REVIEW 3 major objections 5 minor 35 references
Gradient-Weighted, Data-Driven Normalization for Approximate Border Bases -- Concept and Computation
T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read Replacing coefficient normalization with a gradient-weighted semi-norm makes approximate border basis computation scale-invariant and robust to perturbations.
desk verdict Good idea and mostly clean math, but Theorem 6.3's proof has a normalization-factor error and Example 6.4 contradicts the theorem; the scale-invariance claim needs major revision. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the gradient-weighted semi-norm and the generalized eigenvalue problem it induces. For a term $t$, the norm is $\|t\|_{\nabla,X} = \|\nabla t(X)\|_2 / D(t)$ with $D(t) = \sqrt{\sum_i \deg_i(t)^2}$, and a polynomial's norm is the coefficient norm weighted by these term norms; the constant term has norm zero. In the ABM+GWN algorithm, the decision whether a trial term $b$ becomes a new inner term or generates an $\epsilon$-vanishing border polynomial is reduced to the generalized eigenvalue problem $M^\top M v = \lambda D^2 v$, where $M$ is the evaluation matrix of the current terms together with $b$ and $D$ is the diagonal matrix of their gradient-weighted norms. The minimal generalized eigenvalue $\lambda_{\min}$ equals the squared minimal evaluation among gradient-weighted-normalized candidates, and the acceptance test is $\lambda_{\min} \le \epsilon^2$. Because the term weights transform as $\|t\|_{\nabla,\alpha X} = \alpha^{\deg t - 1}\,\|t\|_{\nabla,X}$, the whole eigenproblem is equivariant under rescaling $X \mapsto \alpha X$, which is what produces the scale-invariance theorem.
What would settle it
A direct test: apply ABM+GWN to the six near-circle points of Example 6.4 with tolerance $\epsilon = 0.1$ and to the scaled set $0.1X$ with tolerance $0.01$. Theorem 6.3 predicts the same term set $O$ and coefficients related by $\alpha^{\deg t - 1}$; any run producing a different $O$ or coefficient relationship falsifies the scale-invariance claim. For the coefficient-normalization side, Theorem 7.1 predicts that whenever the computed term set $O$ has a term of degree at least 2, some scaling $\alpha$ with $\alpha^{\deg t - 1}\gamma < 1$ changes $O$; a search over $\alpha$ on a fixed point set either finds such an $\alpha$ or refutes that theorem.
Extended reading notes
Core claim
The paper's central discovery is that choosing a data-dependent norm for polynomials changes approximate border basis computation in a structural way. For a term $t$, the gradient-weighted norm is $\|t\|_{\nabla,X} = \|\nabla t(X)\|_2 / D(t)$, with $D(t) = \sqrt{\sum_i \deg_i(t)^2}$ the Euclidean degree, and for a polynomial it is the coefficient norm built from these term weights; the constant term has norm zero. The paper proves (Theorem 6.3) that running ABM+GWN on $X$ with tolerance $\epsilon$ and on $\alpha X$ with tolerance $\alpha\epsilon$ yields the same term set, $O_\alpha = O$, and border polynomials related by $g_{\alpha,i}(\alpha p) = \alpha g_i(p)$, so the coefficients transform as $c_{\alpha,i,t} = \alpha^{1 - \deg t} c_{i,t}$. By contrast, coefficient normalization can produce a different term set after scaling (Theorem 7.1). In addition, Proposition 6.2 bounds the effect of a small perturbation $\Delta$ on a gradient-weighted-normalized polynomial: $\|g(X+\Delta)\|_2 \le \epsilon + \deg(g)\,\|\Delta\|_{\max} + o(\|\Delta\|_{\max})$, so strict $\epsilon$-vanishing survives sufficiently small noise.
Load-bearing premise
The argument assumes that small evaluation values on the noisy sample points are a faithful proxy for small geometric distance between the points and the polynomial's zero set, and that perturbations are small enough for the first-order Taylor bound to dominate; for large perturbations the gradient-weighted norm gives no geometric guarantee.
Editorial extensions
If this is right
- With gradient-weighted normalization, rescaling the data by any positive $\alpha$ and the tolerance by the same $\alpha$ leaves the computed term set unchanged, and the border polynomials' coefficients change by the known factor $\alpha^{1 - \deg t}$.
