Pith. sign in

REVIEW 4 major objections 5 minor 21 references

Geometric GNNs for Charged Particle Tracking at GlueX

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a graph-neural-network edge classifier, fed by a geometrically constrained graph builder with skip edges, finds charged-particle track segments in GlueX data with 7.5% higher efficiency at equal purity and 71% lower…

desk verdict Solid engineering benchmark for GNN tracking on GlueX FDC, but the headline efficiency gain is only as good as the test-set hygiene, and that hygiene isn't documented. read the letter →

arxiv 2505.22504 v1 pith:YJ7NMKCH submitted 2025-05-28 cs.LG

Ahmed Hossam Mohammed , Kishansingh Rajput , Simon Taylor , Denis Furletov , Sergey Furletov , Malachi Schram This is my paper · ORCID
classification cs.LG
keywords chargedparticletrackinggraphneuralnetworkedgeclassificationtrackfindingGlueXexperimentGPUinferenceFPGAdeploymentskipedges
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 argues that track finding in the GlueX detector can be cast as an edge-classification problem on graphs whose nodes are hits and whose edges are candidate track segments, and that a compact Graph Neural Network solving that classification beats the experiment's current combinatorial tracker. On simulated test events, the ML pipeline reaches 98.06% segment efficiency at 94.62% purity, compared with 91.19% efficiency at the same purity for the traditional method. By batching many events, the pipeline runs at about 44 microseconds per event on an A100 GPU versus 152 microseconds for the traditional method, a 71% reduction. The same model can be deployed on an FPGA at a small efficiency cost, which the paper presents as a trade-off for even faster inference.

What carries the argument

The central mechanism is a two-stage pipeline. The graph-builder connects hits on adjacent, and via skip edges near-adjacent, detector layers only when the hits satisfy the three geometric constraints, so the input graph has high recall and moderate purity. The GNN then alternates between an edge-network, which assigns a probability to every edge, and a node-network, which updates hit embeddings by aggregating neighbor embeddings weighted by those probabilities; after one message-passing iteration the edge probabilities are the predicted track segments. Batching multiple event graphs into one tensorized graph on the GPU is what yields the 44-microsecond per-event inference time, and the skip edges are what let a single message-passing iteration reach hits several layers away despite missing intermediate hits.

What would settle it

Run the graph-builder constraint optimization on a split of simulation that is strictly disjoint from the test events, then recompute the ML pipeline's efficiency at purity 0.9462. If the retuned constraints no longer yield a graph-builder efficiency near 0.9905 and the pipeline efficiency drops to or below the traditional method's 0.9119, the reported 7.5% gain is an artifact of tuning on the evaluation distribution.

Watch

Extended reading notes

Core claim

The central claim is that an edge-classifier GNN, trained on simulated GlueX forward-drift-chamber data, can replace the combinatorial segment-linking tracker with a faster and more efficient pipeline. The authors define efficiency and purity at the segment level, where a true segment is an edge joining two hits left by the same particle. At the purity value the traditional method achieves (0.9462), the GNN pipeline keeps 0.9806 of true segments, a 7.5% relative improvement over the traditional method's 0.9119. The gain comes with a substantial speed advantage: with batched graph construction and inference on an A100 GPU, the ML pipeline takes about 44 microseconds per event, versus 152 microseconds for the traditional method. A key part of the design is the graph-builder, whose geometric cutoffs ($d_{xy} < 34.4$ cm, $d_{xy}/d_z < 5.4$, and $|\Delta\varphi| < 2.3$ rad) and skip edges that jump over up to a fixed number of missing layers are chosen to keep nearly all true segments available to the classifier, so the GNN only needs to discard false edges.

Load-bearing premise

The load-bearing assumption is that the geometric cutoffs for building the graphs were chosen without ever looking at the test events; if the same events were used to tune those cutoffs, the reported gain is partly a fitted number.

