Pith. sign in

REVIEW 5 major objections 7 minor 25 references

Enhancing Multi-Robot Semantic Navigation Through Multimodal Chain-of-Thought Score Collaboration

T0 review · 5 major / 7 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read A local 9B vision-language model with a shared semantic map can beat a remote GPT-3.5 centralized planner in multi-robot object search.

desk verdict The reported system is plausible and the paper is well-specified, but Eq. 3 makes the history-return decision impossible as written, so the central contribution is currently not well-defined. read the letter →

arxiv 2412.18292 v5 pith:JEZ3RNUX submitted 2024-12-24 cs.RO

classification cs.RO
keywords multi-robotsemanticnavigationobjectgoalzero-shotvision-languagemodelchain-of-thoughtpromptingmappingfrontier-basedexplorationdecentralizedplanning
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 proposes MCoCoNav, a decentralized planning framework in which several robots searching for a target object in an unseen home share a single top-down semantic map and use a locally deployed vision-language model to score how promising each view and each candidate frontier is. The central claim is that this modular, zero-shot design beats both a centralized GPT-3.5 planner (Co-NavGPT) and trained exploration baselines on the HM3D v0.2 and MP3D benchmarks, while keeping communication cost low because the shared map is the only message exchanged. On the two-robot HM3D v0.2 evaluation with predicted segmentation, MCoCoNav reports a success-weighted path length (SPL) of 0.387 and a success rate of 0.716, compared with Co-NavGPT's 0.331 and 0.661. If the claim holds, it means a 9-billion-parameter quantized local model can replace a remote large model for multi-robot object search without losing performance.

What carries the argument

The load-bearing object is the cross-image multimodal chain-of-thought: three VLM calls (Perception, Judgment, Decision) that consume both the robot's first-person image and the annotated shared semantic map, with each call's text output becoming part of the next prompt. The navigation decision is carried by the VLM's token probability of 'Yes' (Eqs. 1 and 2), temperature-scaled with $\tau_{ES}=2.0$ and $\tau_{JS}=1.0$ and combined into the horizontal-field-of-view score of Eq. 3, whose 0.5 threshold routes each robot either to explore a frontier or to revisit a history node. A history-score algorithm accumulates these scores into 360 directional bins and normalizes by exploration count, so the global semantic map acts simultaneously as the communication channel, the scoreboard, and the source of the visual prompt for the Decision VLM.

What would settle it

Re-run the two-robot HM3D v0.2 evaluation with the exact hyperparameters of Table 4 on scenes never used for prompt or threshold selection, and compare SPL and success rate against Co-NavGPT and Random Sampling; if the margin collapses to within episode-level noise, the zero-shot transfer claim fails. A second check is to record the distributions of 'Yes' token probabilities in successful versus failed episodes: if the two distributions do not separate around 0.5, the score is not behaving as an exploration-value estimate.

Watch

Extended reading notes

Core claim

The paper's discovery claim is that the next-token 'Yes' probability of a local vision-language model, carried through a chain of three prompts, can act as a stable exploration-value score for multi-robot navigation. The Perception module turns a first-person RGB view into a spatial-relationship description; the Judgment module reads an annotated global semantic map and produces a judgment score; the two are temperature-scaled and combined into a horizontal-field-of-view score $HFOVS=\exp(\tau_{ES}ES+\tau_{JS}JS)$ (Eq. 3), and a 0.5 threshold decides whether the robot continues to frontier points or returns to the highest-scoring history node. The Decision module then selects among labeled frontier points by normalized token probabilities over 'A', 'B', 'C', 'D'. With two robots and predicted segmentation, this procedure reports SPL 0.387 and SR 0.716 on HM3D v0.2 against Co-NavGPT's 0.331 and 0.661, and SPL 0.334 and SR 0.568 on MP3D. The paper argues that this specific combination of perception, judgment, and decision scores is what makes zero-shot local multi-robot semantic navigation work.