- Adapting the ABM algorithm to gradient-weighted normalization preserves its $O(|O|^3)$ complexity, so the scale-invariance and perturbation guarantees come without an asymptotic computational penalty.
- A polynomial that is strictly $\epsilon$-vanishing on $X$ remains $\epsilon$-vanishing on every sufficiently small perturbation $X + \Delta$, with the first-order bound $\deg(g)\,\|\Delta\|_{\max}$ on the change in its evaluation norm.
- Coefficient-normalized approximate border bases are not scale-invariant: if the computed term set contains a term of degree at least 2, there is a scaling that changes the term set, making naive rescaling preprocessing unreliable.
- The paper's definition of an $\epsilon$-approximate border basis uses the same tolerance $\epsilon$ both to require that border polynomials $\epsilon$-vanish and to require that non-constant inner-term polynomials do not, eliminating the second tolerance parameter of the classical definition.
Reading between the lines
- The scale-invariance proof rests on the homogeneity of the term weights rather than on the Euclidean-gradient formula itself: any norm whose term weights scale as $\alpha^{\deg t - 1}$ under $X \mapsto \alpha X$ would inherit Theorem 6.3 with the same argument.
- Because the perturbation guarantee is only first-order, the practical advantage over coefficient normalization should be largest in low-noise regimes; in high-noise settings neither normalization certifies geometric proximity, so the method is best paired with preprocessing that keeps perturbations small.
- The generalized eigenvalue formulation makes normalization a tunable diagonal matrix $D$, so one could interpolate between coefficient normalization and gradient weighting, or design weights reflecting anisotropic noise in the coordinate directions, without leaving the same algorithmic framework.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a gradient-weighted, data-dependent semi-norm for normalizing polynomials in algorithms that compute approximate border bases of approximately vanishing ideals. It defines the semi-norm, formulates the constrained minimization as a generalized eigenvalue problem, adapts the Approximate Buchberger–Möller (ABM) algorithm to this normalization (ABM+GWN), and analyzes two advertised properties: robustness against point perturbations and invariance under scaling of the point set and tolerance. The paper also gives a counterexample showing that coefficient normalization lacks scaling invariance and reports numerical experiments on three varieties supporting the claimed advantages.
Significance. If the claims are established, the paper makes a useful contribution to approximate commutative algebra: it identifies a principled, data-dependent normalization that is easy to integrate into existing algorithms, and it provides a structural invariance property that is absent under coefficient normalization. The generalized eigenvalue formulation is a clean and transferable idea, and the numerical comparison with coefficient normalization is extensive. However, the significance is currently tempered by three technical problems: a normalization-factor error in the proof of the scaling theorem, an inconsistent numerical illustration of that theorem, and a false bound behind the perturbation-robustness claim.
major comments (3)
- [Section 6.2, proof of Theorem 6.3] The proof of Theorem 6.3 contains a normalization-factor error. After substituting M_alpha = M A and D_alpha = (1/alpha) D A, the scaled generalized eigenproblem gives M^T M (A v_alpha) = (lambda_alpha/alpha^2) D^T D (A v_alpha). Hence the unscaled eigenvector relation is v = A v_alpha, i.e. v_alpha = A^{-1} v, only up to the normalization constraint. But the scaled normalization condition is v_alpha^T D_alpha^T D_alpha v_alpha = 1, which becomes v_alpha^T A^T D^T D A v_alpha = alpha^2, not 1. Therefore the eigenvector satisfying both the eigenproblem and the scaled normalization is v_alpha = alpha A^{-1} v, not A^{-1} v. With this correction, the coefficient rule c_{alpha,t} = alpha^{1-deg t} c_t does follow, and the term-set invariance part (i) is unaffected, but the proof as written is incorrect.
