Pith. sign in

REVIEW 5 major objections 5 minor 24 references

Quantum Variational Transformer Model for Enhanced Cancer Classification

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that replacing a transformer's self-attention with 8-qubit variational quantum circuits lifts cancer classification accuracy to 92.8% and AUC to 0.96, while cutting training time by 35%.

desk verdict The central claim of quantum advantage is unverifiable because neither the quantum attention mechanism nor the classical baseline is defined, and the manuscript is littered with unresolved artifacts. read the letter →

arxiv 2506.21641 v1 pith:LHM6NSGZ submitted 2025-06-25 q-bio.QM

Don Roosan , Rubayat Khan , Md Rahatul Ashakin , Tiffany Khou , Saif Nirzhor , Mohammad Rifat Haider This is my paper · ORCID
classification q-bio.QM
keywords QuantumMachineLearningCancerClassificationHybridTransformerModelVariationalCircuitsGenomicDataAnalysisAttentionAmplitudeEncodingParameter-ShiftRule
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that replacing the self-attention layers of a transformer with quantum attention circuits improves cancer type and primary tumor site prediction. On 30,000 anonymized cancer samples, the hybrid model is reported to reach 92.8% accuracy and an AUC of 0.96, against 87.5% and 0.89 for the classical transformer, while training 35% faster with 25% fewer parameters. The authors attribute the gain to the quantum attention mechanism's ability to represent feature interactions through superposition and entanglement. If true, this would be a concrete example of quantum computing changing a biomedical machine-learning outcome.

What carries the argument

The load-bearing object is the variational quantum circuit (VQC) attention layer: a parameterized quantum circuit of 8 qubits that replaces the transformer's classical self-attention. After amplitude encoding maps a normalized feature vector $x$ to a quantum state $|\psi_x\rangle$, the circuit applies a unitary $U(\theta)$ to produce $|\phi_x\rangle = U(\theta)|\psi_x\rangle$, and the trainable angles $\theta$ are updated using the parameter-shift rule. This circuit is what the paper credits with capturing complex feature correlations through superposition and entanglement, and with reducing the parameter count and training time.

What would settle it

Retrain the classical transformer with the same 150 features, same training schedule, optimizer, and hyperparameter tuning as the quantum model; if its test accuracy reaches or exceeds 92.8%, the claimed quantum advantage would not hold. A second check is to run the quantum model on noisy quantum hardware rather than a simulator, since the paper reports most training used simulators.

Watch

Extended reading notes

Core claim

The central claim is that a hybrid quantum-classical transformer, with classical self-attention replaced by variational quantum circuits, classifies cancer samples more accurately and more efficiently than an otherwise comparable classical transformer. Input features are amplitude-encoded into quantum states and processed by an 8-qubit parameterized circuit, with quantum gradients computed by the parameter-shift rule and classical layers trained by backpropagation. On a held-out test set of 4,500 samples the paper reports 92.8% accuracy, AUC 0.96, macro-averaged F1 0.91, precision 0.93, and recall 0.89, compared with 87.5%, 0.89, 0.84, 0.85, and 0.83 for the classical model, and says the difference is statistically significant at p < 0.01. The paper also reports 35% shorter training time and 25% fewer parameters, and notes that most training was performed on simulators rather than noisy quantum hardware.

Load-bearing premise

The load-bearing premise is that the 'classical transformer model' used for comparison is a strong, fairly tuned control with matched architecture, features, and training budget, and the paper does not specify that baseline's architecture.

Editorial extensions

If this is right

  • On the reported dataset, the quantum-enhanced transformer outperforms the classical transformer by 5.3 percentage points in accuracy and 0.07 in AUC.
  • The hybrid model also improves macro-averaged F1, precision, and recall, from 0.84/0.85/0.83 to 0.91/0.93/0.89.
  • Training time falls by 35% and parameter count by 25%, which the authors connect to deployment in resource-limited clinical settings.
  • The paper reports the accuracy gain as statistically significant under a paired t-test, with $p < 0.01$.
  • Because most training used simulators, the authors state that future work should validate the model on noisy quantum systems.