Load-bearing premise

The load-bearing premise is that a VLM's next-token probability of 'Yes' is a stable, well-calibrated measure of exploration value across different scenes, so the fixed 0.5 threshold and the temperature scales $\tau_{ES}=2.0$, $\tau_{JS}=1.0$ transfer to new environments without per-scene retuning.

Editorial extensions

If this is right

  • Multi-robot object-goal navigation in unseen homes can be done entirely from local inference, without task-specific reinforcement learning and without remote foundation-model calls.
  • Communication cost drops to the cost of maintaining one shared semantic map, because the map carries all the information robots exchange.
  • Adding more robots continues to raise success rate and SPL, but with diminishing returns and rising detection traps, so two robots give the best efficiency trade-off on HM3D v0.2.
  • With ground-truth semantic segmentation, the reported success rate for two robots on HM3D v0.2 reaches 0.872, implying that detector and segmentation quality, rather than VLM reasoning, is the current bottleneck.
  • Failure analysis attributes most remaining failures to detection rather than planning or exploration, so better object detection and 3D scan quality would improve the method more than further prompt engineering.

Reading between the lines

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

  • Editorial extension: Replacing the VLM's 'Yes' probabilities with a fixed or random score while keeping the map, history-node algorithm, and frontier selection would test whether the gains come from VLM semantics or from the cost-geometry of the shared map; the paper's ablations remove whole modules but never replace the score with a non-semantic baseline.
  • Editorial extension: The 0.5 threshold and temperature scales are constants tuned on the same benchmark family, so the system's zero-shot claim would be stronger if the threshold were shown to be stable across scene types, for instance from bedroom-dominated episodes to kitchen-dominated episodes.
  • Editorial extension: The cross-image chain-of-thought pattern, one VLM describing a local view, another reading an annotated global map, and a third choosing among labeled candidates, could transfer to other multi-agent tasks such as search-and-rescue exploration where shared map scores and local language models matter.
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

5 major / 7 minor

Summary. The manuscript proposes MCoCoNav, a decentralized multi-robot object-goal navigation system built around a local quantized 9B VLM (GLM-4V-9B). A Perception module obtains an exploration score ES as the VLM's probability of "Yes" for a scene view; a Judgment module obtains JS from the global top-down semantic map annotated with history nodes and frontiers; Eq. (3) combines these into a horizontal field-of-view score HFOVS, thresholded at 0.5 to choose between exploring frontier points and returning to history nodes; a Decision module selects one of A-D frontier points; a Logical Analysis module handles continuation and collision traps; FMM provides local actions. Experiments on HM3D v0.2 and MP3D report SPL/SR improvements over Greedy, Cost-Utility, Random Sampling, Multi-SemExp, and Co-NavGPT, plus ablations and robot-number/semantic-accuracy studies.

Significance. If substantiated, the main result is notable: a local, quantized 9B VLM with a shared semantic map beats a remote GPT-3.5 centralized planner (Co-NavGPT) on HM3D v0.2 (SPL 0.387 vs 0.331; SR 0.716 vs 0.661), with code released and no task-specific training. The cross-image CoT prompting and map-as-communication-bridge design are interesting, and the modular ablation story is coherent. However, the published decision rule is mathematically vacuous as written, the reported numbers lack variance and statistical support, and the communication-cost claim is not measured; the contribution is therefore conditional on correcting and re-validating the central mechanism.