- [Example 6.4] The printed scaled polynomials do not match the coefficient-scaling rule stated in Theorem 6.3 and in the surrounding text. For alpha = 0.1, the rule c_{alpha,t} = alpha^{1-deg t} c_t predicts that each quadratic coefficient of g1 is multiplied by 10, each linear coefficient is unchanged, and the constant is multiplied by 0.1. The printed g1,alpha instead leaves the quadratic coefficients unchanged, multiplies the linear coefficients by 0.1, and multiplies the constant by 0.01, i.e. it uses alpha^{deg t} scaling. The reported evaluation norm 0.0028 is consistent with the theorem's rule (0.1 times the original 0.028), not with the printed coefficients. The same discrepancy appears in g2,alpha. This is not merely a typographical issue: the numerical illustration is the only concrete demonstration of the coefficient-structure claim, and as printed it contradicts the theorem. The example should be recomputed or corrected.
- [Proposition 6.1] Proposition 6.1 is false as stated. Let X = {(1,1)} and g = x + x^2 in R[x]. Then deg(g) = 2, ||g||_{nabla,X} = sqrt(||x||^2 + ||x^2||^2) = sqrt(1 + 1) = sqrt(2), and ||nabla g(X)||_2 = ||(1+2x,0)|| at (1,1) = 3. Thus ||nabla g(X)||_2 = 3 > 2 sqrt(2) = deg(g) ||g||_{nabla,X}. The proof's key inequality replaces a sum of gradient vectors by a sum of squared coefficient-weighted gradient norms; this step is not generally valid because the gradients of different terms need not be orthogonal and because summing D(t)^2 over the support is not bounded by deg(g)^2. Since Proposition 6.2 uses Proposition 6.1 to bound the perturbation-induced change in algebraic distance, the stated robustness guarantee is currently unsupported. The bound may be repairable with an additional factor depending on the support size, but the proposition and Proposition 6.2 must be revised.
minor comments (5)
- [Throughout] There are repeated typos such as "Alhough" for "Although" and broken renderings of names such as "Gr" followed by "obner".
- [Proposition 4.4] In the Cauchy-Schwarz step, the second square-root factor is printed as sqrt(sum_{t != 1} ||t||^2_{nabla,X}); it should be sqrt(sum_{t != 1} g_t^2 ||t||^2_{nabla,X}).
- [Theorem 6.3] The displayed identity g_{\alpha,i}(\alpha p) = \alpha g_i(\alpha^{-1}(\alpha p)) is a tautology; the intended statement is that g_{\alpha,i}(\alpha p) = \alpha g_i(p) for all p, which is what the subsequent coefficient rule expresses.
- [Example 5.10] In step 2, the matrix denoted M_y^T M_x should be M_x^T M_x; the same notation is used correctly elsewhere.
- [Section 8, Tables 1 and 2] The text says that gradient-weighted normalization succeeded in all datasets and scales except (V2, alpha = 0.01), but Table 1 also shows zero success at (V1, alpha = 0.01) for gradient-weighted normalization. This discrepancy should be clarified.
Circularity Check
No significant circularity: the scale-invariance and robustness results are proved from the stated definitions rather than assumed or fitted.
full rationale
The paper's central claims are mathematical theorems proved from the stated definitions of the gradient-weighted semi-norm, the generalized eigenvalue formulation, and the ABM+GWN algorithm. No parameter is fitted and then renamed as a prediction, and no load-bearing conclusion is imported solely from a self-citation. The key scaling behavior ||t||_{∇,αX} = α^{deg t − 1} ||t||_{∇,X} follows directly from the definition of the semi-norm, and Theorem 6.3 derives term-set invariance and the coefficient transformation rule by substituting M_α = M A and D_α = (1/α) D A into the generalized eigenproblem, giving λ_α/α^2 = λ and the corresponding eigenvector relation. This is a proof from definitions, not an assumption of the conclusion. The robustness statement in Proposition 6.2 is an application of Taylor expansion plus Proposition 6.1, which is itself proved from the norm definition; the heuristic identification of algebraic and geometric distance in Sections 1.1 and 1.2 is explicitly labeled as a heuristic and not presented as a derivation. Related work by the same authors ([15], [19], [20]) is cited for background on gradient semi-norms and the spurious vanishing problem, but the results that carry the paper's claims are either proved here with full arguments or are standard facts about border bases cited from external sources. The numerical experiments run the algorithm on independently sampled perturbed point sets and compare against an external consistency criterion; the success rates are not fitted to the theorem being tested. A possible dropped 1/α^2 factor in the proof of Theorem 6.3 and the apparent mismatch in Example 6.4 would be correctness risks or proof errors, not circularity, because they would not make the claimed result equivalent to its inputs by construction. For these reasons, no circular step can be exhibited under the required standard, and the appropriate finding is no significant circularity.
