Pith. sign in

REVIEW 4 major objections 4 minor 27 references

GeoChain: Multimodal Chain-of-Thought for Geographic Reasoning

T0 review · 4 major / 4 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 GeoChain shows state-of-the-art multimodal models can name street objects but cannot consistently turn those cues into a location.

desk verdict Useful large-scale geolocation reasoning benchmark that overclaims 'robust' until the MaskFormer-derived labels are validated; deserves review with required revisions. read the letter →

arxiv 2506.00785 v3 pith:RX6MGNLI submitted 2025-06-01 cs.AI cs.CVcs.LG

Sahiti Yerramilli , Nilay Pande , Rynaa Grover , Jayant Sravan Tamarapalli This is my paper · ORCID
classification cs.AIcs.CVcs.LG
keywords geographicreasoningmultimodallargelanguagemodelschain-of-thoughtbenchmarkinggeolocalizationstreet-levelimagerysemanticsegmentationlocatabilityscorevisualgrounding
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

GeoChain is a benchmark that tries to establish how multimodal models reason geographically, step by step, rather than only judging final guesses. It builds on roughly 1.46 million street-level images, each paired with a fixed 21-question chain running from hemisphere and continent down to city, language, and coordinates, producing over 30 million question-answer pairs. On a curated 2,088-image test set, the paper finds a consistent pattern: leading multimodal models answer simple visual questions well, with pass scores above 91 percent, but accuracy collapses as questions demand precise location, and mean coordinate errors range from hundreds to thousands of kilometres. The intended payoff is diagnostic: by watching where in the chain accuracy fails, researchers can separate visual grounding problems from missing world knowledge or weak spatial chaining.

What carries the argument

The load-bearing mechanism is the static 21-step chain-of-thought question sequence, identical for every image, that moves from binary visual checks such as presence of boats, traffic lights, or flags to free-text prediction of latitude and longitude. Each question is tagged by reasoning category, visual cues, spatial localization, culture and infrastructure, terrain and environment, or exact location, and by an Easy, Medium, or Hard difficulty label, so the benchmark can report exactly where in the chain models fall. Supporting this structure are semantic segmentation maps generated by a transformer-based MaskFormer model pretrained on ADE20K with 150 classes, which provide ground-truth answers for visual questions and the class-area percentages used to compute a visual locatability score; that score, adapted from GeoReasoner, stratifies images into Easy, Medium, and Hard tiers. Final coordinate answers are scored by Haversine distance, with under 50 kilometres counting as a pass for the chain score.

What would settle it

Take a random sample of GeoChain images, have human annotators label whether each of the 150 semantic classes is present, and compare those labels with the MaskFormer-derived ground truth; if agreement on the visual questions is low, or if re-scoring models against human labels changes the relative rankings, the paper's diagnostic conclusions would need revision.

Watch

Extended reading notes

Core claim

The central claim is that current multimodal large language models are not yet geographically grounded: they can perceive and name street-scene objects, but they cannot reliably convert those perceptions into a location. GeoChain's 21-step chain makes this disconnect visible as a performance cliff, with pass rates falling from the high 80s on early visual questions to below 60 percent on hard reasoning questions, and coordinate predictions routinely missing by hundreds of kilometres. The paper further argues that broad inferential ability and precise localization are separable skills: the model with the best overall step-by-step score is not the model with the smallest coordinate error, and models that answer intermediate questions well can still locate very poorly. Difficulty annotations on both images and questions compound each other, so the hardest condition, hard questions paired with hard images, pushes several models below 40 percent pass rate.

Load-bearing premise

The benchmark's correct answers and difficulty tiers rest on automatic semantic segmentation labels rather than human verification, so if the segmentation model mislabels objects, the ground truth, the difficulty ranking, and the model scores all inherit that error.