major comments (5)
  1. [Eq. (3), §Method (Global Map Exploration Judgment), Table 4] Because ES and JS are next-token probabilities in [0,1] and τ_ES=2.0, τ_JS=1.0, HFOVS = exp(τ_ES·ES + τ_JS·JS) always lies in [1, e^3] ≈ [1,20.1]. The decision rule "If HFOVS ≥ 0.5 ... otherwise return to history nodes" therefore always takes the frontier-exploration branch; the history-return branch and the Logical Analysis condition "HFOVS < 0.5" are unreachable. Since the Decision/History switching is the mechanism that the ablations in Table 3 are meant to validate, the paper as written does not specify the algorithm it evaluates. Please correct Eq. (3) or the threshold (or both) to match the code and report whether the history-return branch fires in any episode.
  2. [Table 4, §Experimental Setup] τ_ES=2.0, τ_JS=1.0, δ=25, and the 0.5 threshold are fixed constants, but the reported HM3D v0.2 results are on the same validation split used to choose them; no sensitivity analysis or hold-out split is provided. Because raw VLM probabilities are not calibrated, the absolute value of ES/JS is prompt- and scene-dependent, so the claimed zero-shot gains may partly reflect fitted constants. Please report validation/test separation and a sweep over τ_ES, τ_JS, and the threshold.
  3. [Table 1, Table 3] All metrics are reported as single numbers with no error bars, number of seeds, or significance tests, yet the Table 1 caption states MCoCoNav "significantly outperforms all baseline methods." Given the small margins over Random Sampling on MP3D and over Co-NavGPT on HM3D, please provide mean±std across multiple runs (at least 3 seeds) and/or paired tests per episode.
  4. [§B.2 Computational Resources] The paper asserts that "physical time for communication between the robots is negligible" without measuring it; the experiments run in a single Habitat process, so no inter-robot communication bandwidth or latency is modeled. The introduction and abstract claim reduced communication overhead, but the global map "bridge" is an architectural choice whose cost is never quantified. Please specify the communication model and provide measurements or a bound.
  5. [Table 1] Co-NavGPT, the strongest centralized LLM baseline, is reported only on HM3D v0.2; the MP3D comparison therefore does not include the main competing method. The claim that MCoCoNav outperforms "all baseline methods on all metrics" across both datasets is not supported by the table as presented; please run the strongest baselines on MP3D or qualify the claim.
minor comments (7)
  1. [Table 2 and §Metrics] DT (Detection Trap) is used as a metric but is never defined in the experimental setup; please define it and explain how it is counted.
  2. [Appendix A.2, Algorithm 1] EXP is used as a denominator (line 21) but is never initialized or updated in the pseudocode, so the algorithm is underspecified for the first visit to a location.
  3. [Eq. (3)] The term "temperature scaling" is nonstandard here: taking exp of a probability-weighted sum does not implement softmax temperature scaling; please clarify the intended transformation.
  4. [Table 3] The checkmark rows are ambiguous because the column alignment makes some rows identical in appearance; please use explicit module names or a legend for each ablation row.
  5. [Table 6 and §B.3] The success-distance criterion differs across single-robot baselines (Pixel-Nav uses 1.0m while others use 0.2m); please make the comparison protocol explicit.
  6. [Figures 5 and 6] The failure-mode percentages are reported without denominators or error bars, making it hard to assess the reliability of the failure-mode analysis.
  7. [Related Work] There are copyediting issues such as "V oronav" in the Related Work section and "Jug" in the Table 3 caption; please revise.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the pipeline's outputs are empirical VLM scores and map statistics, not re-derivations of its inputs.

full rationale

MCoCoNav is an empirical navigation pipeline rather than a derived model, so the circularity patterns enumerated do not apply. The ES, JS, HFOVS, HS, and decision scores are computed from frozen VLM probabilities and hand-specified arithmetic; no equation reconstructs an input from an output. The temperature-scaling values (τ_ES=2.0, τ_JS=1.0) are stated as hyperparameters without any claim that they were fit to the HM3D test set, so charging fitted-input-called-prediction would be speculation. The paper contains no load-bearing self-citations and no uniqueness theorem imported from the authors. The publicly stated equations do contain a serious internal-consistency defect: Eq. (3) defines HFOVS = exp(τ_ES·ES + τ_JS·JS), which is always ≥1 for ES,JS∈[0,1], so the 'if HFOVS ≥ 0.5' test always selects frontier exploration and the history-return branch is unreachable. This is a mathematical/correctness problem with the written algorithm, but it is not a circular derivation: the reported performance is still an empirical output of a VLM ranking procedure, not an input reconstructed from the output. Accordingly, the circularity score is 0.