Assumptions & free parameters
assumptions (6)
- standard math The evaluation map eval_X from the polynomial ring to R^m is a ring homomorphism and P/I(X) is isomorphic to R^m for a finite point set X.
- standard math Cauchy-Schwarz inequality and first-order Taylor expansion hold for polynomial functions on R^n.
- standard math For positive semi-definite matrices M^T M and D^2 with D diagonal, the generalized eigenvalue problem M^T M v = lambda D^2 v encodes the constrained minimization of ||g(X)||_2 subject to ||g||_nabla,X = 1.
- domain assumption Data points carry uncertainty, so approximately vanishing polynomials with respect to a tolerance are the right object, and algebraic distance is a heuristic proxy for geometric distance.
- domain assumption Perturbations are small enough that the o(||Delta||_max) term in Proposition 6.2 is negligible; the robustness claim is first-order.
- domain assumption Non-zero constant polynomials are treated as exceptional and never considered approximately vanishing on a non-empty point set.
Cite this review
Pith. "Pith review of Gradient-Weighted, Data-Driven Normalization for Approximate Border Bases -- Concept and Computation." pith.science (2026). https://pith.science/paper/MAPJLNJS
@misc{pith2026250609529,
author = {Pith},
title = {Pith review of: Gradient-Weighted, Data-Driven Normalization for Approximate Border Bases -- Concept and Computation},
year = {2026},
howpublished = {\url{https://pith.science/paper/MAPJLNJS}},
note = {Machine review of arXiv:2506.09529}
}
read the original abstract
This paper studies the concept and the computation of approximately vanishing ideals of a finite set of data points. By data points, we mean that the points contain some uncertainty, which is a key motivation for the approximate treatment. A careful review of the existing border basis concept for an exact treatment motivates a new adaptation of the border basis concept for an approximate treatment. In the study of approximately vanishing polynomials, the normalization of polynomials plays a vital role. So far, the most common normalization in computational commutative algebra uses the coefficient norm of a polynomial. Inspired by recent developments in machine learning, the present paper proposes and studies the use of gradient-weighted normalization. The gradient-weighted semi-norm evaluates the gradient of a polynomial at the data points. This data-driven nature of gradient-weighted normalization produces, on the one hand, better stability against perturbation and, on the other hand, very significantly, invariance of border bases with respect to scaling the data points. Neither property is achieved with coefficient normalization. In particular, we present an example of the lack of scaling invariance with respect to coefficient normalization, which can cause an approximate border basis computation to fail. This is extremely relevant because scaling of the point set is often recommended for preprocessing the data. Further, we use an existing algorithm with coefficient normalization to show that it is easily adapted to gradient-weighted normalization. The analysis of the adapted algorithm only requires tiny changes, and the time complexity remains the same. Finally, we present numerical experiments on three affine varieties to demonstrate the superior stability of our data-driven normalization over coefficient normalization. We obtain robustness to perturbations and invariance to scaling.
Figures
Reference graph
Works this paper leans on
-
[1]
Stable border bases for ideals of points
John Abbott, Claudia Fassino, and Maria-Laura Torrente. Stable border bases for ideals of points. Journal of Symbolic Computation , 43(12):883–894, 2008
work page 2008
-
[2]
Analytic 32 Table 1: Summary of the Scaling Consistency Test of 20 independent runs with 1 % noise
Rika Antonova, Maksim Maydanskiy, Danica Kragic, Sam Devlin, and Katja Hofmann. Analytic 32 Table 1: Summary of the Scaling Consistency Test of 20 independent runs with 1 % noise. Columns coeff. and grad. w. present the coefficient and gradient-weighted normalization, respectively. Column coeff. dist. presents the distance of the normalized coefficient ve...
