Pith. sign in

REVIEW 4 major objections 5 minor 42 references

The paper's vector-valued Lorenz surface, paired with conditional means, uniquely determines the bivariate distribution.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 11:44 UTC pith:4OUUXDFV

load-bearing objection New directional Lorenz surface with a real boundary flaw: the core construction is worth refereeing, but Theorem 3.1(v) is false and Example 2.1's closed form is wrong. the 4 major comments →

arxiv 2607.19780 v1 pith:4OUUXDFV submitted 2026-07-22 stat.OT

A Conditional Quantile Approach to Vector-Valued Bivariate Lorenz Surfaces: Properties and Applications

Shifna P R , S.M. Sunoj This is my paper
classification stat.OT MSC 62H99
keywords Lorenz curvebivariate quantile functionconditional quantileGini indexdirectional inequalitynonparametric estimationbivariate distribution
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper introduces a vector-valued bivariate Lorenz surface (VBLS) built from two conditional Lorenz curves: L12 measures inequality of X1 within the subpopulation whose X2 exceeds its upper quantile, and L21 does the reverse. The central claim is that the pair of curves, together with the two conditional mean functions, uniquely characterizes the full joint distribution of (X1, X2). If true, this gives a directional inequality summary that captures dependence structure, not just marginal concentration, and it can be consistently estimated from data. The paper also defines a corresponding egalitarian surface and component-wise Gini indices that quantify asymmetry in concentration between the two variables. Applications to household expenditure and workers' compensation data illustrate how the two components expose directional patterns that symmetric multivariate Lorenz surfaces miss.

Core claim

On the paper's own terms, the discovery is Definition 2.1/2.2: a vector-valued bivariate Lorenz surface L(u1,u2)=(L12(u1,u2), L21(u1,u2)), where each component is the ordinary univariate Lorenz curve of a conditional distribution—X1 given X2 exceeding its u2-quantile, and X2 given X1 exceeding its u1-quantile. Theorem 5.1 then shows that if two non-negative random vectors have identical VBLS components and identical conditional mean functions, they have the same joint distribution. In other words, the VBLS plus conditional means contains the complete conditional quantile structure of the underlying distribution, via the recovery formulas Q12(u1,u2)=μ12(u2)∂L12/∂u1 and Q21(u1,u2)=μ21(u1)∂L21/

What carries the argument

The central object is the bivariate conditional quantile representation (Q1(u1), Q21(u1,u2)), equivalently (Q12(u1,u2), Q2(u2)), inherited from earlier work on bivariate quantile functions. The VBLS is defined through integrals of the conditional quantile functions Q12 and Q21, and the characterization theorem recovers those quantile functions from the Lorenz surfaces by differentiation: Q12=μ12 ∂L12/∂u1. The argument's load-bearing step is that this conditional quantile pair uniquely characterizes an absolutely continuous bivariate distribution; the paper imports that result and builds the uniqueness and estimation theorems on it.

Load-bearing premise

The argument assumes that a bivariate distribution is fully pinned down by its two conditional quantile functions; if that uniqueness fails for some continuous distributions, the paper's characterization and consistency results do not follow.

What would settle it

Find (or simulate) two absolutely continuous bivariate distributions with the same conditional quantile functions Q12(u1,u2) and Q21(u1,u2) for all u1,u2 in [0,1) but different joint distributions; or exhibit two distributions with identical VBLS components and identical conditional mean functions yet different laws. Either example would refute Theorem 5.1 directly.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the characterization holds, a two-curve summary replaces the full bivariate distribution for inequality comparison: equal VBLS and equal conditional means imply equal distributions.
  • The component-wise Gini indices G12 and G21 provide a directional asymmetry measure: G12>G21 means inequality in the first variable within upper-tail subpopulations of the second exceeds the reverse pattern.
  • The nonparametric estimator converges almost surely to the true VBLS, giving practitioners a consistent, distribution-free tool for directional inequality analysis.
  • Applications to expenditure and insurance data show the VBLS can reveal concentration patterns (e.g., medical expenditure more concentrated than total expenditure) that symmetric Lorenz surfaces hide.

