Pith. sign in

REVIEW 6 major objections 5 minor 23 references

Bridging Vision and Language: Modeling Causality and Temporality in Video Narratives

T0 review · 6 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Inserting a Causal-Temporal Reasoning Module into large vision-language models improves the causal and temporal coherence of generated video captions, raising CIDEr to 18.20 on MSVD and 17.90 on MSR-VTT.

desk verdict The paper's causal mechanism is a mislabeled self-attention layer; undefined losses and a missing baseline leave the reported gains unexplained. read the letter →

arxiv 2412.10720 v1 pith:SABE7GBG submitted 2024-12-14 cs.CV

Ji-jun Park , Soo-joon Choi This is my paper · ORCID
classification cs.CV
keywords videocaptioninglargevision-languagemodelscausalreasoningtemporalCausal-TemporalModuleself-attentionMSVDMSR-VTT
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 argues that large vision-language models generate captions that are descriptively accurate but often miss the causal chain and temporal order of events, and that a bolt-on reasoning module can fix that. The module, called the Causal-Temporal Reasoning Module (CTRM), uses a self-attention Causal Dynamics Encoder to expose cause-effect links between frames and a transformer-based Temporal Relational Learner to preserve order; captions are then generated from the resulting temporal embeddings. The training recipe has three stages: pre-training on general video-text pairs, fine-tuning on a curated dataset with causal-temporal annotations, and a contrastive loss that aligns video embeddings with text embeddings. The reported numbers on MSVD and MSR-VTT put the full model ahead of every baseline on CIDEr, BLEU-4, and ROUGE-L, and the human evaluation rates it higher on fluency, coherence, and relevance.

What carries the argument

The load-bearing object is the Causal-Temporal Reasoning Module (CTRM), built from the Causal Dynamics Encoder (CDE) and the Temporal Relational Learner (TRL). The CDE forms a causal attention matrix $A_c = \mathrm{softmax}(Q_cK_c^\top / \sqrt{d_k})$ over frame features and produces causal embeddings $H_c = A_cV_c$; the TRL then applies a transformer encoder with positional encodings, $H_t = \mathrm{Transformer}(H_c + P)$, to keep the events in order. These temporal embeddings, rather than the original frame features, are what the autoregressive caption decoder conditions on, and the same embeddings are aligned to text with a contrastive loss. This two-stage bottleneck is the mechanism claimed to carry the causal and temporal information into the generated caption.

What would settle it

Re-run the same three-stage training while deleting the two auxiliary training terms, or replacing the specially annotated dataset with an ordinary captioning dataset, and see whether the score advantages on MSVD and MSR-VTT remain; if they disappear, the causal-temporal mechanism is what carries the result, and if they stay, it is not.

Watch

Extended reading notes

Core claim

The central result is that causal and temporal structure can be pulled out of raw frame features by a two-part module, and that conditioning the caption decoder on that structure, rather than on raw frame features, is what makes generated narratives coherent. The CDE computes a causal self-attention matrix over frames; the TRL passes the output through a transformer with positional encodings; and the decoder samples each word conditioned on the resulting temporal embeddings. The authors attribute the benchmark gains to this mechanism, and their ablation supports the attribution: removing either submodule drops the MSVD CIDEr score from 18.20 to 17.10 or 16.80, and removing the whole CTRM drops it to 15.74, the same as the plain Transformer baseline. Human ratings follow the same pattern, with the full model scoring 4.4 on fluency, 4.5 on coherence, and 4.6 on relevance out of 5.

Load-bearing premise

The reported gains depend on a curated dataset annotated with causal and temporal relationships and on two auxiliary training terms that the paper names but never defines, releases, or quantifies; if that dataset and those terms are not actually usable, the fine-tuning stage credited for the improvement has no describable content.