Editorial extensions

If this is right

  • At the threshold where the ML pipeline matches the traditional method's purity of 0.9462, the ML pipeline's segment efficiency is 0.9806 versus 0.9119 for the traditional method, a 7.5% relative improvement.
  • With batch size 128 on an A100 GPU, the full ML pipeline, including graph building, GNN inference, and data transfer, takes about 44 microseconds per event, 71% faster than the traditional method's 152 microseconds.
  • Increasing message-passing iterations from 1 to 3 raises efficiency only from 0.9806 to 0.9857, less than 0.6%, while more than doubling inference time, so the authors keep one iteration.
  • Skip edges with a maximum reach of three intermediate layers allow one message-passing iteration to inform a node about up to fourth-degree neighbors, which is why a shallow GNN suffices.
  • On an FPGA, a reduced-width version of the same GNN runs with an initiation interval of about 2.5 microseconds, reaching 0.9565 efficiency at the same purity of 0.9462, trading a small efficiency drop for a large speedup.

Reading between the lines

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

  • Inference: The paper does not report a held-out split for the multi-objective genetic algorithm that set the graph-builder cutoffs; if those cutoffs were tuned on the same simulated distribution used for the final numbers, the 0.9806 efficiency is partly a fitted value and the margin over the traditional method would be smaller on an untouched sample.
  • Inference: Because the batching speedup comes from amortizing graph construction and inference over many events, the pipeline's 71% time advantage is likely to grow on higher-multiplicity events, where the combinatorial baseline scales worse than linearly; the paper only tests one simulation distribution.
  • Inference: The FPGA numbers cover only the GNN edge classifier, not graph building, so the total FPGA pipeline gain remains an open question until graph construction is also ported; at a 2.5-microsecond initiation interval the classifier would no longer be the bottleneck.
  • Inference: Since the GNN can only delete edges that the graph-builder created, the pipeline's ceiling is the graph-builder recall of 0.9905, and any event with more consecutive missing hits than the skip-edge allowance permits is unrecoverable by the classifier, so robustness on real data depends on the missing-hit distribution matching the simulation.
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

4 major / 5 minor

Summary. The paper presents a Graph Neural Network (GNN) based edge-classification pipeline for charged-particle track finding in the GlueX Forward Drift Chamber. A graph builder connects hits on adjacent or near-adjacent layers using three geometric constraints, and a small edge-classifying GNN filters false edges. On simulated test events, the pipeline is reported to achieve 0.9806 segment efficiency at purity 0.9462, versus 0.9119 efficiency for the traditional method, and a batched A100 GPU implementation runs in about 44 us per event versus 152 us for the traditional method. The paper also describes an FPGA implementation with lower latency but reduced efficiency. The central claim is a 7.5% efficiency improvement at fixed purity and a 71% inference-time reduction.

Significance. If the reported numbers survive a properly held-out evaluation, this is a practically useful result for GlueX and similar drift-chamber tracking applications. The paper gives concrete architecture details, an ablation of message-passing iterations, batched GPU timing measurements, and FPGA resource estimates, which are valuable engineering contributions. However, the headline comparison is currently vulnerable to evaluation-protocol issues: the graph-builder constraints are optimized on simulation without a documented held-out split, the operating threshold is selected on the test curve, and the skipmax value used for the reported results is not stated. These issues directly affect the claimed efficiency gain, so the contribution is significant but needs stronger validation before the quantitative claims can be accepted.