Where Pith is reading between the lines

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

  • The recovery formulas suggest that any distributional quantity expressible through conditional quantiles—such as conditional tail expectations or risk measures—can be read off from the VBLS plus conditional means, extending the paper's characterization beyond its stated scope.
  • A natural testable extension is to define an ordering (VBLS dominance in both components for all u1,u2) and check whether it implies known multivariate stochastic orders, connecting the surface to established dependence orderings.
  • The two-direction asymmetry (G12 versus G21) could serve as a diagnostic for model misspecification in regression or copula fits, since a misspecified conditional dependence structure would distort the two components differently.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces a vector-valued bivariate Lorenz surface (VBLS), L(u1,u2)=(L12(u1,u2),L21(u1,u2)), defined by integrating conditional quantiles Q12(p,u2)=Q(X1 | X2>Q2(u2)) and Q21 analogously, normalized by conditional means. It derives boundary, monotonicity, convexity, scale, location, and independence properties; defines an egalitarian surface and component-wise Gini indices; states characterization theorems, including a result that equality of the VBLS plus conditional means implies equality in distribution; proposes a nonparametric empirical estimator; and illustrates the method on simulated Pareto data and two real datasets (Vietnam expenditure and workers compensation).

Significance. If corrected, the VBLS is a straightforward but useful directional alternative to symmetric bivariate Lorenz surfaces. The quantile definition is transparent, the estimator is natural, and the real-data applications show the directional information (asymmetry between L12 and L21) that marginal or symmetric methods miss. The characterization theorem is true essentially by construction—differentiating L12 recovers Q12/mu12—so its depth is limited, but it is a clean inversion property. The main weaknesses are rigor in several load-bearing places: a false boundary theorem, an incorrect closed-form formula, and an incomplete consistency proof. The central definition and most properties appear sound; with corrections this could be a solid contribution to quantile-based multivariate inequality measurement.

major comments (4)
  1. [Theorem 3.1(v)] The two-variable boundary claim in Theorem 3.1(v) is false. Let Y,W be iid Uniform(0,1), X2=1-Y, X1=W^(1/Y). Then (X1,X2) is absolutely continuous on (0,1)^2 with finite mean 1-ln(2). Conditional on X2>1-delta (i.e. Y<delta), the CDF is F_delta(t)=(t^delta-1)/(delta log t), mean mu_delta=1-log(1+delta)/delta, and the u-quantile is Q_delta(u)=exp(-x(u)/delta), where (1-e^{-x})/x=u. On the path u1=1-sqrt(delta), u2=1-delta, x~2sqrt(delta), so L12<=Q_delta/mu_delta ~ (2/delta)exp(-2/sqrt(delta))->0 while (u1,u2)->(1,1). Thus the double limit is path-dependent. The proof of (v) also moves from the one-variable limit in (iv) to the double limit without justification. Correct or replace this property.
  2. [Example 2.1] The closed-form L12 in Example 2.1 does not agree with Definition 2.2. For c=3 and u1=u2=0.3, the displayed formula gives about 0.2188, whereas integrating the stated Q12(p,u2) with the stated mu12 gives about 0.2049. The discrepancy indicates an algebraic error in the formula. Since this example is used for Figures 1-2 and motivates the simulation setting, the correct expression should be provided.
  3. [Theorem 6.1 proof] The proof of Theorem 6.1 is not a complete proof of a.s. consistency. It asserts that uniform convergence of distribution functions implies convergence of the corresponding conditional quantile functions, and then applies dominated convergence, but neither step is justified for the empirical conditional distribution with random denominator and estimated threshold hat Q2(u2). In particular, uniform a.s. convergence of hat F12(.|u2) to F12(.|u2) needs a Glivenko-Cantelli argument for data-dependent sets, and convergence of the interpolated empirical Lorenz integral needs a uniform quantile-process result, not just pointwise convergence. Please supply a rigorous argument for fixed (u1,u2) in (0,1)^2, or state and prove the required empirical process theorem.
  4. [Theorem 5.1, final step] The characterization theorem depends on the unproved assertion that a bivariate distribution is uniquely determined by the pair (Q1,Q21) (or equivalently (Q12,Q2)). The final sentence of the proof imports this from Vineshkumar and Nair without stating the exact conditions. This step is load-bearing: without it, equality of the two conditional quantile functions does not imply equality of the joint laws. Please include the uniqueness result as a lemma with assumptions (e.g., absolute continuity) and a proof, or at least a precise statement with all conditions.
minor comments (5)
  1. [Section 2, notation] The display defining Q21 as inf{x2:F21(x2,Q1(u1))>=u2} is ambiguous; it should use conditional distribution notation such as F_{2|1}(x2 | X1>Q1(u1)) or an explicit conditioning statement.
  2. [Abstract and Section 5] The abstract says the VBLS itself is unique, but Theorem 5.1 requires both the VBLS and the conditional mean functions. Please state this qualification clearly.
  3. [Theorem 4.1 proof] The text says Q2(0) is the lower endpoint of the support. For a non-negative absolutely continuous X2, Q2(0)=0 and the event {X2>0} has probability one; the proof should be rephrased accordingly.
  4. [Throughout] State explicitly that mu12(u2)>0 and mu21(u1)>0 for all arguments in the domain; otherwise the denominators in Definition 2.2 are undefined.
  5. [Theorem 6.1] The estimator uses hat Q2(u2)=X_{2(ceil(nu2))}, which can equal the sample maximum and give an empty subsample for finite n. Please specify the treatment of boundary cases or state that fixed u2 in (0,1) and n large enough are assumed.