Editorial extensions

If this is right

  • If the CTRM result is correct, video captioning can gain narrative coherence without replacing the underlying vision-language backbone, because the module is added on top of existing frame features.
  • The ablation on MSVD implies that both submodules pull their weight: removing the CDE costs 1.10 CIDEr, removing the TRL costs 1.40 CIDEr, and removing the whole module loses 2.46 CIDEr.
  • The human evaluation implies that the improvement is visible to people, not just to automatic metrics, since annotators rate the full model above every baseline on fluency, coherence, and relevance.
  • The timing table implies that the extra reasoning comes at a small cost, with training time only slightly above the plain Transformer and inference time still 0.20 seconds per sample.

Reading between the lines

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

  • A natural test the paper does not run is to replace the curated causal-temporal fine-tuning data with ordinary captioning data while keeping CTRM; if the gains persist, the module alone, not the annotations, is doing the work.
  • The same two-part design could transfer to other tasks where event ordering matters, such as story ending generation or temporal grounding of narrated instructions.
  • Because the auxiliary losses $L_\text{causal}$ and $L_\text{temporal}$ are only named, not specified, a reader cannot currently tell whether the contrastive alignment loss alone would reproduce the reported gains.
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

6 major / 5 minor

Summary. The paper proposes a Causal-Temporal Reasoning Module (CTRM) for video captioning, consisting of a Causal Dynamics Encoder (CDE) and a Temporal Relational Learner (TRL), integrated into a large vision-language model. The training strategy is described as three-stage: pre-training on video-text data, fine-tuning on a curated dataset with causal-temporal annotations, and contrastive alignment. Experiments on MSVD and MSR-VTT report CIDEr, BLEU-4, and ROUGE-L scores, supplemented by ablations, human evaluations, and analyses of causality and temporal consistency.

Significance. If the proposed mechanism were correctly specified and the results reproducible, the paper would offer a modest contribution by showing that an explicit causal-temporal module plus auxiliary losses improves narrative coherence in video captioning. However, the manuscript does not define the auxiliary losses, does not describe the curated dataset, reports single-run numbers without variance or significance tests, and the central equation named "causal attention" is not causally masked. As a result, the paper's significance cannot be assessed from the information provided, and the current claims are not supported by the evidence.

major comments (6)
  1. [§III-B1, Eq. (2)-(3)] The "causal attention" in Eq. (2) is an unmasked softmax over all T key positions, so H_c at any time step aggregates information from future frames and the matrix is not causal. The name "causal" is therefore not justified by the equations; if causality is introduced only through L_causal and L_temporal, those terms must be defined (see next comment).
  2. [§III-D2, Eq. (7)] L_causal and L_temporal are never defined; the weights λ1 and λ2 and the temperature τ in Eq. (8) are also unspecified. Without definitions of these losses and of the annotations used to compute them, the fine-tuning stage has no describable content, and the improvements in Table I cannot be attributed to causal-temporal supervision.
  3. [§III-D, Eq. (9)] The final loss L = L_pre + L_fine + L_contrast is presented as a single objective, which contradicts the claimed three-stage training strategy (pre-training, fine-tuning, contrastive alignment). It is unclear whether these losses are optimized jointly or sequentially, and the reappearance of L_pre in the final loss suggests that pre-training and fine-tuning are not actually separated as stages.
  4. [Table I and §IV-A] The quantitative results are reported as single numbers with no standard deviations, no number of runs, no significance tests, and no dataset split details. Moreover, the CIDEr values (18.20 on MSVD, 17.90 on MSR-VTT) are far below typical published values on these benchmarks, and the paper does not comment on this discrepancy. The comparison is also incomplete: NarrativeBridge [4], the closest prior work cited in the introduction, is omitted from the baseline table.
  5. [§III-D2 and §IV-E] The curated dataset enriched with causal and temporal annotations is not described: its size, source, annotation protocol, and train/test split are all absent. Similarly, the manual annotation used for the causality and temporal consistency analyses in Tables IV and V is not described (number of annotators, inter-annotator agreement, sample size). The human evaluation in Table III also lacks these details, so the reliability of these analyses cannot be assessed.
  6. [§III-A and §IV-A] The large vision-language model backbone is never specified. The text says "such as Flamingo or BLIP-2" but does not state which model is actually used, how frame features are extracted, what resolution or frame rate is used, or the parameter count. This makes the experimental setup irreproducible even in principle.
minor comments (5)
  1. [Abstract and running text] There are numerous typographical and spacing errors, e.g., "descripti ve" in the abstract and "fine-tuning" with a broken ligature; these should be corrected in revision.
  2. [§II-B, reference [17]] The text attributes a live video captioning framework to "Blanco-Fern´ andez et al." but the cited reference [17] is by W. Choi and J. Yoon; the citation does not match the description.
  3. [Table II] The "Without CTRM" row reproduces the same numbers as the "Transformer" baseline in Table I; this should be explicitly acknowledged or clarified.
  4. [§III-C, Eq. (5)] The caption decoder conditions on H_t, but the paper does not describe how the decoder attends to H_t (e.g., cross-attention layer, number of layers, fusion method); a specification is needed.
  5. [§IV-E, Tables IV and V] The metrics "Captions with Causality (%)" and "Consistent Captions (%)" are not defined; the numerator, denominator, and annotation criteria should be stated.

