allow cross-resistance. Stated as naturally satisfied under pharmacological cross-resistance scenarios.
  • ad hoc to paper Assumption 6 (Gaussianity): (Y(A), Y(B)) | X=x ~ N((f_A(x), f_B(x))^T, Σ) with constant correlation ρ
    Invoked in Section 4.2 for sensitivity analysis (Theorem 6). A parametric working model for the cross-world joint distribution. Used to derive closed-form sensitivity parameter ρ*. Not required for the main estimator.
invented entities (1)
  • None independent evidence
    purpose: N/A
    The paper introduces no new physical entities, particles, forces, or dimensions. All mathematical objects (potential outcomes, matching estimators, pseudo-oracle quantities) are standard causal inference constructs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluation of Combination Therapy amid Patient-Level Heterogeneity." pith.science (2026). https://pith.science/paper/RNHTXDAR

@misc{pith2026260708058,
  author       = {Pith},
  title        = {Pith review of: Evaluation of Combination Therapy amid Patient-Level Heterogeneity},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RNHTXDAR}},
  note         = {Machine review of arXiv:2607.08058}
}
abstract

Combination therapy, a treatment approach that involves two or more monotherapies, is widely considered to enhance therapeutic efficacy across different medical conditions. It was often believed that combination therapy is beneficial because of pharmacological interactions between its component monotherapies. However, through laboratory experiments, pharmacologists have recently noted that the benefits of some combination therapies might be largely driven by varying patient-level responses to their component monotherapies. Without accounting for such patient-level heterogeneity, classical statistical inference frameworks for combination therapy might be inadequate and overly optimistic. In this paper, we introduce a novel and model-free statistical inference framework to complement the classical one and evaluate combination therapy after adjusting for patient-level heterogeneity in responses to monotherapies. We address the non-identifiability and nonlinearity issue inherent in adjustment of patient-level heterogeneity and establish conditions for the (partial) identifiability of the cross-world target parameter. We develop an outcome-based optimal matching scheme to achieve asymptotic normality and construct $\sqrt{N}$-rate confidence intervals for the target parameter, thereby enabling reliable, efficient and transparent evaluation of combination therapy amid patient-level heterogeneity. The benefits of the proposed framework are demonstrated through a reanalysis of the ACTG 175 trial.

Figures

Figures reproduced from arXiv: 2607.08058 by the authors.

Figure 1
Figure 1. Illustration of patient-level heterogeneity. The solid horizontal line indicates the [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. It is in this sense that our proposed test in Eq. (3) complements, though not [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Sensitivity analysis for the ACTG 175 trial: the sensitivity parameters for the [PITH_FULL_IMAGE:figures/full_fig_p029_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    and Imbens, G

    Abadie, A. and Imbens, G. W. (2006). Large sample properties of matching estimators for average treatment effects.Econometrica, 74(1):235–267

    work page 2006

  2. [2]

    and Imbens, G

    Abadie, A. and Imbens, G. W. (2011). Bias-corrected matching estimators for average treatment effects.Journal of Business & Economic Statistics, 29(1):1–11

    work page 2011

  3. [3]

    and Imbens, G

    Abadie, A. and Imbens, G. W. (2012). A martingale representation for matching estimators. Journal of the American Statistical Association, 107(498):833–843

    work page 2012

  4. [4]

    and Imbens, G

    Abadie, A. and Imbens, G. W. (2016). Matching on the estimated propensity score.Econo- metrica, 84(2):781–807

    work page 2016

  5. [5]

    and Spiess, J

    Abadie, A. and Spiess, J. (2022). Robust post-matching inference.Journal of the American Statistical Association, 117(538):983–995

    work page 2022

  6. [6]

    Antonelli, J., Cefalu, M., Palmer, N., and Agniel, D. (2018). Doubly robust matching estimators for high dimensional confounding adjustment.Biometrics, 74(4):1171–1179

    work page 2018

  7. [7]

    Bliss, C. I. (1939). The toxicity of poisons applied jointly 1.Annals of Applied Biology, 26(3):585–615

    work page 1939

  8. [8]

    Bodik, J., Huang, Y., and Yu, B. (2025). Cross-world assumption and refining prediction intervals for individual treatment effects.arXiv preprint arXiv:2507.12581

    work page arXiv 2025

Show all 42 references
  1. [9]