Circularity Check

0 steps flagged

No circular derivation: the VBLS is defined directly from conditional quantiles, and the main characterization theorem inverts the defining integral rather than assuming its conclusion.

full rationale

I traced the main derivation chain. Definition 2.2 defines L12(u1,u2)=1/mu12(u2) * integral_0^{u1} Q12(p,u2) dp, and symmetrically for L21. The bulk of Sections 3--4 (monotonicity, convexity, scale invariance, location-shift behavior, egalitarian characterization, Gini bounds) are direct analytic consequences of this definition, not circular steps. Theorem 5.1 differentiates L12 to recover Q12(u1,u2)=mu12(u2)*partial L12/partial u1, uses the assumed equality of L and mu to conclude equality of conditional quantiles, and then invokes the bivariate-quantile uniqueness representation. This is an algebraic inversion of the defining integral, not a prediction that was already fitted or an assumption smuggled in as a result. The uniqueness representation is imported from prior work (Vineshkumar & Nair 2019; Nair & Vineshkumar 2021) and is not itself derived by a self-citation chain from this paper. The Section 6 estimator is a plug-in empirical conditional quantile/mean estimator, so there is no fitted-parameter-called-prediction circularity; the simulation and applications use independent data. I did identify two non-circular rigor concerns: the proof of Theorem 3.1(v) reduces a two-dimensional limit to a one-dimensional limit without the needed uniformity and is invalid as written (indeed a counterexample can make L12 fail to converge to 1 along a path to (1,1)), and Theorem 6.1 asserts uniform a.s. convergence of conditional quantile processes with an estimated threshold without a full proof. These concern correctness/rigor, not circularity, and do not affect the score.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

The paper's central construction introduces no fitted parameters; the theory relies on the imported bivariate-quantile representation from the authors' own previous work and on standard measure-theoretic results. No new physical or empirical entities are postulated; the VBLS is a mathematical definition.