major comments (4)
  1. [Sec. 4 and App. A] The MOGA optimization of the graph-builder constraints is not documented as using a held-out split. Section 4 states that the cutoff values d_xy < 34.4 cm, d_xy/dz < 5.4, and |d_phi| < 2.3 rad were selected by a multi-objective genetic algorithm 'to optimize both efficiency and purity on realistic simulation data,' and Appendix A reports the resulting graph-builder efficiency and purity (0.9905 and 0.5473), but nowhere does the paper state that the MOGA used only training or validation events. If the same 15% test set was seen during constraint selection, the graph-builder efficiency is in-sample and, because the GNN cannot recover true edges already dropped by the graph builder, the pipeline efficiency of 0.9806 inherits an optimistic bias. The authors must either document that MOGA was run on a separate split or rerun the constraint selection on training/validation data and re-evaluate on the untouched test set.
  2. [Sec. 6.1, Fig. 5] The probability threshold is chosen by matching the traditional method's purity on the test set. The text says 'we applied a threshold on the output that yields the same purity value (0.9462),' and Fig. 5 shows the resulting operating point. This is a test-set-dependent selection of the operating point. Even if the threshold is considered a simple calibration parameter, the reported 7.5% efficiency gain is not a fully out-of-sample comparison unless the threshold is fixed using validation data before evaluating on the test set. The authors should either pre-specify the threshold on the validation split and report test-set efficiency at that fixed threshold, or clearly state that the comparison uses a test-set-optimized operating point and provide confidence intervals.
  3. [Sec. 6.1 and Fig. 7] The value of skipmax used for the reported pipeline results is never stated. Section 4 defines skipmax and Fig. 3 illustrates skipmax = 2, while Fig. 7a discusses an event built with skipmax = 3, but the main results in Sec. 6.1 do not specify which skipmax produced the 0.9806 efficiency and the 44 us timing. Because skipmax changes graph density, efficiency, and inference time, this is a load-bearing hyperparameter for both performance claims. The authors must report the exact skipmax used and, if it was selected by validation or test performance, treat it as an additional tuned hyperparameter in the evaluation protocol.
  4. [Sec. 6.1 and Fig. 6] The timing comparison lacks hardware context for the traditional method. The paper reports that the traditional method requires approximately 152 us per GlueX event, while the ML pipeline timing in Fig. 6 is measured on an A100 GPU, but it does not state on what hardware the traditional method was run (e.g., CPU model and clock, or GPU). If the traditional method is CPU-only and the ML pipeline uses a modern GPU, the 71% speedup is partly a hardware comparison rather than an algorithmic one. The authors should specify the platform for the traditional-method timing and ideally report the ML pipeline timing on the same hardware as the traditional method, or at least discuss the practical implications of the heterogeneous comparison.
minor comments (5)
  1. [Abstract and Sec. 6.1] The abstract says the model is tested on both simulation and real GlueX measurements, but the paper only presents two qualitative real-data examples in Fig. 7. There is no quantitative real-data efficiency/purity measurement. The authors should either add quantitative real-data results or soften the claim in the abstract.
  2. [Abstract] The abstract uses 'Graphical Processing Units'; the standard term is 'Graphics Processing Units.'
  3. [Fig. 5 caption] The caption says 'at a threshold ≈ 0.3' but does not explain what the threshold is. Please state explicitly that this is the edge-probability threshold applied to the GNN output.
  4. [Table 3] The abbreviation 'II' is used in the table without being defined in the table or nearby text. Please define 'Initiation Interval' at first use.
  5. [General] No code or data availability statement is provided. For an applied ML paper whose main contribution is an empirical comparison, releasing the trained model, graph-building code, and evaluation scripts would substantially aid reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; the reported gains are empirical out-of-sample measurements, with only an unstated data-selection protocol concern that does not constitute circularity.

full rationale

The paper's central claim is an empirical comparison between a GNN pipeline and the traditional GlueX track finder on simulated test events (Sec. 6.1). The only fitted quantities are the graph-builder cutoffs (Sec. 4, Appendix A) and the decision threshold used to match the traditional method's purity. The graph-builder cutoffs are selected by MOGA, but the reported pipeline efficiency of 0.9806 is computed from the trained GNN on the test set; there is no equation in which this efficiency equals the MOGA objective by construction. The threshold is chosen to match the traditional method's purity of 0.9462, which is an evaluation convention rather than a fitted parameter that alone determines the reported efficiency independently of the GNN's outputs. The paper does not invoke a uniqueness theorem, and its citations to prior GNN tracking work are contextual rather than load-bearing for the central comparison. The manuscript never states that the MOGA constraint selection used a training/validation split disjoint from the test set, and the threshold is applied on the test set; this is a potential statistical-validity limitation, but it is not a circular derivation. No step in the paper's derivation chain reduces to its own input.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the fidelity of GEANT4 simulation, the correctness of simulation-derived edge labels, and the fairness of the fixed-purity comparison against the traditional method. The graph-builder cutoffs are fitted constants from MOGA; the skipmax and output threshold are additional choices. No new physical entities are introduced.