Reading between the lines

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

  • We infer that the reported 35% training-time saving is not evidence about real quantum hardware: simulator timings depend on the classical simulation backend, so this claim should be re-measured on actual devices before being compared.
  • A natural ablation the paper does not report: replace the quantum attention layer with a classical attention layer of matched parameter count and measure the accuracy gap; that would isolate whether the quantum circuit itself, rather than the smaller parameter count, drives the gain.
  • The 8-qubit encoding has capacity for 256 amplitudes while the model uses 150 features; scaling qubit counts and feature dimensionality would test how much of the benefit comes from quantum expressivity versus model capacity.
  • The same quantum attention design could be transferred to other genomics tasks with long-range dependencies, such as mutation-effect prediction or single-cell type classification, to see whether the accuracy gain generalizes beyond this dataset.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper proposes a hybrid quantum-classical transformer model for cancer type and primary tumor site classification, in which classical self-attention modules are replaced by 8-qubit variational quantum circuits (VQCs). Using 30,000 anonymized cancer samples from the Genome Warehouse, the authors report that their quantum-enhanced transformer outperforms an unspecified 'classical transformer model' (92.8% vs 87.5% accuracy; AUC 0.96 vs 0.89) while reducing training time by 35% and parameter count by 25%. The methods section describes amplitude encoding, a generic parameterized unitary, cross-entropy loss, and the parameter-shift rule, but never specifies the quantum attention mechanism or the classical baseline architecture. The results section reports aggregate test-set metrics, a confusion matrix, ROC curves, and a paired t-test that is inconsistent with the described data split. The discussion concedes that most training was performed on simulators. No code, dataset identifiers, or detailed hyperparameters are provided.

Significance. If the central claim were supported, replacing classical self-attention with a quantum variational layer that improves both accuracy and training efficiency for a real biomedical classification task would be a notable result for the quantum machine learning and bioinformatics communities. The paper does not achieve this: the comparison to the classical baseline is uncontrolled, the quantum attention layer is not formally defined, the statistical test contradicts the evaluation protocol, and the efficiency claims rest on simulator training without measured runtimes. The manuscript provides no code, data identifiers, or reproducible architecture details, so the empirical results cannot be independently verified. The research question is interesting, but the evidence presented is insufficient to support the paper's conclusions.

major comments (5)
  1. [Section 2 (Methods)] The classical transformer baseline is never specified. The text introduces 'the classical transformer model' without giving its architecture, number of layers, hidden dimension, number of attention heads, regularization, parameter count, or training budget. Without a matched and clearly described baseline, the reported improvements in accuracy (92.8% vs 87.5%) and efficiency (35% faster training, 25% fewer parameters) cannot be attributed to the quantum attention layer rather than to differences in model capacity, regularization, or tuning effort.
  2. [Sections 1.2 and 2 (Quantum attention mechanism)] The paper does not define how the VQC implements attention. Equations (1)–(4) describe amplitude encoding, a generic unitary U(θ), cross-entropy loss, and the parameter-shift rule, but no equation or textual description explains how queries, keys, values, or attention weights are computed, nor how the VQC output is integrated with the rest of the transformer. If the VQC is merely a learned nonlinear feature map rather than an attention mechanism, the experiment compares two different architectures, and the claim of a 'quantum attention layer' is unsupported.
  3. [Section 3 (Results, statistical significance)] The statistical test description is inconsistent with the evaluation protocol. Section 2 states that the data was randomly partitioned into training (70%), validation (15%), and test (15%) subsets, with no mention of cross-validation. Section 3, however, claims that 'a paired t-test was conducted on the accuracy scores obtained from cross-validation folds for both models.' No number of folds, per-fold accuracy scores, or confidence intervals are reported, so the claimed p < 0.01 cannot be verified, and the test appears to refer to an experimental procedure that was not described.
  4. [Section 4 (Discussion) and Table 1] The efficiency claims are not supported by the reported data. The Discussion states that 'Most training used simulators,' yet the paper reports a 35% training-time reduction and 25% parameter reduction without providing wall-clock times, simulator backend, qubit noise model, or hardware details. Parameter-shift-rule gradients on an 8-qubit circuit do not trivially run faster than classical attention, and the parameter count of the classical baseline is never given, so the claimed reductions are uninterpretable.
  5. [Section 2 (Data and reproducibility)] The dataset is described only as '30,000 anonymized cancer samples from the Genome Warehouse (GWH), compiled from ten biomedical repositories,' without accession numbers, dataset identifiers, or download instructions. Combined with the absence of code and the missing architecture details, the experiments cannot be reproduced or independently checked, which is essential for a paper making empirical performance claims.
minor comments (5)
  1. [Section 3 (Results)] The Results section begins with the leftover fragments 'estimation..' and 'RMSD' before the first sentence; these appear to be artifacts from a previous draft and should be removed.
  2. [Section 3 (Figure 2)] Figure 2 is described as a 'Visualization of attention weights,' but the manuscript does not explain how these weights are extracted from the quantum circuit or what the axes, colors, or plotted values represent.
  3. [Section 2 (Methods)] The phrase 'achieving an imputation accuracy of 95.2%' is undefined; since missing values have no ground truth, the authors should specify how this accuracy was computed or remove the claim.
  4. [Equation (3)] Equation (3) is written as a sum over classes c but does not explicitly define y_c and ŷ_c within the equation; although the surrounding text mentions them, the notation should be made self-contained.
  5. [References] Several references are self-citations of the authors' prior work (e.g., [20]) that are not clearly connected to the specific claims they support; the authors should consider citing more independent sources for general statements about quantum machine learning and attention mechanisms.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracy, AUC, and efficiency gains are empirical held-out test measurements, not derived from or equivalent to the model's defining equations.