axioms (4)
  • domain assumption Bivariate quantile functions (Q1(u1), Q21(u1,u2)) (or symmetrically (Q12,Q2)) provide a unique characterization of absolutely continuous bivariate distributions.
    Invoked in Section 2 to build the VBLS and in Theorem 5.1 to conclude equality of joint laws from equality of conditional quantiles; imported from the authors' own prior work (Vineshkumar & Nair 2019; Nair & Vineshkumar 2021) without proof in this paper.
  • standard math For a non-negative absolutely continuous random variable X with finite mean, the ordinary Lorenz curve L(u) = (1/μ)∫_0^u Q(p)dp is convex, lies below the egalitarian line, and has the standard boundary and Gini properties.
    Used throughout as background (Section 1, Theorems 3.1-4.2).
  • domain assumption Conditional distributions of X1 given {X2 > Q2(u2)} are absolutely continuous with finite means for u2∈[0,1).
    Needed for the conditional quantile functions Q12 and the conditional means μ12 to be well-defined and differentiable; stated in Section 2.
  • standard math Dominated convergence theorem and Glivenko-Cantelli theorem.
    Used in proofs of Theorem 3.1 and Theorem 6.1.

pith-pipeline@v1.3.0-alltime-deepseek · 18838 in / 21597 out tokens · 185918 ms · 2026-08-01T11:44:26.638828+00:00 · methodology

0 comments
read the original abstract

The Lorenz curve is a fundamental tool for measuring inequality, but its extension to multivariate settings remains challenging due to the complex dependence structure among variables and the need to capture directional aspects of inequality. In this paper, we introduce a novel vector-valued bivariate Lorenz surface (VBLS) based on conditional distributions and conditional quantile functions. Unlike existing symmetric bivariate Lorenz surfaces, the proposed VBLS effectively captures the directional inequality arising from the conditional dependence between two variables. We establish several fundamental properties of the proposed surface and investigate its mathematical properties. The corresponding egalitarian surface is defined, leading to the development of associated vector-valued bivariate Gini measures for quantifying inequality. We further derive characterization results that demonstrate the uniqueness of the proposed VBLS within the underlying distributional framework. Nonparametric estimators of the VBLS are developed and their finite-sample performance is evaluated through a simulation study. The usefulness of the proposed methodology is also illustrated with applications to income inequality and actuarial data.

Figures

Figures reproduced from arXiv: 2607.19780 by Shifna P R, S.M. Sunoj.

Figure 1
Figure 1. Figure 1: Plot of L12(u1, u2) for bivariate Pareto distribution [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Plot of L21(u1, u2) for bivariate Pareto distribution framework accommodates a wide class of random variables, including discrete, continu￾ous, and mixed types, making it suitable for applications in higher-dimensional settings. Applying the transformation p = F12 in (2.1), we have x1 = Q12(p, u2), dF12(x1; u2) = dp. Since F12(0; u2) = 0 and F12(Q12(u1, u2); u2) = u1, it follows that L12(u1, u2) = 1 µ12(u2… view at source ↗
Figure 3
Figure 3. Figure 3: Graphical representation for L12(u1, u2) for bivariate linear hazard quantile model [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Graphical representation for L21(u1, u2) for bivariate linear hazard quantile model To verify the equivalence of the distribution based and quantile based formulations of the VBLS, we illustrate using the following example the consistency through a bivariate model for which both the distribution functions and the corresponding quantile functions are available in closed form. 7 [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figure 5
Figure 5. Figure 5: Estimated VBLS surface Lˆ 12(u1, u2) for the Vietnam expenditure data [PITH_FULL_IMAGE:figures/full_fig_p025_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Estimated VBLS surface Lˆ 21(u1, u2) for the Vietnam expenditure data bivariate Gini indices are G12 = 0.5215 and G21 = 0.8488. Since G12 < G21, the av￾erage level of the surface L12 exceeds that of L21, indicating that inequality associated with medical expenditure is substantially greater than inequality associated with total household expenditure when the dependence structure between the variables is ta… view at source ↗
Figure 7
Figure 7. Figure 7: Diagonal comparison of the estimated VBLSs for the Vietnam expenditure data [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Estimated VBLS surface Lˆ 12(u1, u2) for workers compensation data [PITH_FULL_IMAGE:figures/full_fig_p026_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Estimated VBLS surface Lˆ 21(u1, u2) for workers compensation data bivariate Gini indices were estimated as G12 = 0.7750 and G21 = 0.6737. The large values 26 [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Slice plots of Lˆ 12(u1, u2) for workers compensation data [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Slice plots of Lˆ 21(u1, u2) for workers compensation data of both indices indicate substantial concentration in the joint distribution of premiums and losses. From an actuarial perspective, the larger G12 compared to G21 indicates that premium inequality is more sensitive to tail events than loss inequality. Insurers may need to adjust risk loadings for high premium contracts to better reflect the observ… view at source ↗