free parameters (5)
  • Graph-builder d_xy cutoff = 34.4 cm
    Selected via NSGA-II MOGA to optimize graph efficiency and purity on simulation; directly controls the input edge set and hence all downstream metrics.
  • Graph-builder d_xy/dz cutoff = 5.4
    Same MOGA optimization; limits how far apart two hits can be in the xy-plane relative to their layer separation.
  • Graph-builder azimuthal cutoff = 2.3 rad
    Same MOGA optimization; restricts edges by the azimuthal angle difference between hits.
  • skipmax = Not reported for final metrics; figures show 2 and 3
    Controls the maximum number of intermediate layers a skip edge can cross; affects graph efficiency, density, and whether missing-hit tracks are connected. The paper does not state which value is used for the headline results.
  • Edge probability threshold = Approximately 0.3
    Chosen on the test set so that ML purity equals the traditional purity of 0.9462; efficiency at this threshold is the headline 7.5% gain.
assumptions (5)
  • domain assumption GEANT4 simulation of the GlueX FDC faithfully represents detector geometry, inefficiencies, and background for training and evaluation.
    All quantitative claims are based on simulated events (Sec. 3, Sec. 6.1); no real-data quantitative validation is provided.
  • domain assumption Ground-truth particle-to-hit association in simulation is correct and yields unambiguous edge labels.
    Sec. 4 derives true and false edge labels from simulated particle identity; errors here would bias efficiency and purity.
  • domain assumption Segment-level efficiency and purity is the appropriate figure of merit for track finding, and an equal-purity comparison is fair.
    Sec. 6.1 defines the metrics and compares at one fixed purity; downstream track-level or physics-object performance is not measured.
  • domain assumption The traditional GlueX track-finding implementation is the correct baseline and was run with representative settings.
    Sec. 3.1 describes the algorithm; no comparison to other GNN trackers or alternative traditional configurations is made.
  • domain assumption The graph-builder constraint optimization was performed without using the test events.
    App. A does not specify a held-out optimization dataset; if the MOGA saw the evaluation distribution, the reported numbers are optimistic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Geometric GNNs for Charged Particle Tracking at GlueX." pith.science (2026). https://pith.science/paper/YJ7NMKCH

@misc{pith2026250522504,
  author       = {Pith},
  title        = {Pith review of: Geometric GNNs for Charged Particle Tracking at GlueX},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YJ7NMKCH}},
  note         = {Machine review of arXiv:2505.22504}
}
read the original abstract

Nuclear physics experiments are aimed at uncovering the fundamental building blocks of matter. The experiments involve high-energy collisions that produce complex events with many particle trajectories. Tracking charged particles resulting from collisions in the presence of a strong magnetic field is critical to enable the reconstruction of particle trajectories and precise determination of interactions. It is traditionally achieved through combinatorial approaches that scale worse than linearly as the number of hits grows. Since particle hit data naturally form a 3-dimensional point cloud and can be structured as graphs, Graph Neural Networks (GNNs) emerge as an intuitive and effective choice for this task. In this study, we evaluate the GNN model for track finding on the data from the GlueX experiment at Jefferson Lab. We use simulation data to train the model and test on both simulation and real GlueX measurements. We demonstrate that GNN-based track finding outperforms the currently used traditional method at GlueX in terms of segment-based efficiency at a fixed purity while providing faster inferences. We show that the GNN model can achieve significant speedup by processing multiple events in batches, which exploits the parallel computation capability of Graphical Processing Units (GPUs). Finally, we compare the GNN implementation on GPU and FPGA and describe the trade-off.

