Pith. sign in

REVIEW 3 major objections 4 minor 28 references

NEUBORN: The Neurodevelopmental Evolution framework Using BiOmechanical RemodelliNg

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read NEUBORN claims that injecting a tissue-level isotropic growth prior into a diffeomorphic deep registration network produces longitudinal warps that match state-of-the-art alignment accuracy while being orders of magnitude more anatomically

desk verdict The registration contribution is real, but the headline growth-trajectory claim is not yet supported: the growth prior is computed from the same follow-up volumes used for evaluation, so the accuracy is partly enforced rather than predicted. read the letter →

arxiv 2508.09757 v1 pith:ID66PL2N submitted 2025-08-13 q-bio.QM cs.AI

Nashira Baena , Mariana da Silva , Irina Grigorescu , Aakash Saboo , Saga Masui , Jaques-Donald Tournier , Emma C. Robinson This is my paper · ORCID
classification q-bio.QMcs.AI
keywords biomechanicalregistrationlongitudinalneonatalbrainMRIcorticalgrowthdiffeomorphicdeformationdeeplearningNeo-Hookeanhyperelasticitydecomposition
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

This paper tries to show that longitudinal brain registration of neonatal MRI can be made both alignment-accurate and anatomically plausible by feeding a deep learning registration network a subject-specific biomechanical growth prior. The prior prescribes uniform expansion within each segmented brain tissue, derived from the volume change between the two time points, and a Neo-Hookean hyperelastic loss enforces mechanical plausibility. In tests on 19 unseen subjects, the method matches the alignment accuracy of a leading learning-based baseline while producing orders of magnitude fewer folded (negative Jacobian) voxels and simulated follow-up cortical volumes much closer to true individual volumes. If right, it makes individual growth-trajectory modelling from routine longitudinal scans both sharper and more biologically interpretable.

What carries the argument

The load-bearing mechanism is the multiplicative decomposition of the deformation gradient $F = F_e F_g$ (elastic times growth), with $F_g$ prescribed isotropically per segmented structure as $g^{-1/3} I$ where $g$ is the linear-aligned volume ratio between time points, and $F_e$ computed through a differentiable Neo-Hookean strain-energy loss. This converts a biomechanical growth prior into a differentiable regulariser that steers the network's deformation away from implausible folds while letting the image-similarity loss handle alignment.

What would settle it

Compare NEUBORN's Jacobian determinant fields against regionally sub-parcellated volume changes computed from an independent longitudinal surface correspondence: if local Jacobian expansion does not track local tissue growth even though global volumes match, the subject-specific growth claim fails. Alternatively, run NEUBORN on test-retest scans with no growth; if it still predicts substantial volume expansion, the target-informed prior is overfitting.

Watch

Extended reading notes

Core claim

The core claim is that prescribing growth at the tissue level inside a diffeomorphic, learning-based registration network yields deformations that are simultaneously as accurate, more diffeomorphic, and more biologically faithful than current baselines. The deformation is built from two hierarchical stationary velocity fields integrated by scaling and squaring; the biomechanical loss penalises strain energy after the deformation is factored as elastic response times prescribed growth, with growth set isotropically inside each segmented structure from the tissue's volume ratio across time. On held-out subjects, the method attains Dice overlap statistically indistinguishable from the learning-

Load-bearing premise

Everything hinges on the assumption that within each segmented brain structure, growth is a single uniform expansion whose rate is taken from the volume ratio after linear alignment; if true growth is uneven inside a structure, the reported volume accuracy is a fitting artifact and not evidence of biological fidelity.

Editorial extensions

If this is right

  • By interpolating the learned velocity field, the framework can produce intermediate brain states between the two scan ages, giving denser growth trajectories from sparse follow-ups (flagged as future work in the paper).
  • The same per-subject deformation can be projected forward to simulate later brain structure at the patient level, supporting early identification of deviant cortical development.
  • Smoother, less folded warps make the resulting anatomical correspondences more reliable for downstream morphometric analyses.
  • The general structure — a differentiable biomechanical loss plus a prescribed growth field — can be adapted to other longitudinal brain-change scenarios, such as atrophy, by changing the growth prescription.