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

42 extracted references · 1 linked inside Pith

  1. [1]

    Publications of the American Statistical Association , volume=

    Methods of measuring the concentration of wealth , author=. Publications of the American Statistical Association , volume=. 1905 , publisher=

    1905

  2. [2]

    A general definition of the

    Gastwirth, Joseph L , journal=. A general definition of the. 1971 , publisher=

    1971

  3. [3]

    The Journal of Economic Inequality , volume=

    Multidimensional deprivation: contrasting social welfare and counting approaches , author=. The Journal of Economic Inequality , volume=. 2003 , publisher=

    2003

  4. [4]

    The Review of Economic Studies , volume=

    The comparison of multi-dimensioned distributions of economic status , author=. The Review of Economic Studies , volume=. 1982 , publisher=

    1982

  5. [5]

    The Quarterly Journal of Economics , volume=

    Multidimensional egalitarianisms , author=. The Quarterly Journal of Economics , volume=. 1977 , publisher=

    1977

  6. [6]

    Econometrica: Journal of the Econometric Society , pages=

    The measurement and decomposition of multi-dimensional inequality , author=. Econometrica: Journal of the Econometric Society , pages=. 1986 , publisher=

    1986

  7. [7]

    Journal of Business & Economic Statistics , volume=

    Relative price changes and inequality in the size distribution of various components of income: A multidimensional approach , author=. Journal of Business & Economic Statistics , volume=. 1987 , publisher=

    1987

  8. [8]

    Journal of Economic Theory , volume=

    Multidimensional generalizations of the relative and absolute inequality indices: the Atkinson-Kolm-Sen approach , author=. Journal of Economic Theory , volume=. 1995 , publisher=

    1995

  9. [9]

    Social Choice and Welfare , volume=

    Multidimensional inequality and multidimensional generalized entropy measures: An axiomatic derivation , author=. Social Choice and Welfare , volume=. 1999 , publisher=

    1999

  10. [10]

    On the two-dimensional concentration surface and extensions of concentration coefficient and

    Taguchi, Tokio , journal=. On the two-dimensional concentration surface and extensions of concentration coefficient and. 1972 , publisher=

    1972

  11. [11]

    Arnold, Barry C , journal=. Pareto. 1983 , publisher=

    1983

  12. [12]

    Majorization and the

    Arnold, Barry C , note=. Majorization and the. 1987 , publisher=

    1987

  13. [13]

    Koshevoy, Gleb and Mosler, Karl , journal=. The. 1996 , publisher=

    1996

  14. [14]

    Statistica , volume=

    Bivariate quantile functions and their applications to reliability modelling , author=. Statistica , volume=

  15. [15]

    American Journal of Mathematical and Management Sciences , volume=

    Properties of bivariate distributions represented through quantile functions , author=. American Journal of Mathematical and Management Sciences , volume=. 2021 , publisher=

    2021

  16. [16]

    Quantile-based

    Nair, N Unnikrishnan and Sankaran, PG and Balakrishnan, Narayanaswamy , year=. Quantile-based

  17. [17]

    Multivariate

    Shifna,P R and Nair, N Unnikrishnan and Sunoj, S M , journal=. Multivariate. 2024 , publisher=

    2024

  18. [18]

    2004 , publisher=

    Convex optimization , author=. 2004 , publisher=

    2004

  19. [19]