full rationale

This paper is an empirical benchmarking study rather than a derivation. The central claims (92.8% accuracy, AUC 0.96, 35% faster training, 25% fewer parameters) are reported as measurements on a held-out test set, with the classical transformer used as a comparison. Equations (1)-(4) only define amplitude encoding, a generic parameterized unitary U(θ), cross-entropy loss, and the parameter-shift rule; none of these equations entail the reported performance numbers, so there is no self-definitional reduction or fitted-input-called-prediction step. The self-citations, including refs [5]-[10] and especially ref [20], are used for background or contextual statements such as 'Quantum algorithms accelerate computations, enabling faster training and better handling of large datasets, improving model generalization vital for limited biomedical datasets [19], [20].' That citation is not load-bearing: the paper's own measured training-time reduction and parameter counts are the evidence for the efficiency claim, not the cited prior work. The significant weaknesses of the paper are rigor issues rather than circularity: the classical transformer baseline is never specified in terms of architecture or training budget, and the 'quantum attention' layer is asserted but not formally defined as an attention mechanism (no Q/K/V mapping or attention-score equation is given). These concerns mean the comparison may be uncontrolled, but they do not make the result equivalent by construction to its inputs. No circular step can be exhibited, so the appropriate score is 0.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim is empirical and depends on a long chain of unverified modeling choices: dataset provenance, preprocessing, feature selection, the quantum circuit, and the baseline. None of these are externally anchored with code or data.

free parameters (7)
  • Number of qubits = 8
    Chosen for the VQC attention circuit; no scaling or ablation justifies this value.
  • Number of selected features = 150
    Output of RFECV feature selection; threshold is a modeling choice.
  • Learning rate = 1e-4
    Optimizer hyperparameter for Adam.
  • Batch size = 64
    Training hyperparameter.
  • Number of epochs = 50
    Training duration; early stopping is mentioned but not specified.
  • Outlier exclusion thresholds = Z > 3, Mahalanobis p < 0.01
    Hand-selected outlier cutoffs; approximately 4% of samples excluded.
  • Train/validation/test split = 70/15/15
    Random split ratio; no seed given.
assumptions (5)
  • standard math Amplitude encoding maps normalized feature vectors to quantum states via Eq. (1).
    Standard quantum encoding; not the source of the claimed gain.
  • standard math Parameter-shift rule gives exact gradients for VQC parameters (Eq. 4).
    Standard training method for variational circuits.
  • domain assumption The 30,000-sample GWH dataset is accurately labeled for cancer type and primary tumor site.
    No independent label validation or data access is provided.
  • ad hoc to paper The 8-qubit VQC implements an attention-like function that is useful for classification.
    No circuit diagram, entanglement pattern, or output mapping is given, so this is an unverified modeling premise.
  • ad hoc to paper The classical transformer baseline is comparable to the quantum model in capacity and tuning.
    Baseline architecture is not described; the comparison assumes fair matching.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Variational Transformer Model for Enhanced Cancer Classification." pith.science (2026). https://pith.science/paper/LHM6NSGZ