Reading between the lines

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

  • The reported volume accuracy may be partly inherited from the target-informed growth map: the growth factor $g$ is computed from the follow-up volumes the network is asked to reproduce. A stricter test would predict growth from the baseline scan alone, without using the follow-up volume ratio, and compare ASPVC.
  • The paper itself notes only relative growth is assessed because linear registration removes absolute volume changes; this means the claim of 'growth trajectories' concerns relative expansion, not absolute brain size change.
  • A natural stress test is to replace the homogeneous isotropic growth with a spatially varying map and see whether the low ASPVC persists; this would separate the biomechanical regulariser's contribution from the information already contained in the volume-ratio prior.
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 / 4 minor

Summary. The paper presents NEUBORN, a deep-learning framework for longitudinal diffeomorphic registration of neonatal brain MRI. It couples a hierarchical two-resolution U-Net stationary-velocity-field architecture with a Neo-Hookean hyperelastic biomechanical loss intended to enforce biologically plausible growth. The model is trained and evaluated on 92 dHCP preterm infants (64 train, 9 validation, 19 test). The authors report that NEUBORN matches VoxelMorph in Dice overlap, produces drastically fewer negative Jacobians than VoxelMorph, and yields lower ASPVC for cortical volume, which they interpret as better adherence to subject-specific and population-level growth trajectories. The abstract and conclusions claim that the framework 'generates growth trajectories that better follow population-level trends' and 'accurately preserves complex subject-specific cortical growth trajectories.'

Significance. If the growth-trajectory claims held, this would be a valuable contribution: it would combine state-of-the-art alignment accuracy with biomechanically interpretable deformations, potentially improving normative modeling of early brain development. The registration-quality findings—comparable Dice to a strong baseline and a four-order-of-magnitude reduction in negative Jacobians—are credible and of practical interest. However, the central growth-accuracy claim is not independently established because the biomechanical growth map is prescribed from the same follow-up volumes used as ground truth for evaluation. The paper deserves attention for its registration methodology, but the predictive and biological-validity claims require substantial reframing or re-evaluation.

major comments (3)
  1. [§2.2, Eq. (7); §3, Eq. (8), Table 1] The growth map is prescribed as F_g = (g^{-1/3})I with g = V_{t1,seg}/V_{t2,seg}, using the follow-up segmentation volumes that later serve as ground truth for ASPVC. Since det(F_g) = g^{-1}, the biomechanical loss (J_e − 1)^2 with J_e = det(F_e) = g·det(F) drives the determinant of the total deformation toward 1/g = V_{t2,seg}/V_{t1,seg} in every voxel of each structure. Thus the reported ASPVC_cortex = 3.82 (Table 1) is a direct consequence of injecting the target volume ratios into the objective, not evidence of prediction. The ablation in Table 2 (ASPVC rises to 16.8 without the biomechanical loss) is fully consistent with this interpretation. To support the growth-trajectory claim, the evaluation must use growth maps derived without access to the follow-up volumes (e.g., predicted from timepoint 1 or from a population prior) rather than the measured V_{t2,seg}.
  2. [§2.2 'Subject-specific growth maps'; §4 Conclusions] The definition of g requires V_{t2,seg}, the follow-up segmentation, at test time. The manuscript does not specify whether this volume is available during inference or whether g is computed from the reference image in a standard registration setting. As written, the method cannot 'generate growth trajectories' or 'enable forward projections of future brain changes' (Conclusions) because it needs the future volumes to define the growth map. Please clarify the exact inference protocol: if g is computed from the reference scan (which is legitimate for registration), the predictive claims must be removed; if g is predicted from timepoint 1 alone, the evaluation must reflect that setting.
  3. [§3, Figure 4; §3 'Evaluation methods'] The population-level age-volume trends in Figure 4 are computed from 'all 92 available subjects,' which includes the 19 test subjects whose simulated follow-up volumes are compared against that same trend. This makes the claim that NEUBORN 'better follow[s] population-level growth trajectories' non-independent. The trend should be estimated from the training/validation subjects only, or via leave-one-out (or split-half) schemes. Additionally, the Wilcoxon test for NEUBORN cortical volumes against ground truth yields p=0.05, and all methods show significant white-matter volume differences; the abstract's statement that the model 'accurately preserves complex subject-specific cortical growth trajectories' overstates the evidence.
minor comments (4)
  1. [Title and §3 heading] The title uses 'BiOmechanical' while the section heading on page 3 reads 'Using BiOmechnical RemodelliNg' (missing 'a'). Please unify the spelling.
  2. [Eq. (3)] The composition formula ϕ = ϕ1 ◦ ϕ2 + ϕ2 is unusual. If ϕ1 and ϕ2 are velocity fields that are integrated separately, the addition is not standard; please clarify the notation and confirm the intended composition.
  3. [Table 1] The column 'Jϕ ≤ 0 (%)' is ambiguous: is this the percentage of negative-Jacobian voxels per image, or per total brain tissue? Please define clearly in the caption. Also, the Dice standard deviations (e.g., 0.0016) appear overly precise; consider reporting to 1-2 significant digits.
  4. [§3, Abstract] The statement 'with fewer negative Jacobians' relative to state-of-the-art baselines is misleading when Elastic Syn has exactly zero negative Jacobians. Please qualify this claim as relative to VoxelMorph specifically.