Circularity Check

2 steps flagged · score 5.0 of 10

Partial circularity: 'causal attention' (Eq. 2) is unmasked full self-attention, the TRL (Eq. 4) is a standard transformer encoder, and Eq. 7's causal-temporal losses are undefined; the causal-temporal claim thus reduces to a renaming of generic transformer blocks, as confirmed by Table II 'Without CTRM' equaling the Table I 'Transformer' baseline.

  1. renaming known result [Section III-B, Eqs. (2)-(4); ablation evidence in Table II vs Table I.]
    "Given frame features V, the CDE employs a self-attention mechanism to model the causal structure. Specifically, a causal attention matrix Ac is computed as: Ac = softmax(QcKc^T/√dk) (2)... The TRL captures temporal dependencies across video frames. Using the causal embeddings Hc, the TRL applies a transformer-based encoder: Ht = Transformer(Hc + P) (4), where P represents positional encodings to preserve temporal order."

    Eq. (2) is the standard scaled dot-product attention with the softmax over all T key positions and no causal mask, so each frame's 'causal' embedding already depends on future frames; Eq. (4) is a vanilla transformer encoder. The paper's claim that CDE and TRL 'explicitly encode causal dependencies and temporal consistency' is therefore a renaming of generic self-attention plus positional encoding. The paper's own ablation confirms the reduction: 'Without CTRM 15.74 42.0 45.8' (Table II) is numerically identical to the 'Transformer 15.74 42.0 45.8' baseline (Table I). Tables IV-V attribute causality and temporal-consistency gains to this renamed block, so the causal-mechanism claim reduces, by the paper's own equations, to generic attention.

  2. fitted input called prediction [Section III-D2, Eq. (7); Section IV-E1, Table IV.]
    "the fine-tuning loss combines the captioning loss with auxiliary losses: Lfine = Lcaption + λ1Lcausal + λ2Ltemporal (7)... Lcausal promotes the alignment of causal features, and Ltemporal ensures temporal consistency. ... we manually annotated a subset of captions from the MSVD dataset with explicit causal relationships. We measured the ratio of captions that successfully captured the causal connections present in the video content."

    The only route by which causal-temporal content enters the model is fine-tuning on 'a curated dataset enriched with causal and temporal annotations' (Sec. III-D2) via Lcausal and Ltemporal, yet neither loss is defined, quantified, or released. The causal evaluation (Table IV) then manually annotates generated captions for exactly the causal-relationship construct the model is claimed to have been trained on. 'Captions with Causality (%)' therefore re-measures the undisclosed training signal rather than an emergent capability: the reported 72.5% vs. 65.2% advantage cannot be separated from direct fitting to the annotation criterion and is presented as prediction while being, if the claimed fine-tuning occurred, the trained objective itself.