Editorial extensions

If this is right

  • If the diagnosis is right, improving geolocation is not mainly a matter of better object recognition; the bottleneck sits downstream, in turning recognized cues into spatial inferences and grounded world knowledge.
  • Question-level pass rates give model builders a targeted failure profile, showing that language, administrative region, city, and coordinate questions are where the chain most often breaks.
  • The 30 million question-answer pairs and semantic labels provide a fixed, apples-to-apples testbed for comparing future multimodal models under identical prompts.
  • Difficulty tiers built from the locatability score let evaluations separate sensitivity to visual ambiguity from sensitivity to reasoning complexity.
  • The consistent drop from binary to multiclass to free-text questions suggests that open-ended generation and factual recall, not perception, are the current limiting factors.

Reading between the lines

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

  • The paper does not claim that the chain-of-thought sequence itself improves localization, but a natural testable extension is to compare coordinate predictions with and without the 21 questions presented as an inference-time prompt.
  • The semantic labels could be repurposed as a hallucination probe by asking models whether classes absent from the segmentation map are present, yielding a direct false-positive rate that the benchmark's pass scores only indirectly capture.
  • Because the benchmark is static, it could serve as a training signal: fine-tuning on the fixed 21-step chain may transfer to better city-level accuracy, an outcome GeoChain does not itself evaluate.
  • The 25, 200, and 750 kilometre thresholds define an operational ladder for real-world geolocation such as disaster response, suggesting that a model passing only continent-level accuracy is not yet useful for ground-level triage.
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 / 4 minor

Summary. This paper presents GeoChain, a benchmark for multimodal geographic reasoning built on 1.46 million Mapillary street-level images. Each image is paired with a fixed 21-step chain-of-thought question sequence spanning visual, spatial, cultural, and exact-location categories, with difficulty labels, semantic segmentation from MaskFormer, and a locatability score. The authors curate a 2,088-image Test-Mini stratified into Easy, Medium, and Hard tiers and evaluate GPT-4.1, GPT-4.1-mini, Claude 3.7 Sonnet, Gemini 2.5 Flash, and Gemini 2.5 Pro, reporting pass scores, question-category and difficulty breakdowns, and Haversine localization errors. The paper claims that GeoChain offers a robust diagnostic methodology revealing weaknesses in visual grounding, reasoning consistency, and precise localization, especially as reasoning complexity increases.

Significance. GeoChain's scale, static question design, and public release are assets: the 30 million Q&A pairs and segmentation metadata could support many future evaluations, and the stepwise 21-question design is a useful departure from endpoint-only geolocation benchmarks. The paper is also transparent about several limitations. However, the central diagnostic claim is currently not fully supported: the ground-truth answers for the visual questions and the difficulty tiers both derive from the same unvalidated MaskFormer predictions, and the free-text scoring is opaque. These are fixable with a human-verified subset, transparent parsing, and uncertainty quantification, but they are central to the paper's stated contribution.