    Chen, C., Liu, F., Ren, Y., Suttner, L., Sun, Z., Shentu, Y., and Schmidt, E. V. (2020). 31 Independent drug action and its statistical implications for development of combination therapies.Contemporary Clinical Trials, 98:106126

    2020

  2. [10]

    Y., Rai, M., Tadikonda, Y., Roy, P., Nollner, D

    Chen, E. Y., Rai, M., Tadikonda, Y., Roy, P., Nollner, D. W., Chitkara, A., Hamilton, J., and Thawani, R. (2025). Trends in complexity of single-agent and combination therapies for solid tumor cancers approved by the us food and drug administration.The oncologist, 30(3):oyae302

    2025

  3. [11]

    Chernozhukov, V., W¨ uthrich, K., and Zhu, Y. (2023). Toward personalized inference on individual treatment effects.Proceedings of the National Academy of Sciences, 120(7):e2300458120

    2023

  4. [12]

    F., and Bartoszewski, R

    Czechowicz, P., Wiech-Walow, A., Slawski, J., Collawn, J. F., and Bartoszewski, R. (2025). Old drugs, new challenges: reassigning drugs for cancer therapies.Cellular & Molecular Biology Letters, 30(1):27

    2025

  5. [13]

    and Miller, T

    Demidenko, E. and Miller, T. W. (2019). Statistical determination of synergy based on bliss definition of drugs independence.PLoS One, 14(11):e0224137

    2019

  6. [14]

    Deng, H., Han, Q., and Zhang, C.-H. (2021). Confidence intervals for multiple isotonic regression and other monotone models.The Annals of Statistics, 49(4):2021–2052

    2021

  7. [15]

    Deresa, N. W. and Van Keilegom, I. (2021). On semiparametric modelling, estimation and inference for survival data subject to dependent censoring.Biometrika, 108(4):965–979

    2021

  8. [16]

    and Vale, N

    Duarte, D. and Vale, N. (2022). Evaluation of synergism in drug combinations and reference models for future orientations in oncology.Current Research in Pharmacology and Drug Discovery, 3:100110

    2022

  9. [17]

    Ferman, B. (2021). Matching estimators with few treated and many control observations. Journal of Econometrics, 225(2):295–307. 32

    2021

  10. [18]

    and Korn, E

    Freidlin, B. and Korn, E. L. (2023). Augmenting randomized clinical trial data with historical control data: precision medicine applications.JNCI: Journal of the National Cancer Institute, 115(1):14–20

    2023

  11. [19]

    and He, X

    Guo, X. and He, X. (2021). Inference on selected subgroups in clinical trials.Journal of the American Statistical Association, 116(535):1498–1506

    2021

  12. [20]

    H., Henry, W

    Haubrich, R. H., Henry, W. K., Lederman, M. M., Phair, J. P., Niu, M., et al. (1996). A trial comparing nucleoside monotherapy with combination therapy in hiv-infected adults with cd4 cell counts from 200 to 500 per cubic millimeter.New England Journal of Medicine, 335(15):1081–1090

    1996

  13. [21]

    Han, Q. (2023). Noisy linear inverse problems under convex constraints: Exact risk asymp- totics in high dimensions.The Annals of Statistics, 51(4):1611–1638

    2023

  14. [22]

    Hansen, B. B. (2008). The prognostic analogue of the propensity score.Biometrika, 95(2):481–488

    2008

  15. [23]

    C., Dai, A., Plana, D., and Palmer, A

    Hwangbo, H., Patterson, S. C., Dai, A., Plana, D., and Palmer, A. C. (2023). Additivity predicts the efficacy of most approved combination therapies for advanced cancer.Nature Cancer, 4(12):1693–1704

    2023

  16. [24]

    and Cand` es, E

    Lei, L. and Cand` es, E. J. (2021). Conformal inference of counterfactuals and individual treatment effects.Journal of the Royal Statistical Society Series B: Statistical Methodology, 83(5):911–938

    2021

  17. [25]

    Lin, Z., Ding, P., and Han, F. (2023). Estimation based on nearest neighbor matching: from density ratio to average treatment effect.Econometrica, 91(6):2187–2217