@misc{pith2026250621641,
  author       = {Pith},
  title        = {Pith review of: Quantum Variational Transformer Model for Enhanced Cancer Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LHM6NSGZ}},
  note         = {Machine review of arXiv:2506.21641}
}
read the original abstract

Accurate prediction of cancer type and primary tumor site is critical for effective diagnosis, personalized treatment, and improved outcomes. Traditional models struggle with the complexity of genomic and clinical data, but quantum computing offers enhanced computational capabilities. This study develops a hybrid quantum-classical transformer model, incorporating quantum attention mechanisms via variational quantum circuits (VQCs) to improve prediction accuracy. Using 30,000 anonymized cancer samples from the Genome Warehouse (GWH), data preprocessing included cleaning, encoding, and feature selection. Classical self-attention modules were replaced with quantum attention layers, with classical data encoded into quantum states via amplitude encoding. The model, trained using hybrid backpropagation and quantum gradient calculations, outperformed the classical transformer model, achieving 92.8% accuracy and an AUC of 0.96 compared to 87.5% accuracy and an AUC of 0.89. It also demonstrated 35% faster training and 25% fewer parameters, highlighting computational efficiency. These findings showcase the potential of quantum-enhanced transformers to advance biomedical data analysis, enabling more accurate diagnostics and personalized medicine.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 12 canonical work pages

  1. [1]

    Global cancer burden growing, amidst mounting need for services

    “Global cancer burden growing, amidst mounting need for services.” Accessed: Nov. 30, 2024. [Online]. Available: https://www.who.int/news/item/01 -02-2024-global-cancer-burden-growing--amidst- mounting-need-for-services

    work page 2024

  2. [2]

    From Chaos to Opportunity: Decoding Cancer Heterogeneity for Enhanced Treatment Strategies,

    A. Ottaiano et al., “From Chaos to Opportunity: Decoding Cancer Heterogeneity for Enhanced Treatment Strategies,” Biology, vol. 12, no. 9, Art. no. 9, Sep. 2023, doi: 10.3390/biology12091183

    work page doi:10.3390/biology12091183 2023

  3. [3]

    Characterization of cancer genomic heterogeneity by next-generation sequencing advances precision medicine in cancer treatment,

    J. Zhang, S. S. Späth, S. L. Marjani, W. Zhang, and X. Pan, “Characterization of cancer genomic heterogeneity by next-generation sequencing advances precision medicine in cancer treatment,” Precis. Clin. Med., vol. 1, no. 1, pp. 29–48, Jun. 2018, doi: 10.1093/pcmedi/pby007

    work page doi:10.1093/pcmedi/pby007 2018

  4. [4]

    Virtual histopathology methods in medical imaging - a systematic review,

    M. T. Imran et al., “Virtual histopathology methods in medical imaging - a systematic review,” BMC Med. Imaging, vol. 24, no. 1, p. 318, Nov. 2024, doi: 10.1186/s12880-024-01498-9

    work page doi:10.1186/s12880-024-01498-9 2024

  5. [5]

    Integrating Artificial Intelligence with Mixed Reality to Optimize Health Care in the Metaverse,

    D. Roosan, “Integrating Artificial Intelligence with Mixed Reality to Optimize Health Care in the Metaverse,” in Augmented and Virtual Reality in the Metaverse , V. Geroimenko, Ed., Cham: Springer Nature Switzerland, 2024, pp. 247–264. doi: 10.1007/978-3-031-57746-8_13

    doi:10.1007/978-3-031-57746-8_13 2024

  6. [6]