Circularity Check

1 steps flagged · score 8.0 of 10

Growth-trajectory claims are enforced by target-derived growth maps, making ASPVC and trend adherence fitting artifacts.

  1. fitted input called prediction [Section 2.2, Eq. (6)-(7); Section 3, Table 1 and Fig. 4]
    "Subject-specific growth maps are prescribed isotropically as Fg = (g−1/3)I, where g = Vt1,seg/Vt2,seg is estimated from the relative volume change from time point 1 to time point 2, for each of the segmented structures (cortical grey matter, white matter, ventricles, cerebellum, deep grey matter, brainstem and hippocampi & amygdala), after linear alignment."

    With F = Fe Fg and det(Fg) = g^{-1}, the Neo-Hookean term (Eq. 7) is minimized when Je = det(Fe) = g det(F) is close to 1, i.e. when det(F) ≈ 1/g = Vt2/Vt1. The loss therefore explicitly drives each structure's Jacobian determinant to the target volume ratio. The reported ASPVCcortex = 3.82 then measures how well the deformation implements the prescribed growth map, not whether growth is predicted. The same Vt2 values define the ground-truth follow-up volumes and the g maps, so the ASPVC and the Fig. 4 trend adherence claims reduce by construction to the fitted inputs.

full rationale

The central growth-trajectory claim of the paper is circular: the biomechanical loss is supervised with a per-structure target volume ratio g = Vt1,seg/Vt2,seg measured from the same follow-up segmentations that are later used as ground truth for ASPVC. Since det(Fg) = g^{-1}, the (Je−1)^2 term in Eq. 7 forces det(F) toward Vt2/Vt1, so the low ASPVC and apparent trend following are consequences of target-derived supervision rather than independent prediction. The ablation result (ASPVC 16.8 without the biomechanical loss) is fully consistent with this: removing the loss removes the enforced target volumes. The 'better follow population-level trends' comparison is further weakened by the fact that Fig. 4's trend is computed from all 92 subjects, including the 19 test subjects whose predictions are plotted, so the reference curve is contaminated with the very data being evaluated. The registration and diffeomorphicity results (Dice comparable to VoxelMorph, ~4 orders of magnitude fewer negative Jacobians) are independent and credible; they do not depend on the growth-map circularity. The self-citations [7,23] for the hyperelastic implementation are not by themselves circular because Eq. 7 is stated in the paper and the constitutive model is a standard Neo-Hookean form; the circularity lies in the target-derived prescription of Fg. Score is set to 8 rather than 10 because the alignment-accuracy component is independently assessed and the circularity affects the growth-trajectory claim specifically.