major comments (4)
  1. [§3.2–3.3] Ground truth for the visual-presence questions and the locatability score are not independent: both are computed from MaskFormer class-area percentages. A segmentation error, e.g., a missed 'boat' class, flips the reference answer for Question 1 and also changes the class-area weighted locatability score that assigns the image to an Easy, Medium, or Hard tier. The paper reports no human verification rate for these labels, and Section 6 only concedes that the score is 'contingent upon the accuracy of an upstream semantic segmentation model.' Because label errors can systematically alter both pass scores and the difficulty-stratified conclusions, the Abstract's claim of a 'robust diagnostic methodology' is not yet established. Please report per-class agreement with human annotation on a random subset, show that corrected labels do not change model rankings or the difficulty trend, or otherwise bound the error.
  2. [§4.1.2] The Pass Score uses 'exact match for free-text' with no parsing or normalization details. Country, state or province, city, and language answers admit many valid surface forms ('USA' vs 'United States', diacritics, multiple languages), so a naive exact-match criterion can penalize correct answers. The absence of a stated normalization or human adjudication weakens the per-category and per-question-type conclusions, including Table 7's claim that free-text is uniformly hardest. Please specify the matching procedure (case folding, alias lists, LLM-based judging with human audit) and report agreement on a sample.
  3. [§4.2] All reported metrics are point estimates from a single evaluation pass, with no confidence intervals, repeated runs, or variance estimates. Several headline comparisons are small: Gemini-2.5-Pro versus Gemini-2.5-Flash pass scores are 81.84% versus 79.77%, and GPT-4.1 versus Gemini-2.5-Flash city-level accuracy is 57.84% versus 55.71%. Without uncertainty quantification, the qualitative claims about model ordering and 'erratic reasoning' are not supported at the reported resolution. Please provide per-model standard errors (e.g., bootstrap over images or repeated API calls with different seeds and temperatures).
  4. [§3.4] The Test-Mini is drawn from only 24 MSLS cities, and the locatability thresholds [0.12,0.22), [0.22,0.45), and [0.45,0.6) are introduced without justification. The difficulty-stratified conclusions therefore rest on a small, city-skewed sample and hand-picked cutoffs. The observed monotone accuracy decline across tiers is evidence that the tiers carry signal, but the paper should report per-city and per-tier sample sizes and test the sensitivity of the main trends to threshold choices.
minor comments (4)
  1. [§6] The word 'Additionally' is misspelled as 'Additonally' in the Limitations section.
  2. [§3.2] The phrase 'alocatability score' should be 'a locatability score' for readability.
  3. [A.1.2] The system prompt says question indices are 0-indexed, while Table 4 lists ranks 1 through 21; please clarify the correspondence to avoid ambiguity in reproducing the evaluation.
  4. [References] Several listed references (e.g., Cao et al., 2025; Grover et al., 2025; Jadhav et al., 2025) are not cited in the main text or appendix; please either cite them or remove them from the reference list.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: GeoChain's benchmark measurements are independent of its constructed labels; the shared MaskFormer label source is a reliability caveat, not a circular reduction.

full rationale

GeoChain's central results are model pass scores and localization errors measured against fixed reference answers; none of these is fit to or defined by the benchmark's own outputs. The locatability score (Section 3.2) is explicitly adopted from external prior work (Li et al., 2024), and the difficulty thresholds are fixed score intervals, not parameters fit to model performance. Visual-question ground truth (Section 3.3) is derived from MaskFormer class areas, and the locatability score is computed from the same class-area percentages; this shared upstream source is a genuine measurement-reliability caveat, acknowledged in Section 6 ('our locatability score's precision is contingent upon the accuracy of an upstream semantic segmentation model'), but it does not make any model prediction equivalent to a fitted input or to the benchmark's definitions. The paper's self-authored references (Grover et al. 2025; Yerramilli et al. 2024a,b; Jain et al. 2023; Jadhav et al. 2025; Pande and Awate 2021) appear only in the bibliography and are not load-bearing in any derivation. Difficulty-stratified accuracy is a conditional measurement that could in principle have failed to show the expected trend, so observing lower accuracy on Hard-tier images is empirical support for the score rather than a tautology. No quoted equation or construction reduces a claimed prediction to the paper's own inputs.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The benchmark depends on three external assumptions (metadata accuracy, segmentation reliability, locatability validity) and two hand-set scoring parameters; no new physical or conceptual entities are introduced.