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

The central claim rests on four free parameters, all hand-set in Table 4, and on the assumption that VLM probabilities are calibrated exploration signals. The paper introduces no new physical or architectural entities; the history score and cross-image chain-of-thought are algorithmic constructs, not independently falsifiable entities.

free parameters (4)
  • Temperature scaling tau_ES = 2.0
    Weights the exploration score in Eq. (3); value chosen in Table 4, likely tuned on HM3D v0.2 validation.
  • Temperature scaling tau_JS = 1.0
    Weights the judgment score in Eq. (3); hand-set in Table 4.
  • HFOVS threshold = 0.5
    Determines whether a robot explores frontiers or returns to history nodes; no principled derivation is given.
  • Long-term goal update interval delta = 25 pixels
    Dictates when navigation goals are re-evaluated; set in Table 4 without sensitivity analysis.
assumptions (4)
  • domain assumption GLM-4V-9B token probabilities for 'Yes' are calibrated exploration-value estimates without fine-tuning.
    The Perception and Judgment modules treat raw next-token probabilities as scores; used in Eqs. (1) to (3).
  • domain assumption A shared semantic map can be treated as a communication bridge with negligible communication cost.
    Stated in the Introduction and Method, but Appendix A.1 requires normalizing and merging all robots' maps into a common coordinate system.
  • ad hoc to paper History score Algorithm 1's division by exploration count yields a useful revisiting priority.
    Algorithm 1 defines HS = SUM(A)/EXP; no theoretical or empirical justification is given beyond the ablation.
  • standard math Frontier-based exploration on the projected 2D map is a valid way to discover goals.
    Inherited from Yamauchi's frontier exploration, cited as Yamauchi 1997.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Multi-Robot Semantic Navigation Through Multimodal Chain-of-Thought Score Collaboration." pith.science (2026). https://pith.science/paper/JEZ3RNUX

@misc{pith2026241218292,
  author       = {Pith},
  title        = {Pith review of: Enhancing Multi-Robot Semantic Navigation Through Multimodal Chain-of-Thought Score Collaboration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JEZ3RNUX}},
  note         = {Machine review of arXiv:2412.18292}
}
read the original abstract

Understanding how humans cooperatively utilize semantic knowledge to explore unfamiliar environments and decide on navigation directions is critical for house service multi-robot systems. Previous methods primarily focused on single-robot centralized planning strategies, which severely limited exploration efficiency. Recent research has considered decentralized planning strategies for multiple robots, assigning separate planning models to each robot, but these approaches often overlook communication costs. In this work, we propose Multimodal Chain-of-Thought Co-Navigation (MCoCoNav), a modular approach that utilizes multimodal Chain-of-Thought to plan collaborative semantic navigation for multiple robots. MCoCoNav combines visual perception with Vision Language Models (VLMs) to evaluate exploration value through probabilistic scoring, thus reducing time costs and achieving stable outputs. Additionally, a global semantic map is used as a communication bridge, minimizing communication overhead while integrating observational results. Guided by scores that reflect exploration trends, robots utilize this map to assess whether to explore new frontier points or revisit history nodes. Experiments on HM3D_v0.2 and MP3D demonstrate the effectiveness of our approach. Our code is available at https://github.com/FrankZxShen/MCoCoNav.git.

Figures

Figures reproduced from arXiv: 2412.18292 by the authors.