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

The central claim relies on neural network training plus a biomechanical prior whose growth input is derived from the very target volumes used in evaluation. The prior literature supplies standard registration tools and material parameters; the main unverified piece is the homogeneous isotropic growth assumption and the reuse of target volumes.

free parameters (4)
  • per-structure growth scalar g per subject = V_t1,seg / V_t2,seg
    Computed from the same follow-up volumes used as evaluation target; prescribes isotropic growth Fg=(g^-1/3)I and drives low ASPVC.
  • loss weights lambda1, lambda2 = 0.01 and 1e-5
    Chosen via 9-subject validation; they balance image similarity, biomechanical strain energy, and smoothness.
  • shear modulus mu by tissue = 0 background, 0.01 CSF, 1 brain
    Hand-assigned from prior literature [23,7], not fitted to this dataset.
  • bulk modulus kappa = 100 mu
    Set from prior work [23,7], following Neo-Hookean convention.
assumptions (4)
  • standard math Diffeomorphic SVF integration via scaling and squaring yields topology-preserving maps
    Assumed throughout Section 2.2, standard in diffeomorphic registration literature but not proven here.
  • domain assumption Neo-Hookean hyperelasticity with mu and kappa assigned by tissue type approximates neonatal brain mechanics
    Borrowed from [23,7,20]; no validation against measured tissue properties in this paper.
  • domain assumption Multiplicative decomposition F=FeFg with isotropic Fg is a valid growth model
    Carried from Rodriguez et al. [20]; assumes homogeneous isotropic growth per segmented structure.
  • ad hoc to paper g=V_t1,seg/V_t2,seg from linear-aligned segmentations measures true growth
    This is the key target-informed input enabling low ASPVC; not independently justified and creates circular evaluation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NEUBORN: The Neurodevelopmental Evolution framework Using BiOmechanical RemodelliNg." pith.science (2026). https://pith.science/paper/ID66PL2N

@misc{pith2026250809757,
  author       = {Pith},
  title        = {Pith review of: NEUBORN: The Neurodevelopmental Evolution framework Using BiOmechanical RemodelliNg},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ID66PL2N}},
  note         = {Machine review of arXiv:2508.09757}
}
read the original abstract

Understanding individual cortical development is essential for identifying deviations linked to neurodevelopmental disorders. However, current normative modelling frameworks struggle to capture fine-scale anatomical details due to their reliance on modelling data within a population-average reference space. Here, we present a novel framework for learning individual growth trajectories from biomechanically constrained, longitudinal, diffeomorphic image registration, implemented via a hierarchical network architecture. Trained on neonatal MRI data from the Developing Human Connectome Project, the method improves the biological plausibility of warps, generating growth trajectories that better follow population-level trends while generating smoother warps, with fewer negative Jacobians, relative to state-of-the-art baselines. The resulting subject-specific deformations provide interpretable, biologically grounded mappings of development. This framework opens new possibilities for predictive modeling of brain maturation and early identification of malformations of cortical development.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 18 canonical work pages

  1. [1]

    NeuroImage 38(1), 95–113 (2007)

    Ashburner, J.: A fast diffeomorphic image registration algorithm. NeuroImage 38(1), 95–113 (2007). https://doi.org/10.1016/j.neuroimage.2007.07.007

    doi:10.1016/j.neuroimage.2007.07.007 2007

  2. [2]

    Insight J 2(365), 1–35 (2009)

    Avants, B.B., Tustison, N., Song, G., et al.: Advanced normalization tools (ants). Insight J 2(365), 1–35 (2009)

    work page 2009

  3. [3]

    IEEE Transactions on Medical Imaging 38(8), 1788–1800 (2019)

    Balakrishnan, G., Zhao, A., Sabuncu, M.R., Guttag, J., Dalca, A.V.: Voxelmorph: A learning framework for deformable medical image regis- tration. IEEE Transactions on Medical Imaging 38(8), 1788–1800 (2019). https://doi.org/10.1109/TMI.2019.2897538

    work page arXiv 2019

  4. [4]

    Brain 120(2), 257–269 (1997)

    Bartley, A.J., Jones, D.W., Weinberger, D.R.: Genetic variability of hu- man brain size and cortical gyral patterns. Brain 120(2), 257–269 (1997). https://doi.org/10.1093/brain/120.2.257

    work page doi:10.1093/brain/120.2.257 1997

  5. [5]