    Modeling bivariate

    Sarabia, Jos. Modeling bivariate. Fifth Meeting of ECINEQ at Bari, Italy , year=

  20. [20]

    Leimkuhler, Ferdinand F , journal=. The. 1967 , publisher=

    1967

  21. [21]

    Nair, N Unnikrishnan and Sankaran, P G and Vineshkumar, B , journal=. The. 2012 , publisher=

    2012

  22. [22]

    A multivariate extension of the

    Grothe, Oliver and K. A multivariate extension of the. The Journal of Economic Inequality , volume=. 2022 , publisher=

    2022

  23. [23]

    arXiv preprint arXiv:2203.09000 , year=

    Lorenz map, inequality ordering and curves based on multidimensional rearrangements , author=. arXiv preprint arXiv:2203.09000 , year=

    Pith/arXiv arXiv

  24. [24]

    Lorenz surfaces based on the

    Sarabia, Jos. Lorenz surfaces based on the. Mathematics , volume=. 2020 , publisher=

    2020

  25. [25]

    1999 , publisher=

    Multidimensional inequality measurement: A proposal , author=. 1999 , publisher=

    1999

  26. [26]

    Economica , volume=

    Inequality of wellbeing: A multidimensional approach , author=. Economica , volume=. 2012 , publisher=

    2012

  27. [27]

    Multidimensional

    Banerjee, Asis Kumar , journal=. Multidimensional

  28. [28]

    , author=

    Contributions to the theory of convex bodies. , author=. Michigan Mathematical Journal , volume=. 1957 , publisher=

    1957

  29. [29]

    The Annals of Mathematical Statistics , volume=

    Remarks on a multivariate transformation , author=. The Annals of Mathematical Statistics , volume=. 1952 , publisher=

    1952

  30. [30]

    On Iterated

    Ignatov, Zvetan and Yordanov, Vilimir , journal=. On Iterated

  31. [31]

    On Iterated

    Yordanov, Vilimir , journal=. On Iterated

  32. [32]

    Multiple-Attribute

    Hallin, Marc and Mordant, Gilles , journal=. Multiple-Attribute. 2025 , publisher=

    2025

  33. [33]

    Model selection based on

    Denuit, Michel and Sznajder, Dominik and Trufin, Julien , journal=. Model selection based on. 2019 , publisher=

    2019

  34. [34]

    Summarizing insurance scores using a

    Frees, Edward W and Meyers, Glenn and Cummings, A David , journal=. Summarizing insurance scores using a. 2011 , publisher=

    2011

  35. [35]

    Insurance ratemaking and a

    Frees, Edward W and Meyers, Glenn and Cummings, A David , journal=. Insurance ratemaking and a. 2014 , publisher=

    2014

  36. [36]

    Christopoulos, Katerina A and Hartogensis, Wendy and Glidden, David V and Pilcher, Christopher D and Gandhi, Monica and Geng, Elvin H , journal=. The. 2017 , publisher=

    2017

  37. [37]

    Using the

    Mauguen, Audrey and Begg, Colin B , journal=. Using the. 2016 , publisher=

    2016

  38. [38]

    Journal of Quantitative Criminology , volume=

    More places than crimes: Implications for evaluating the law of crime concentration at place , author=. Journal of Quantitative Criminology , volume=. 2017 , publisher=

    2017

  39. [39]

    European Journal of Applied Mathematics , volume=

    A brief history of the analysis of crime concentration , author=. European Journal of Applied Mathematics , volume=. 2010 , publisher=

    2010

  40. [40]

    Ecology , volume=

    Describing inequality in plant size or fecundity , author=. Ecology , volume=. 2000 , publisher=

    2000

  41. [41]

    Oecologia , volume=

    The meaning and measurement of size hierarchies in plant populations , author=. Oecologia , volume=. 1984 , publisher=

    1984

  42. [42]

    Communications in Statistics-Theory and Methods , volume=

    Bivariate total time on test transforms , author=. Communications in Statistics-Theory and Methods , volume=. 2025 , publisher=

    2025