Figure 1
Figure 1. Examples of (a) VLM multimodal CoT reasoning [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Components of MCoCoNav. The MCoCoNav architecture consists of Visual Foundation Models, the MCoCoNav Planner and the Local Policy Planner. At its core is the MCoCoNav Planner, which is composed of three main components: the Perception module, the Judgment module and the Decision module. the ESt i,p for the current pose p: ESt i,p = ˆf”Y es”(I t i , V LM(I t i , IP ) + O t i + IE) (1) Global Map Exploration Judgment … view at source ↗
Figure 3
Figure 3. Workflow of cross-image multimodal CoT. Information from multiple images is unified into a multimodal CoT through VLM. At each time step, the next action for each robot and the global semantic map are updated based on the observed scene objects. The cross-image multimodal CoT enables semantic label alignment for navigation planning, environmental understanding, and common-sense reasoning. the current position. If ro… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: 2-Robot navigation episode with MCoCoNav in HM3D v0.2. The top row shows the robots’ RGB scene views and the global semantic map. The bottom row shows the VLM Visual Prompt used for Decision (large image) and the VLM Visual Prompt used for Judgment (small image). Best …
Figure 5
Figure 5. Figure 5: Percentage of failure cases for 2-robot in dif￾ferent baselines. Our MCoCoNav experienced the fewest Exploration failures. the multi-robot of (Chaplot et al. 2020), where the exper￾iment is set up with two robots collaboratively exploring the environment. The task is c…
Figure 6
Figure 6. Figure 6: Percentage distribution of failure cases in dif￾ferent numbers of robots. The percentage of detection fail￾ures increases with the number of robots. as illustrated in [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Perception Instruction Template [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Exploration Instruction Template [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]
Figure 9
Figure 9. Figure 9: Judgment Instruction Template [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Decision Instruction Template [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 21 canonical work pages

  1. [1]

    **Object List:** - [Other objects if any]

  2. [2]

    **Spatial Relationships:** - **[Describe other objects if any]**

  3. [3]

    - Mention any items on top of furniture or decorations on the walls

    **Additional Context:** - Detail any significant aspects of the lighting, colors, textures, and overall ambiance of the room. - Mention any items on top of furniture or decorations on the walls. ------ Perception Instruction Figure 7: Perception Instruction Template. Based on your description, you need to determine whether the current scene is worth explo...

  4. [4]

    In Proceedings of the IEEE/CVF international confer- ence on computer vision , 15130–15140

    Hierarchical object-to-zone graph for object naviga- tion. In Proceedings of the IEEE/CVF international confer- ence on computer vision , 15130–15140. Zhang, Z.; Zhang, A.; Li, M.; and Smola, A. 2022. Auto- matic chain of thought prompting in large language models. arXiv preprint arXiv:2210.03493. Zhang, Z.; Zhang, A.; Li, M.; Zhao, H.; Karypis, G.; and S...

  5. [10]

    **Examine the Relationship Between the Target Object and the Scene:** - If the object detection model predicts the presence of the target object with a high probability (e.g., >85%), determine if the scene is worth exploring

  6. [11]

    - Use your knowledge of typical object locations to evaluate the likelihood of the target object being present (e.g., beds are usually found in bedrooms, TVs in living rooms)

    **Analyze the Scene Context:** - Assess the context of the scene, including ceilings, walls, floors, or windows. - Use your knowledge of typical object locations to evaluate the likelihood of the target object being present (e.g., beds are usually found in bedrooms, TVs in living rooms)

  7. [12]

    - Ensure your judgment does not violate the high probability criterion from point (1)

    **Considering Object Proximity and Context:** - Evaluate the proximity of the target object to the scene in the image. - Ensure your judgment does not violate the high probability criterion from point (1). For instance, a bathtub is unlikely in a bedroom, but the presence of a door could indicate the target object might be nearby

  8. [13]

    [Yes, No]

    **Disregard Generic Objects:** - Ignore objects commonly found in various rooms (like light switches and doors) as they do not provide strong evidence for the target object’s presence. ------ **Output Format:** Your output should be a simple "[Yes, No]" statement indicating whether the scene is worth exploring based on the given criteria. Exploration Inst...

Show all 25 references
  1. [14]

    **Level of Exploration:** - If there are a very small number of Historical Observation Points and a large number of gaps in the semantic map, prefer exploring Frontier Points

  2. [15]

    **Explorability Worthiness:** - Scenes that are worth exploring ({ISWORTH}) are usually more likely to be explored by choosing Frontier Points

  3. [16]

    Frontier Points or Historical Observation Points that are close and free of obstacles tend to have higher exploration priority

    **Proximity and Accessibility:** - Evaluate how Your location relates to surrounding obstacles. Frontier Points or Historical Observation Points that are close and free of obstacles tend to have higher exploration priority

  4. [17]

    [Yes, No]

    **Relationship Between Location and Previous Movement:** - If Your location is too close to Previous Movement, it may indicate a collision trap. In such cases, prefer to explore Historical Observation Points that are close to Your location. **Decision Format:** Your recommenda...