free parameters (2)
  • Locatability score difficulty thresholds = [0.12, 0.22, 0.45, 0.6]
    Hand-chosen boundaries for Hard/Medium/Easy tiers in Section 3.2; they determine the composition of Test-Mini and the difficulty-stratified results, with no validation against human-rated difficulty.
  • Pass Score localization radius = 50 km
    In Section 4.1.2, a final coordinate prediction is counted as correct if Haversine distance is under 50 km; this hand-set threshold directly shapes the Pass Score and model rankings.
assumptions (3)
  • domain assumption MSLS geo-tags and city metadata are accurate ground truth for country, city, and coordinates.
    Section 3.3 uses MSLS metadata for Q19-Q21 ground truth without error analysis or temporal checks.
  • domain assumption MaskFormer pre-trained on ADE20K reliably segments street-level scenes in all 24 cities.
    Section 3.1 and 3.3 derive visual-question ground truth and locatability scores from these segmentation maps; no accuracy validation on this domain is reported.
  • domain assumption The GeoReasoner locatability score is a valid proxy for human-perceived geolocation difficulty.
    Section 3.2 adopts the scoring method from Li et al. 2024 and uses it for tier stratification without an independent human study.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoChain: Multimodal Chain-of-Thought for Geographic Reasoning." pith.science (2026). https://pith.science/paper/RX6MGNLI

@misc{pith2026250600785,
  author       = {Pith},
  title        = {Pith review of: GeoChain: Multimodal Chain-of-Thought for Geographic Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RX6MGNLI}},
  note         = {Machine review of arXiv:2506.00785}
}
read the original abstract

This paper introduces GeoChain, a large-scale benchmark for evaluating step-by-step geographic reasoning in multimodal large language models (MLLMs). Leveraging 1.46 million Mapillary street-level images, GeoChain pairs each image with a 21-step chain-of-thought (CoT) question sequence (over 30 million Q&A pairs). These sequences guide models from coarse attributes to fine-grained localization across four reasoning categories - visual, spatial, cultural, and precise geolocation - annotated by difficulty. Images are also enriched with semantic segmentation (150 classes) and a visual locatability score. Our benchmarking of contemporary MLLMs (GPT-4.1 variants, Claude 3.7, Gemini 2.5 variants) on a diverse 2,088-image subset reveals consistent challenges: models frequently exhibit weaknesses in visual grounding, display erratic reasoning, and struggle to achieve accurate localization, especially as the reasoning complexity escalates. GeoChain offers a robust diagnostic methodology, critical for fostering significant advancements in complex geographic reasoning within MLLMs.

Figures

Figures reproduced from arXiv: 2506.00785 by the authors.

Figure 1
Figure 1. Components of a GeoChain instance: (Top-Left) Easy Mapillary Street-Level Sequences (MSLS) image with locatability score of 0.45. (Top-Right) Example chain-of-thought questions with difficulty indicators. (Bottom￾Left) Derived semantic segmentation map. (Bottom-Right) Extracted key semantic labels. Together, these elements enable step-by-step diagnostic evaluation of geographic reasoning. carefully selected images, … view at source ↗
Figure 2
Figure 2. Count of images per city, illustrating the city distribution within the GeoChain dataset. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Distribution of Locatability Scores in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Pass score (%) by city, highlighting the influence of geographical location on model accuracy. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: reveals a typical degradation pattern: All models perform well in the initial questions (1–9), which ask about visual or global cues such as vehi￾cles, hemisphere, or continent. These are relatively easy to infer on the basis of surface-level features. As the questions…
Figure 6
Figure 6. Figure 6: Mean Haversine distance (km) by city and model. Larger values indicate poor localization precision. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Model vs Question Type 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 14 canonical work pages

  1. [1]

    Dragomir Anguelov, Carole Dulong, Daniel Filip, Christian Frueh, Stéphane Lafon, Richard Lyon, Abhijit Ogale, Luc Vincent, and Josh Weaver. 2010. https://doi.org/10.1109/MC.2010.170 Google street view: Capturing the world at street level . Computer, 43(6):32--38

    work page doi:10.1109/mc.2010.170 2010

  2. [2]

    Relja Arandjelovic, Petr Gronat, Akihiko Torii, Tomas Pajdla, and Josef Sivic. 2016. NetVLAD : CNN architecture for weakly supervised place recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition ( CVPR ) , pages 5287--5297

    work page 2016

  3. [3]