Figures

Figures reproduced from arXiv: 2505.22504 by the authors.

Figure 1
Figure 1. A 3-D figure illustrating the geometry of the GlueX detector. It shows Forward [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. ML Pipeline Summary: The detector yields hit coordinates information. The [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Graphs built with no skip edges (i.e., skipmax = 0) are more susceptible to the problem of missing hits as shown in part (a). This problem is resolved by introducing edges between hits on non-consecutive layers with a maximum separation of 1+skipmax. Part (b) shows the same event built with a skipmax of 2 that does not suffer from the gap patterns which makes detecting the full track possible (i.e., higher efficienc… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Edge-Classifier GNN takes a graph (with X and E representing coordinates of graph hits and adjacency list, respectively) and returns the probability of each edge being true after performing I message passing iterations. study are calculated on the segment (edge) level.…
Figure 5
Figure 5. Figure 5: Overall efficiency & purity of the ML pipeline evaluated on simulated test [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: A100 GPU execution time per GlueX event. The analysis is conducted on [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Real GlueX examples hls4ml [19]. Each model was designed to process events with up to 150 nodes and 256 edges. The U200 FPGA board was used to obtain resource estimates during synthesis, whose maximum resources can be seen in [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Digital Signal Processing (DSP), Flip-Flop (FF), and Look Up Table (LUT) [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 15 canonical work pages

  1. [1]

    Adhikari S, Akondi C, Al Ghoul H, Ali A, Amaryan M, Anassontzis E, Austregesilo A, Barbosa F, Barlow J, Barnes A, Barriga E, Barsotti R, Beattie T, Benesch J, Berdnikov V, Biallas G, Black T, Boeglin W, Brindza P, Briscoe W, Britton T, Brock J, Brooks W, Cannon B, Carlin C, Carman D, Carstens T, Cao N, Chernyshov O, Chudakov E, Cole S, Cortes O, Crahen W,...

    work page 2021

  2. [2]

    Hansroul M, Jeremie H and Savard D 1988 Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment270 498–501 ISSN 0168-9002 URL https://www.sciencedirect.com/science/article/pii/016890028890722X

    arXiv 1988

  3. [3]

    Kisel I 2006 Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment566 85–88 ISSN 0168-9002 tIME 2005 URL https://www.sciencedirect.com/science/article/pii/S0168900206008151

    work page 2006

  4. [4]

    com/science/article/pii/S0168900219315517

    Brondolin E, Leogrande E, Hynds D, Gaede F, Petriˇ c M, Sailer A and Simoniello R 2020 Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors 17 and Associated Equipment 956 163304 ISSN 0168-9002 URL https://www.sciencedirect. com/science/article/pii/S0168900219315517

    work page 2020

  5. [5]

    Zhou H, Sun K, Lu Z, Li H, Ai X, Zhang J, Huang X and Liu J 2025 Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 1075 170357 ISSN 0168-9002 URL https://www.sciencedirect.com/science/ article/pii/S0168900225001585

    work page 2025

  6. [6]

    Fr¨ uhwirth R and Strandlie A 2021Track Finding (Springer International Publishing) chap 5, pp 81–102

    work page

  7. [7]

    Farrell S, Calafiura P, Mudigonda M, Prabhat, Anderson D, Vlimant J R, Zheng S, Bendavid J, Spiropulu M, Cerati G, Gray L, Kowalkowski J, Spentzouris P and Tsaris A 2018 Novel deep learning methods for track reconstruction ( Preprint 1810.06111)

    arXiv 2018

  8. [8]