    Pharmacogenomics cascade testing (PhaCT): a novel approach for preemptive pharmacogenomics testing to optimize medication therapy,

    D. Roosan, A. Hwang, and M. R. Roosan, “Pharmacogenomics cascade testing (PhaCT): a novel approach for preemptive pharmacogenomics testing to optimize medication therapy,” Pharmacogenomics J., vol. 21, no. 1, pp. 1–7, Feb. 2021, doi: 10.1038/s41397-020-00182-9

    doi:10.1038/s41397-020-00182-9 2021

  7. [7]

    Artificial Intelligent Context-Aware Machine-Learning Tool to Detect Adverse Drug Events from Social Media Platforms,

    D. Roosan, A. V. Law, M. R. Roosan, and Y. Li, “Artificial Intelligent Context-Aware Machine-Learning Tool to Detect Adverse Drug Events from Social Media Platforms,” J. Med. Toxicol. Off. J. Am. Coll. Med. Toxicol., vol. 18, no. 4, pp. 311–320, Oct. 2022, doi: 10.1007/s13181-022-00906-2

    doi:10.1007/s13181-022-00906-2 2022

  8. [8]

    Effectiveness of ChatGPT in clinical pharmacy and the role of artificial intelligence in medication therapy management,

    D. Roosan, P. Padua, R. Khan, H. Khan, C. Verzosa, and Y. Wu, “Effectiveness of ChatGPT in clinical pharmacy and the role of artificial intelligence in medication therapy management,” J. Am. Pharm. Assoc., Dec. 2023, doi: 10.1016/j.japh.2023.11.023

    doi:10.1016/j.japh.2023.11.023 2023

Show all 24 references
  1. [9]

    Applying Artificial Intelligence to create risk stratification visualization for underserved patients to improve population health in a community health setting,

    D. Roosan, M. R. Roosan, S. Kim, A. V. Law, and C. Sanine, “Applying Artificial Intelligence to create risk stratification visualization for underserved patients to improve population health in a community health setting,” May 20, 2022, Research Square. doi: 10.21203/rs.3.rs-1...

    2022 doi

  2. [10]

    Augmented Reality and Artificial Intelligence: Applications in Pharmacy,

    D. Roosan, “Augmented Reality and Artificial Intelligence: Applications in Pharmacy,” in Augmented Reality and Artificial Intelligence: The Fusion of Advanced Technologies , V. Geroimenko, Ed., in Springer Series on Cultural Computing. , Cham: Springer Nature Switzerland, 2023...

    2023 doi

  3. [11]

    From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment,

    K. Swanson, E. Wu, A. Zhang, A. A. Alizadeh, and J. Zou, “From patterns to patients: Advances in clinical machine learning for cancer diagnosis, prognosis, and treatment,” Cell, vol. 186, no. 8, pp. 1772– 1791, Apr. 2023, doi: 10.1016/j.cell.2023.01.035

    2023 doi

  4. [12]

    Deep learning in cancer diagnosis, prognosis and treatment selection,

    K. A. Tran, O. Kondrashova, A. Bradley, E. D. Williams, J. V. Pearson, and N. Waddell, “Deep learning in cancer diagnosis, prognosis and treatment selection,” Genome Med., vol. 13, p. 152, Sep. 2021, doi: 10.1186/s13073-021-00968-x

    2021 doi

  5. [13]

    Quantum computing at the frontiers of biological sciences,

    P. S. Emani et al., “Quantum computing at the frontiers of biological sciences,” Nat. Methods, vol. 18, no. 7, pp. 701–709, Jul. 2021, doi: 10.1038/s41592-020-01004-3

    2021 doi

  6. [14]

    Transformer Architecture and Attention Mechanisms in Genome Data Analysis: A Comprehensive Review,

    S. R. Choi and M. Lee, “Transformer Architecture and Attention Mechanisms in Genome Data Analysis: A Comprehensive Review,” Biology, vol. 12, no. 7, p. 1033, Jul. 2023, doi: 10.3390/biology12071033