full rationale

The headline benchmark results (Table I: CIDEr 18.20 on MSVD, 17.90 on MSR-VTT) are measured against external datasets and external baselines, and no parameter is fit to a subset of benchmark captions and then re-predicted, so the core numerical claims are not forced by construction; that ordinarily grounds a low score. There is also no self-citation chain: none of the twenty references is authored by Park or Choi, and no uniqueness theorem is imported. The circularity is in the central mechanistic claim. Eq. (2) is named the 'causal attention matrix' but is the standard unmasked softmax attention — the softmax runs over all T key positions, so the Causal Dynamics Encoder is ordinary full self-attention; Eq. (4) is a standard transformer encoder with positional encodings. The attribution that CDE/TRL 'explicitly encode causal dependencies and temporal consistency' is therefore a renaming of generic transformer blocks, and the paper's own ablation makes the reduction exact: 'Without CTRM 15.74 42.0 45.8' (Table II) is identical to the 'Transformer 15.74 42.0 45.8' baseline (Table I). The only path by which genuine causal-temporal content could enter — Lcausal and Ltemporal in Eq. (7) and the 'curated dataset enriched with causal and temporal annotations' in Section III-D2 — is never specified: the losses are undefined, the dataset is neither described nor released, and Section III-E's Eq. (9) sums Lpre + Lfine + Lcontrast into one objective, contradicting the claimed three-stage schedule. Because the causality and temporal-consistency analyses (Tables IV-V) manually annotate captions for causal relationships and event order — the same constructs the model is claimed to have been trained on — those 'predictions' collapse into re-measurement of the undisclosed training signal. This is partial circularity of the renaming and fitted-input kinds: the central causal-temporal narrative claim reduces to naming, but the external benchmark results retain independent content, so the score is 5 rather than 6-8.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on an unvalidated assumption that self-attention captures causality, an undescribed causal-temporal dataset, and undefined auxiliary losses. These are not independently supported and are load-bearing for the reported results.

free parameters (3)
  • lambda1 (causal loss weight)
    Introduced in Eq. (7) as a weight on the undefined Lcausal; the paper gives no value, and the central training objective depends on it.
  • lambda2 (temporal loss weight)
    Same as lambda1; weight on undefined Ltemporal in Eq. (7).
  • tau (contrastive temperature)
    Temperature in the contrastive loss (Eq. 8); standard but no value given.
assumptions (3)
  • domain assumption Self-attention over frame features encodes causal relationships (Eq. 2-3).
    The paper labels Ac a 'causal attention matrix' but provides no causal masking or structural constraint; the assumption that unconstrained self-attention captures causality is stated without evidence.
  • ad hoc to paper A curated dataset with causal-temporal annotations exists and can be used for fine-tuning (Section III-D2).
    The dataset is claimed but never described, quantified, or released; the entire fine-tuning stage rests on it.
  • ad hoc to paper The auxiliary losses Lcausal and Ltemporal can be computed (Eq. 7).
    These losses are never defined; the training objective in Eq. (9) is therefore underspecified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bridging Vision and Language: Modeling Causality and Temporality in Video Narratives." pith.science (2026). https://pith.science/paper/SABE7GBG

@misc{pith2026241210720,
  author       = {Pith},
  title        = {Pith review of: Bridging Vision and Language: Modeling Causality and Temporality in Video Narratives},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SABE7GBG}},
  note         = {Machine review of arXiv:2412.10720}
}
read the original abstract

Video captioning is a critical task in the field of multimodal machine learning, aiming to generate descriptive and coherent textual narratives for video content. While large vision-language models (LVLMs) have shown significant progress, they often struggle to capture the causal and temporal dynamics inherent in complex video sequences. To address this limitation, we propose an enhanced framework that integrates a Causal-Temporal Reasoning Module (CTRM) into state-of-the-art LVLMs. CTRM comprises two key components: the Causal Dynamics Encoder (CDE) and the Temporal Relational Learner (TRL), which collectively encode causal dependencies and temporal consistency from video frames. We further design a multi-stage learning strategy to optimize the model, combining pre-training on large-scale video-text datasets, fine-tuning on causally annotated data, and contrastive alignment for better embedding coherence. Experimental results on standard benchmarks such as MSVD and MSR-VTT demonstrate that our method outperforms existing approaches in both automatic metrics (CIDEr, BLEU-4, ROUGE-L) and human evaluations, achieving more fluent, coherent, and relevant captions. These results validate the effectiveness of our approach in generating captions with enriched causal-temporal narratives.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 5 canonical work pages