    Acta Biomaterialia 48, 319–340 (2017)

    Budday, S., Sommer, G., Birkl, C., Langkammer, C., et al.: Mechanical char- acterization of human brain tissue. Acta Biomaterialia 48, 319–340 (2017). https://doi.org/10.1016/j.actbio.2016.10.036

    doi:10.1016/j.actbio.2016.10.036 2017

  6. [6]

    Scientific Reports4, 5644 (2014)

    Budday, S., Raybaud, C., Kuhl, E.: A mechanical model predicts morphological abnormalities in the developing human brain. Scientific Reports4, 5644 (2014). https://doi.org/10.1038/srep05644

    doi:10.1038/srep05644 2014

  7. [7]

    arXiv preprint arXiv:2108.08214 (2021)

    Da Silva, M., Sudre, C.H., Garcia, K., et al.: Distinguishing healthy ageing from dementia: a biomechanical simulation of brain atrophy using deep networks. arXiv preprint arXiv:2108.08214 (2021)

    arXiv 2021

  8. [8]

    Medical Image Analysis 57, 226–236 (2019)

    Dalca, A.V., Balakrishnan, G., Guttag, J., Sabuncu, M.R.: Unsupervised learning of probabilistic diffeomorphic registration for images and surfaces. Medical Image Analysis 57, 226–236 (2019). https://doi.org/10.1016/j.media.2019.07.006

    doi:10.1016/j.media.2019.07.006 2019

Show all 28 references
  1. [9]

    NeuroImage 185, 575–592 (2019)

    Duan, D., Xia, S., Rekik, I., Meng, Y., et al.: Exploring folding pat- terns of infant cerebral cortex based on multi-view curvature fea- tures: Methods and applications. NeuroImage 185, 575–592 (2019). https://doi.org/10.1016/j.neuroimage.2018.08.041

    2019 doi

  2. [10]

    bioRxiv (2025)

    Guo, Y., Suliman, M.A., Williams, L.Z.J., Besenczi, R., et al.: Motifs of brain cor- tical folding from birth to adulthood: structural asymmetry and folding-functional links. bioRxiv (2025). https://doi.org/10.1101/2025.03.04.641459

    2025 doi

  3. [11]

    NeuroImage 62(2), 782–790 (2012)

    Jenkinson, M., Beckmann, C.F., Behrens, T.E.J., Woolrich, M.W., Smith, S.M.: Fsl. NeuroImage 62(2), 782–790 (2012). https://doi.org/10.1016/j.neuroimage.2011.09.015

    2012 doi

  4. [12]

    Medical Image Analysis5(2), 143–156 (2001)

    Jenkinson, M., Smith, S.: A global optimisation method for robust affine registra- tion of brain images. Medical Image Analysis5(2), 143–156 (2001)

    2001

  5. [13]

    arXiv preprint arXiv:1711.05101 (2019)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2019). https://doi.org/10.48550/arXiv.1711.05101

    2019 doi arXiv

  6. [14]

    NeuroImage125, 456–478 (2016)

    Makropoulos, A., Aljabar, P., Wright, R., Hüning, B., et al.: Regional growth and atlasing of the developing human brain. NeuroImage125, 456–478 (2016)

    2016

  7. [15]

    IEEE Transactions on Medical Imaging 33(9), 1818–1831 (2014)

    Makropoulos, A., Gousias, I.S., Ledig, C., Aljabar, P., et al.: Automatic whole brain mri segmentation of the developing neonatal brain. IEEE Transactions on Medical Imaging 33(9), 1818–1831 (2014)

    2014

  8. [16]

    NeuroImage 173, 88–112 (2018)

    Makropoulos, A., Robinson, E.C., Schuh, A., Wright, R., et al.: The de- veloping human connectome project: a minimal processing pipeline for neonatal cortical surface reconstruction. NeuroImage 173, 88–112 (2018). https://doi.org/10.1101/125526

    2018 doi

  9. [17]

    https://doi.org/10.1002/hbm.24199 Neurodevelopmental Evolution Using Biomechanical Remodelling 11