    2023 doi

  7. [15]

    The state of quantum computing applications in health and medicine,

    F. F. Flöther, “The state of quantum computing applications in health and medicine,” Res. Dir. Quantum Technol., vol. 1, p. e10, Jan. 2023, doi: 10.1017/qut.2023.4

    2023 doi

  8. [16]

    Quantum Computing in the Next -Generation Computational Biology Landscape: From Protein Folding to Molecular Dynamics,

    S. Pal, M. Bhattacharya, S. -S. Lee, and C. Chakraborty, “Quantum Computing in the Next -Generation Computational Biology Landscape: From Protein Folding to Molecular Dynamics,” Mol. Biotechnol., vol. 66, no. 2, pp. 163–178, Feb. 2024, doi: 10.1007/s12033-023-00765-4

    2024 doi

  9. [17]

    Future Potential of Quantum Computing and Simulations in Biological Science,

    S. Pal, M. Bhattacharya, S. Dash, S. -S. Lee, and C. Chakraborty, “Future Potential of Quantum Computing and Simulations in Biological Science,” Mol. Biotechnol., vol. 66, no. 9, pp. 2201–2218, Sep. 2024, doi: 10.1007/s12033-023-00863-3

    2024 doi

  10. [18]

    Transition Role of Entangled Data in Quantum Machine Learning,

    X. Wang, Y. Du, Z. Tu, Y. Luo, X. Yuan, and D. Tao, “Transition Role of Entangled Data in Quantum Machine Learning,” May 12, 2024, arXiv: arXiv:2306.03481. doi: 10.48550/arXiv.2306.03481

    2024 doi arXiv

  11. [19]

    Quantum Machine Learning Applications in the Biomedical Domain: A Systematic Review | IEEE Journals & Magazine | IEEE Xplore

    “Quantum Machine Learning Applications in the Biomedical Domain: A Systematic Review | IEEE Journals & Magazine | IEEE Xplore.” Accessed: Nov. 30, 2024. [Online]. Available: https://ieeexplore.ieee.org/document/9844718

    2024

  12. [20]

    Utilizing Quantum Computing -based Large Language Transformer Models to Identify Social Determinants of Health from Electronic Health Records,

    D. Roosan, J. Chok, Y. Li, and T. Khou, “Utilizing Quantum Computing -based Large Language Transformer Models to Identify Social Determinants of Health from Electronic Health Records,” Jul. 2024, pp. 1–6. doi: 10.1109/ICECET61485.2024.10698600

    2024

  13. [21]

    Supporting novice clinicians cognitive strategies: System design perspective,

    R. Islam, M. Roosan, J. Mayer, and J. Clutter, “Supporting novice clinicians cognitive strategies: System design perspective,” in IEEE -EMBS International Conference on Biomedical and Health Informatics, 2016, p. 509, doi: 10.1109/BHI.2016.7455946

    2016

  14. [22]

    Clinical complexity in medicine: A measurement model of task and patient complexity,

    R. Islam, C. Weir, and G. Del Fiol, “Clinical complexity in medicine: A measurement model of task and patient complexity,” Methods Inf. Med., vol. 55, no. 1, pp. 14–22, 2016, doi: 10.3414/ME15-01-0031

    2016 doi

  15. [23]

    Heuristics in managing complex clinical decision tasks in experts’ decision making,

    R. Islam, C. Weir, and G. Del Fiol, “Heuristics in managing complex clinical decision tasks in experts’ decision making,” in Proc. IEEE Int. Conf. Healthc. Inform., 2014, pp. 186 –193, doi: 10.1109/ICHI.2014.32

    2014 doi

  16. [24]

    Understanding complex clinical reasoning in infectious diseases for improving clinical decision support design,

    R. Islam, C. R. Weir, M. Jones, G. Del Fiol, and M. H. Samore, “Understanding complex clinical reasoning in infectious diseases for improving clinical decision support design,” BMC Med. Inform. Decis. Mak., vol. 15, p. 101, 2015, doi: 10.1186/s12911-015-0221-z

    2015 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.