arXiv 2006
-
[3]
Ideals, varieties, and algorithms , volume 3
David Cox, John Little, and Donal O’shea. Ideals, varieties, and algorithms , volume 3. Springer, 1992
work page 1992
-
[4]
Alicia Dickenstein and Joannis Z. Emiris, editors. Solving Polynomial Equations - Foundations, Algorithms, and Applications . Springer-Verlag, Berlin, 2005
work page 2005
-
[5]
Symboli-numeric methods for solving polynomial equa- tions
Mohamed Elkadi and Bernard Mourrain. Symboli-numeric methods for solving polynomial equa- tions. In Alicia Dickenstein and Joannis Z. Emiris, editors, Solving Polynomial Equations , pages 125–168. Springer-Verlag, Berlin, 2005
work page 2005
-
[6]
Almost vanishing polynomials for sets of limited precision points
Claudia Fassino. Almost vanishing polynomials for sets of limited precision points. Journal of Symbolic Computation, 45(1):19–37, 2010
work page 2010
-
[7]
Simple varieties for limited precision points
Claudia Fassino and Maria-Laura Torrente. Simple varieties for limited precision points. Theoretical Computer Science, 479:174–186, 2013
work page 2013
-
[8]
Approximate computation of zero-dimensional polynomial ideals
Daniel Heldt, Martin Kreuzer, Sebastian Pokutta, and Hennie Poulisse. Approximate computation of zero-dimensional polynomial ideals. Journal of Symbolic Computation , 44(11):1566–1591, 2009
work page 2009
Show all 35 references
-
[9]
Discriminative vanishing component analysis
Chenping Hou, Feiping Nie, and Dacheng Tao. Discriminative vanishing component analysis. In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence (AAAI) , pages 1666–1672, Palo Alto, California, 2016. AAAI Press
2016
-
[10]
Principal variety analysis
Reza Iraji and Hamidreza Chitsaz. Principal variety analysis. In Proceedings of the 1st Annual Conference on Robot Learning (ACRL), pages 97–108. PMLR, 2017
2017
-
[11]
Nepomuceno, Aleksandra Tutueva, and Denis Butusov
Artur Karimov, Erivelton G. Nepomuceno, Aleksandra Tutueva, and Denis Butusov. Algebraic method for the reconstruction of partially observed nonlinear systems using differential and integral embedding. Mathematics, 8(2):300–321, Feb 2020
2020
-
[12]
Identifying empirical equations of chaotic circuit from data
Artur Karimov, Vyacheslav Rybin, Ekaterina Kopets, Timur Karimov, Erivelton Nepomuceno, and Denis Butusov. Identifying empirical equations of chaotic circuit from data. Nonlinear Dynamics, 111(1):871–886, 2023
2023
-
[13]
Characterizations of border bases
Achim Kehrein and Martin Kreuzer. Characterizations of border bases. Journal of Pure and Applied Algebra, 196(2):251–270, 2005
2005
-
[14]
An algebraist’s view on border bases
Achim Kehrein, Martin Kreuzer, and Lorenzo Robbiano. An algebraist’s view on border bases. In Alicia Dickenstein and Joannis Z. Emiris, editors, Solving Polynomial Equations , pages 169–202. Springer-Verlag, Berlin, 2005
2005
-
[15]
Border basis computation with gradient-weighted normalization
Hiroshi Kera. Border basis computation with gradient-weighted normalization. In Proceedings of the 2022 International Symposium on Symbolic and Algebraic Computation , pages 225–234, New York, 2022. Association for Computing Machinery
2022
-
[16]
Noise-tolerant algebraic method for reconstruction of non- linear dynamical systems
Hiroshi Kera and Yoshihiko Hasegawa. Noise-tolerant algebraic method for reconstruction of non- linear dynamical systems. Nonlinear Dynamics, 85(1):675–692, 2016
2016
-
[17]
Approximate vanishing ideal via data knotting
Hiroshi Kera and Yoshihiko Hasegawa. Approximate vanishing ideal via data knotting. In Pro- ceedings of the Thirty-Second AAAI Conference on Artificial Intelligence (AAAI) , pages 3399–3406, Palo Alto, California, 2018. AAAI Press. 35
2018
-
[18]
Spurious vanishing problem in approximate vanishing ideal