    2023

  18. [26]

    Luedtke, A. R. and Van Der Laan, M. J. (2016). Statistical inference for the mean outcome under a possibly non-unique optimal treatment strategy.Annals of Statistics, 44(2):713. 33

    2016

  19. [27]

    Manosuthi, W., Wiboonchutikul, S., and Sungkanuparph, S. (2016). Integrated therapy for hiv and tuberculosis.AIDS Research and Therapy, 13(1):22

    2016

  20. [28]

    W., Bell, D., Bonow, R

    Nesto, R. W., Bell, D., Bonow, R. O., Fonseca, V., Grundy, S. M., Horton, E. S., Le Winter, M., Porte, D., Semenkovich, C. F., Smith, S., et al. (2003). Thiazolidinedione use, fluid retention, and congestive heart failure: a consensus statement from the american heart associat...

    2003

  21. [29]

    Palmer, A. C. and Sorger, P. K. (2017). Combination cancer therapy can confer benefit via patient-to-patient variability without drug additivity or synergy.Cell, 171(7):1678–1691

    2017

  22. [30]

    Perini, P., Calabrese, M., Rinaldi, L., and Gallo, P. (2008). Cyclophosphamide-based com- bination therapies for autoimmunity.Neurological Sciences, 29(Suppl 2):233–234

    2008

  23. [31]

    C., and Sorger, P

    Plana, D., Palmer, A. C., and Sorger, P. K. (2022). Independent drug action in combination therapy: implications for precision oncology.Cancer Discovery, 12(3):606–624

    2022

  24. [32]

    Rosenbaum, P. R. (2010).Design of Observational Studies. Springer

    2010

  25. [33]

    Rubin, D. B. (1986). Statistical matching using file concatenation with adjusted weights and multiple imputations.Journal of Business & Economic Statistics, 4(1):87–94

    1986

  26. [34]

    Sackett, D. L. (1989). Rules of evidence and clinical recommendations on the use of an- tithrombotic agents.Chest, 95(2):2S–4S

    1989

  27. [35]

    V., Chisamore, M

    Schmidt, E. V., Chisamore, M. J., Chaney, M. F., Maradeo, M. E., Anderson, J., Baltus, G. A., Pinheiro, E. M., and Uebele, V. N. (2020). Assessment of clinical activity of pd- 1 checkpoint inhibitor combination therapies reported in clinical trials.JAMA Network Open, 3(2):e192...

    2020

  28. [36]

    V., Sun, L

    Schmidt, E. V., Sun, L. Z., Palmer, A. C., and Chen, C. (2023). Rationales for combining therapies to treat cancer: independent action, response correlation, and collateral sensi- tivity versus synergy.Annual Review of Cancer Biology, 7(1):247–263. 34

    2023

  29. [37]

    Stuart, E. A. (2010). Matching methods for causal inference: A review and a look forward. Statistical Science, 25(1):1–21. US Food and Drug Administration (2013). Codevelopment of two or more new investigational drugs for use in combination.Guidance for Industry, US Department...

    2010

  30. [38]

    Wei, Q., Li, P., Yang, T., Zhu, J., Sun, L., Zhang, Z., Wang, L., Tian, X., Chen, J., Hu, C., et al. (2024). The promise and challenges of combination therapies with antibody-drug conjugates in solid tumors.Journal of Hematology & Oncology, 17(1):1

    2024

  31. [39]

    Wu, P., Ding, P., Geng, Z., and Liu, Y. (2024a). Quantifying individual risk for binary outcome.arXiv preprint arXiv:2402.10537

    arXiv

  32. [40]

    and Mao, X

    Wu, P. and Mao, X. (2025). The promises of multiple experiments: Identifying joint distri- bution of potential outcomes.arXiv preprint arXiv:2504.20470

    2025 arXiv

  33. [41]

    and Guo, X

    Xu, S. and Guo, X. (2025). Optimal inference of the mean outcome under optimal treatment regime.arXiv preprint arXiv:2509.09773

    2025 arXiv

  34. [42]

    and Zhang, Y

    Yang, S. and Zhang, Y. (2023). Multiply robust matching estimators of average and quantile treatment effects.Scandinavian Journal of Statistics, 50(1):235–265. 35

    2023

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.