  5. [18]

    Frontier Points that are close and free of obstacles tend to have higher exploration priority

    **Proximity and Accessibility:** - Evaluate how Your location relates to surrounding obstacles. Frontier Points that are close and free of obstacles tend to have higher exploration priority

  6. [19]

    Prefer to explore Frontier Points that are farther from Your location

    **Relationship Between Location and Previous Movement:** - If Your location is too close to Previous Movement, it may indicate the robot is entering a collision trap. Prefer to explore Frontier Points that are farther from Your location

  7. [20]

    The robot should maintain its exploration direction unless an efficient switch is evident

    **Exploration Consistency:** - Minimize frequent switches between Frontier Points. The robot should maintain its exploration direction unless an efficient switch is evident

  8. [21]

    **Target-Oriented Exploration:** - If there are accessible Frontier Points in the direction of Your location (red arrow), and the scenario information includes Your Navigation Target, give the highest priority to exploring these Frontier Points without violating point (1). **Y...

  9. [22]

    **Proximity and Accessibility:** Frontier Point A at (8, 9) is relatively close and free of obstacles

  10. [23]

    **Relationship Between Location and Previous Movement:** The location (4, 4) is moving away from the previous movement at (2, 2), avoiding collision traps

  11. [24]

    **Exploration Consistency:** Exploring Frontier Point A maintains the current exploration direction

  12. [25]

    **Recommendation:** I am likely to choose Frontier Point A for exploration

    **Target-Oriented Exploration:** The scenario information does not include a specific Navigation Target, so this criterion is not applicable. **Recommendation:** I am likely to choose Frontier Point A for exploration. ------ Now, begin your analysis with the provided scene inf...

  13. [2019]

    In Proceedings of the IEEE/CVF international conference on computer vision, 9339–9347

    Habitat: A platform for embodied ai research. In Proceedings of the IEEE/CVF international conference on computer vision, 9339–9347. Sethian, J. A. 1999. Fast marching methods. SIAM review, 41(2): 199–235. Shah, D.; Equi, M. R.; Osi ´nski, B.; Xia, F.; Ichter, B.; and Levine, ...

  14. [2021]

    arXiv preprint arXiv:2109.08238

    Habitat-matterport 3d dataset (hm3d): 1000 large- scale 3d environments for embodied ai. arXiv preprint arXiv:2109.08238. Ramrakhya, R.; Batra, D.; Wijmans, E.; and Das, A. 2023. Pirlnav: Pretraining with imitation and rl finetuning for ob- jectnav. In Proceedings of the IEEE/...

  15. [2023]

    Advances in Neural Information Processing Systems, 36: 5168–5191

    Ddcot: Duty-distinct chain-of-thought prompting for multimodal reasoning in language models. Advances in Neural Information Processing Systems, 36: 5168–5191. Zhou, K.; Zheng, K.; Pryor, C.; Shen, Y .; Jin, H.; Getoor, L.; and Wang, X. E. 2023. Esc: Exploration with soft commo...

  16. [2024]

    arXiv preprint arXiv:2406.12793

    ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools. arXiv preprint arXiv:2406.12793. Ho, N.; Schmid, L.; and Yun, S.-Y . 2022. Large lan- guage models are reasoning teachers. arXiv preprint arXiv:2212.10071. Jiang, J.; Zheng, L.; Luo, F.; and Zhang, Z....

  17. [4090]

    Since the robots communicate using a global semantic map, we consider the physical time for communication between the robots to be negligible

    for the experiments, which can effectively reduce the evaluation runtime. Since the robots communicate using a global semantic map, we consider the physical time for communication between the robots to be negligible. B.3 Baseline Details • Greedy (Visser et al. 2013): The Gree...

Pith tools

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