Hiroshi Kera and Yoshihiko Hasegawa. Spurious vanishing problem in approximate vanishing ideal. IEEE Access, 7:178961–178976, 2019
2019
-
[19]
Gradient boosts the approximate vanishing ideal
Hiroshi Kera and Yoshihiko Hasegawa. Gradient boosts the approximate vanishing ideal. In Pro- ceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence (AAAI) , pages 4428–4425, Palo Alto, California, 2020. AAAI Press
2020
-
[20]
Monomial-agnostic computation of vanishing ideals
Hiroshi Kera and Yoshihiko Hasegawa. Monomial-agnostic computation of vanishing ideals. Journal of Computational Algebra , 11:100022, 2024
2024
-
[21]
Vanishing ideal genetic programming
Hiroshi Kera and Hitoshi Iba. Vanishing ideal genetic programming. In Proceedings of the 2016 IEEE Congress on Evolutionary Computation (CEC) , pages 5018–5025, Piscataway, NJ, 2016. IEEE
2016
-
[22]
Dual-to-kernel learning with ideals
Franz J Kir´ aly, Martin Kreuzer, and Louis Theran. Dual-to-kernel learning with ideals. 2014
2014
-
[23]
Computational commutative algebra 2 , volume 2
Martin Kreuzer and Lorenzo Robbiano. Computational commutative algebra 2 , volume 2. Springer Science & Business Media, 2005
2005
-
[24]
Computation of approximate border bases and applications
Jan Limbeck. Computation of approximate border bases and applications . PhD thesis, Passau, Universit¨ at Passau, 2013
2013
-
[25]
Vanishing component analysis
Roi Livni, David Lehavi, Sagi Schein, Hila Nachliely, Shai Shalev-Shwartz, and Amir Globerson. Vanishing component analysis. In Sanjoy Dasgupta and David McAllester, editors, Proceedings of the 30th International Conference on Machine Learning , volume 28(1) of Proceedings of ...
2013
-
[26]
Approximating latent manifolds in neural networks via vanishing ideals, 2025
Nico Pelleriti, Max Zimmer, Elias Wirth, and Sebastian Pokutta. Approximating latent manifolds in neural networks via vanishing ideals, 2025
2025
-
[27]
Approximate Commutative Algebra
Lorenzo Robbiano and John Abbott. Approximate Commutative Algebra. Springer, Vienna, 2010
2010
-
[28]
Hans J. Stetter. Numerical Polynomial Algebra . Society for Industrial and Applied Mathematics, USA, 2004
2004
-
[29]
Application of algebra in the oil industry
Maria-Laura Torrente. Application of algebra in the oil industry . PhD thesis, Scuola Normale Superiore, Pisa, 2008
2008
-
[30]
Wang and T
L. Wang and T. Ohtsuki. Nonlinear blind source separation unifying vanishing component analysis and temporal structure. IEEE Access, 6:42837–42850, 2018
2018
-
[31]
Polynomial representation for persistence diagram
Zhichao Wang, Qian Li, Gang Li, and Guandong Xu. Polynomial representation for persistence diagram. In Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 6116–6125, Los Alamitos, California, 2019. Conference Publishing Ser- vices
2019
-
[32]
Conditional gradients for the approximately vanishing ideal
Elias Wirth and Sebastian Pokutta. Conditional gradients for the approximately vanishing ideal. 2022
2022
-
[33]
Approximate vanishing ideal computa- tions at scale
Elias Samuel Wirth, Hiroshi Kera, and Sebastian Pokutta. Approximate vanishing ideal computa- tions at scale. In The Eleventh International Conference on Learning Representations , 2023
2023
-
[34]
Deep vanishing component analysis network for pattern classification
Hongliang Yan, Zifei Yan, Gang Xiao, Weizhi Wang, and Wangmeng Zuo. Deep vanishing component analysis network for pattern classification. Neurocomputing, 316:240–250, 2018
2018
-
[35]
Hand posture recognition using approximate vanishing ideal genera- tors
Yan-Guo Zhao and Zhan Song. Hand posture recognition using approximate vanishing ideal genera- tors. In Proceedings of the 2014 IEEE International Conference on Image Processing (ICIP) , pages 1525–1529. IEEE, 2014. 36 Appendices A Comparison of border basis computation method...
2014
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.