  1. [4]

    Narrativebridge: Enhancing video caption ing with causal-temporal narrative,

    A. Nadeem, F. Sardari, R. Dawes, S. S. Husain, A. Hilton, and A. Mustafa, “Narrativebridge: Enhancing video caption ing with causal-temporal narrative,” CoRR, vol. abs/2406.06499, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2406.06499

    arXiv doi:10.48550/arxiv.2406.06499 2024

  2. [1]

    Vid2seq: Large-scale pretraining of a visual language model for dense video captioning,

    A. Y ang, A. Nagrani, P . H. Seo, A. Miech, J. Pont-Tuset, I. Laptev, J. Sivic, and C. Schmid, “Vid2seq: Large-scale pretraining of a visual language model for dense video captioning,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, V anc ouver , BC, Canada, June 17-24, 2023 . IEEE, 2023, pp. 10 714–10 726. [Online]. Availabl...

    arXiv 2023

  3. [2]

    Triple sequence generativ e adversarial nets for unsupervised image captioning,

    Y . Zhou, W. Tao, and W. Zhang, “Triple sequence generativ e adversarial nets for unsupervised image captioning,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal P rocessing (ICASSP). IEEE, 2021, pp. 7598–7602

    2021

  4. [3]

    Sketch storytelling,

    Y . Zhou, “Sketch storytelling,” in ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICA SSP). IEEE, 2022, pp. 4748–4752

    2022

  5. [5]

    Rethinking visual de pendency in long-context reasoning for large vision-language models,

    Y . Zhou, Z. Rao, J. Wan, and J. Shen, “Rethinking visual de pendency in long-context reasoning for large vision-language models, ” arXiv preprint arXiv:2410.19732, 2024

    arXiv 2024

  6. [6]

    Visual in-context le arning for large vision-language models,

    Y . Zhou, X. Li, Q. Wang, and J. Shen, “Visual in-context le arning for large vision-language models,” in Findings of the Association for Computational Linguistics, ACL 2024, Bangkok, Thailand an d virtual meeting, August 11-16, 2024 . Association for Computational Linguis- tics, 2024, pp. 15 890–15 902

    2024

  7. [7]

    Are we on the right way fo r evaluating large vision-language models?

    L. Chen, J. Li, X. Dong, P . Zhang, Y . Zang, Z. Chen, H. Duan, J. Wang, Y . Qiao, D. Lin, and F. Zhao, “Are we on the right way fo r evaluating large vision-language models?” CoRR, vol. abs/2403.20330,

    arXiv

  8. [8]

    Visionllm v2: An end-to-end generalist multimodal large language model for hundreds of vision-language tasks,

    J. Wu, M. Zhong, S. Xing, Z. Lai, Z. Liu, W. Wang, Z. Chen, X. Zhu, L. Lu, T. Lu, P . Luo, Y . Qiao, and J. Dai, “Visionllm v2: An end-to-end generalist multimodal large language model for hundreds of vision-language tasks,” CoRR, vol. abs/2406.08394, 2024. [Online]. Available: https://doi.org/10.48550/arXiv.2406.08394

    arXiv doi:10.48550/arxiv.2406.08394 2024

Show all 23 references
  1. [9]

    A su rvey on multimodal large language models,

    S. Yin, C. Fu, S. Zhao, K. Li, X. Sun, T. Xu, and E. Chen, “A su rvey on multimodal large language models,” National Science Review , p. nwae403, 2024

    2024

  2. [10]

    Multimodal event transformer for i mage-guided story ending generation,

    Y . Zhou and G. Long, “Multimodal event transformer for i mage-guided story ending generation,” in Proceedings of the 17th Conference of the European Chapter of the Association for Computational Ling uistics, 2023, pp. 3434–3444

    2023

  3. [11]

    Fine-tuning large vision-language models as decision-making agents via rein forcement learning,

    Y . Zhai, H. Bai, Z. Lin, J. Pan, S. Tong, Y . Zhou, A. Suhr, S. Xie, Y . LeCun, Y . Ma, and S. Levine, “Fine-tuning large vision-language models as decision-making agents via rein forcement learning,” CoRR, vol. abs/2405.10292, 2024. [Online]. Available: https://doi.org/10.48...