    Meng, Y., Li, G., Wang, L., Lin, W., Gilmore, J.H., Shen, D.: Discovering cortical sulcalfoldingpatternsinneonatesusinglarge-scaledataset.HumanBrainMapping 39(9), 3625–3635 (2018). https://doi.org/10.1002/hbm.24199 Neurodevelopmental Evolution Using Biomechanical Remodelling 11

    2018 doi

  10. [18]

    https://doi.org/10.48550/arXiv.2006.16148

    Mok, T.C.W., Chung, A.C.S.: Large deformation diffeomor- phic image registration with laplacian pyramid networks (2020). https://doi.org/10.48550/arXiv.2006.16148

    2020 doi arXiv

  11. [19]

    Paszke, A., Gross, S., Massa, F., Lerer, A., et al.: Pytorch: An imperative style, high-performancedeeplearninglibrary.AdvancesinNeuralInformationProcessing Systems 32 (2019)

    2019

  12. [20]

    Journal of Biomechanics 27(4), 455–467 (1994)

    Rodriguez, E.K., Hoger, A., McCulloch, A.D.: Stress-dependent finite growth in soft elastic tissues. Journal of Biomechanics 27(4), 455–467 (1994). https://doi.org/10.1016/0021-9290(94)90021-3

    1994 doi

  13. [21]

    Cerebral Cortex 31(1), 702–715 (2020)

    Schmitt, J.E., Raznahan, A., Liu, S., Neale, M.C.: The heritability of cortical folding: Evidence from the human connectome project. Cerebral Cortex 31(1), 702–715 (2020). https://doi.org/10.1093/cercor/bhaa254

    2020 doi

  14. [22]

    Brain143(10), 2874–2894 (2020)

    Severino, M., Geraldo, A.F., Utz, N., Tortora, D., et al.: Definitions and classifica- tion of malformations of cortical development: practical guidelines. Brain143(10), 2874–2894 (2020). https://doi.org/10.1093/brain/awaa174

    2020 doi

  15. [23]

    arXiv preprint arXiv:2012.07596 (2020)

    da Silva, M., Garcia, K., Sudre, C.H., et al.: Biomechanical modelling of brain atrophy through deep learning. arXiv preprint arXiv:2012.07596 (2020)

    2012 arXiv

  16. [24]

    Medical Image Analysis 26(1), 203–216 (2015)

    Simpson, I.J.A., Cardoso, M.J., Modat, M., Cash, D.M., et al.: Probabilistic non- linear registration with spatially adaptive regularisation. Medical Image Analysis 26(1), 203–216 (2015). https://doi.org/10.1016/j.media.2015.08.006

    2015 doi

  17. [25]

    Proceedings of the National Academy of Sciences 111(35), 12667–12672 (2014)

    Tallinen, T., Chung, J.Y., Biggins, J.S., Mahadevan, L.: Gyrification from con- strained cortical expansion. Proceedings of the National Academy of Sciences 111(35), 12667–12672 (2014). https://doi.org/10.1073/pnas.1406015111

    2014 doi

  18. [26]

    Nature Physics12(6), 588–593 (2016)

    Tallinen, T., Chung, J.Y., Rousseau, F., Girard, N., Lefèvre, J., Mahadevan, L.: On the growth and form of cortical convolutions. Nature Physics12(6), 588–593 (2016). https://doi.org/10.1038/nphys3632

    2016 doi

  19. [27]

    Brain Multiphysics 2, 100025 (2021)

    Zarzor, M.S., Kaessmair, S., Steinmann, P., Blümcke, I., Budday, S.: A two-field computational model couples cellular brain development with cortical folding. Brain Multiphysics 2, 100025 (2021). https://doi.org/10.1016/j.brain.2021.100025

    2021

  20. [28]

    NeuroImage 218, 116946 (2020)

    Zöllei, L., Iglesias, J.E., Ou, Y., Grant, P.E., Fischl, B.: Infant freesurfer: An automated segmentation and surface extraction pipeline for t1-weighted neuroimaging data of infants 0–2 years. NeuroImage 218, 116946 (2020). https://doi.org/10.1016/j.neuroimage.2020.116946

    2020

Pith tools

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