    Ju X, Farrell S, Calafiura P, Murnane D, Prabhat, Gray L, Klijnsma T, Pedro K, Cerati G, Kowalkowski J, Perdue G, Spentzouris P, Tran N, Vlimant J R, Zlokapa A, Pata J, Spiropulu M, An S, Aurisano A, Hewes V, Tsaris A, Terao K and Usher T 2020 Graph Neural Networks for Particle Reconstruction in High Energy Physics detectors ( Preprint 2003.11603)

    arXiv 2020

Show all 21 references
  1. [9]

    2021 The European Physical Journal C81 1–14

    Ju X, Murnane D, Calafiura P, Choma N, Conlon S, Farrell S, Xu Y, Spiropulu M, Vlimant J R, Aurisano A et al. 2021 The European Physical Journal C81 1–14

    2021

  2. [10]

    DeZoort G, Thais S, Duarte J, Razavimaleki V, Atkinson M, Ojalvo I, Neubauer M and Elmer P 2021 Computing and Software for Big Science5 1–13

    2021

  3. [11]

    Heintz A, Razavimaleki V, Duarte J, DeZoort G, Ojalvo I, Thais S, Atkinson M, Neubauer M, Gray L, Jindariani S, Tran N, Harris P, Rankin D, Aarrestad T, Loncar V, Pierini M, Summers S, Ngadiuba J, Liu M, Kreinar E and Wu Z 2020 Accelerated Charged Particle Tracking with Graph ...

    2020 arXiv

  4. [12]

    Elabd A, Razavimaleki V, Huang S Y, Duarte J, Atkinson M, DeZoort G, Elmer P, Hauck S, Hu J X, Hsu S C, Lai B C, Neubauer M, Ojalvo I, Thais S and Trahms M 2022 Frontiers in Big Data 5 ISSN 2624-909X

    2022

  5. [13]

    Allison J, Amako K, Apostolakis J, Arce P, Asai M, Aso T, Bagli E, Bagulya A, Banerjee S, Barrand G, Beck B, Bogdanov A, Brandt D, Brown J, Burkhardt H, Canal P, Cano- Ott D, Chauvie S, Cho K, Cirrone G, Cooperman G, Cort´ es-Giraldo M, Cosmo G, Cuttone G, Depaola G, Desorgher...

    2016

  6. [14]

    Fonseca C and Fleming P 1993 Multiobjective genetic algorithms IEE Colloquium on Genetic Algorithms for Control Systems Engineeringpp 6/1–6/5

    1993

  7. [15]

    Nair V and Hinton G E 2010 Rectified linear units improve restricted boltzmann machines Proceedings of the 27th international conference on machine learning (ICML-10)pp 807–814

    2010

  8. [16]

    Kingma D P and Ba J 2014 arXiv preprint arXiv:1412.6980

    2014 arXiv

  9. [17]

    org/abs/2006.13318

    Cai C and Wang Y 2020 CoRR abs/2006.13318 (Preprint 2006.13318) URL https://arxiv. org/abs/2006.13318

    2020 arXiv

  10. [18]

    Rusch T K, Bronstein M M and Mishra S 2023 A Survey on Oversmoothing in Graph Neural Networks (Preprint 2303.10993) URL https://arxiv.org/abs/2303.10993

    2023 arXiv

  11. [19]

    FastML Team 2024 fastmachinelearning/hls4ml URL https://github.com/ 18 fastmachinelearning/hls4ml

    2024

  12. [20]

    AMD 2021 Vitis High-Level Synthesis User Guide: Introduction (UG1399)AMD accessed: 2025- 03-21 URL https://docs.amd.com/r/en-US/ug1399-vitis-hls/Introduction

    2021

  13. [21]

    Graph-Builder Constraint optimization using genetic algorithm The determination of graph-builder constraint values is driven by the need to optimize efficiency and purity

    Deb K, Pratap A, Agarwal S and Meyarivan T 2002 IEEE Transactions on Evolutionary Computation 6 182–197 19 Appendix A. Graph-Builder Constraint optimization using genetic algorithm The determination of graph-builder constraint values is driven by the need to optimize efficienc...

    2002

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.