    2024 doi arXiv

  4. [12]

    Style-aware contrastive learning for multi-style image captioning,

    Y . Zhou and G. Long, “Style-aware contrastive learning for multi-style image captioning,” in Findings of the Association for Computational Linguistics: EACL 2023 , 2023, pp. 2257–2267

    2023

  5. [13]

    Improving cross-modal alignment for text-guided image inpaint- ing,

    ——, “Improving cross-modal alignment for text-guided image inpaint- ing,” in Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics , 2023, pp. 3445–3456

    2023

  6. [14]

    Internlm-xcomposer-2. 5: A versatile large vision language model supporting long-con textual input and output,

    P . Zhang, X. Dong, Y . Zang, Y . Cao, R. Qian, L. Chen, Q. Guo, H. Duan, B. Wang, L. Ouyang, S. Zhang, W. Zhang, Y . Li, Y . Gao, P . Sun, X. Zhang, W. Li, J. Li, W. Wang, H. Y an, C. He, X. Zhang, K. Chen, J. Dai, Y . Qiao, D. Lin, and J. Wang, “Internlm-xcomposer-2. 5: A ve...

    2024 doi arXiv

  7. [15]

    Thread of thought unraveling chaotic contexts,

    Y . Zhou, X. Geng, T. Shen, C. Tao, G. Long, J.-G. Lou, and J. Shen, “Thread of thought unraveling chaotic contexts,” arXiv preprint arXiv:2311.08734, 2023

    2023 arXiv

  8. [16]

    Streaming dense video captioning,

    X. Zhou, A. Arnab, S. Buch, S. Y an, A. Myers, X. Xiong, A. N agrani, and C. Schmid, “Streaming dense video captioning,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVP R 2024, Seattle, WA, USA, June 16-22, 2024 . IEEE, 2024, pp. 18 243–18 252. [Online]....

    2024

  9. [17]

    Livecap: Live video captioning wit h sequential encoding network,

    W. Choi and J. Y oon, “Livecap: Live video captioning wit h sequential encoding network,” in 13th International Conference on Information and Communication Technology Convergence, ICTC 2022, Jeju Island, Korea, Republic of, October 19-21, 2022 . IEEE, 2022, pp. 1894–1896. [Onl...

    2022

  10. [18]

    Ret rieval enhanced zero-shot video captioning,

    Y . Ma, L. Qing, G. Li, Y . Qi, Q. Z. Sheng, and Q. Huang, “Ret rieval enhanced zero-shot video captioning,” CoRR, vol. abs/2405.07046,

  11. [19]

    Accura te and fast compressed video captioning,

    Y . Shen, X. Gu, K. Xu, H. Fan, L. Wen, and L. Zhang, “Accura te and fast compressed video captioning,” in IEEE/CVF International Conference on Computer Vision, ICCV 2023, Paris, France, Oc tober 1-6, 2023 . IEEE, 2023, pp. 15 512–15 521. [Online]. Available: https://doi.org/1...

    2023

  12. [20]

    Available: https://doi.org/10.48550/ar Xiv.2405.07046

    [Online]. Available: https://doi.org/10.48550/ar Xiv.2405.07046

    doi

  13. [22]

    Video captioning with aggre gated features based on dual graphs and gated fusion,

    Y . Jin, B. Liu, and J. Wang, “Video captioning with aggre gated features based on dual graphs and gated fusion,” CoRR, vol. abs/2308.06685,

    arXiv

  14. [2023]

    Available: https://doi.org/10.48550/ar Xiv.2308.06685 6

    [Online]. Available: https://doi.org/10.48550/ar Xiv.2308.06685 6

    doi arXiv

  15. [2024]

    Available: https://doi.org/10.48550/ar Xiv.2403.20330

    [Online]. Available: https://doi.org/10.48550/ar Xiv.2403.20330

    doi arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.