    Ron Campos, Ashmal Vayani, Parth Parag Kulkarni, Rohit Gupta, Aritra Dutta, and Mubarak Shah. 2025. https://arxiv.org/abs/2503.16423 GAEA : A geolocation aware conversational model . Preprint, arXiv:2503.16423

    arXiv 2025

  4. [4]

    Schwing, and Alexander Kirillov

    Bowen Cheng, Alexander G. Schwing, and Alexander Kirillov. 2021. Per-pixel classification is not all you need for semantic segmentation. In Advances in Neural Information Processing Systems 34 ( NeurIPS 2021) , pages 17864--17875. Curran Associates, Inc

    work page 2021

  5. [5]

    Wenliang Dai, Junnan Li, Dongxu Li, Anthony Meng Huat Tiong, Junqi Zhao, Weisheng Wang, Boyang Li, Pascale Fung, and Steven Hoi. 2023. https://arxiv.org/abs/2305.06500 Instructblip: Towards general-purpose vision-language models with instruction tuning . Preprint, arXiv:2305.06500

    arXiv 2023

  6. [6]

    GeoGuessr. 2013. https://www.geoguessr.com/ Geoguessr - let's explore the world!

    work page 2013

  7. [7]

    Google. 2025 a . Gemini 2.5 flash is now in preview. https://blog.google/products/gemini/gemini-2-5-flash-preview/. Accessed 2025-05-18

    work page 2025

  8. [8]

    Google. 2025 b . Gemini 2.5 pro model. https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/\#gemini-2-5-thinking

    work page 2025

Show all 27 references
  1. [9]

    Muki Haklay and Patrick Weber. 2008. https://doi.org/10.1109/MPRV.2008.80 Openstreetmap: User-generated street maps . IEEE Pervasive Computing, 7(4):12--18

    2008 doi

  2. [10]

    James Hays and Alexei A. Efros. 2008. https://doi.org/10.1109/CVPR.2008.4587784 IM2GPS: estimating geographic information from a single image . In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition ( CVPR ) , pages 1--8. IEEE

    2008

  3. [11]

    Alexandre Lacoste, Nils Lehmann, Pau Rodriguez, Evan David Sherwin, Hannah Kerner, Björn Lütjens, Jeremy Andrew Irvin, David Dao, Hamed Alemohammad, Alexandre Drouin, Mehmet Gunturkun, Gabriel Huang, David Vazquez, Dava Newman, Yoshua Bengio, Stefano Ermon, and Xiao Xiang Zhu....

    2023 arXiv

  4. [12]

    Ling Li, Yu Ye, Bingchuan Jiang, and Wei Zeng. 2024. Georeasoner: Geo-localization with reasoning in street views using a large vision-language model. In International Conference on Machine Learning (ICML)

    2024

  5. [13]

    Yifan Li, Yifan Du, Kun Zhou, Jinpeng Wang, Xin Zhao, and Ji-Rong Wen. 2023. https://openreview.net/forum?id=xozJw0kZXF Evaluating object hallucination in large vision-language models . In The 2023 Conference on Empirical Methods in Natural Language Processing

    2023

  6. [14]

    OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff ...

    2024 arXiv

  7. [15]

    Saksham Pramanik, Aayush Mundra, Ashutosh Mittal, Sreyas Mohan, Saim Wani, Shramay S Vernekar, Pranav M Dixit, Shanti Priya, Ankur Beniwal, Ojaswa Sharma, and Senthil Mani. 2024. https://arxiv.org/abs/2502.14412 Evaluating precise geolocation inference capabilities of vision l...

    2024 arXiv

  8. [16]

    Anna Rohrbach, Lisa Anne Hendricks, Kaylee Burns, Trevor Darrell, and Kate Saenko. 2018. https://doi.org/10.18653/v1/D18-1437 Object hallucination in image captioning . In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4035--4045,...

    2018 doi

  9. [17]

    Zirui Song, Jingpu Yang, Yuan Huang, Jonathan Tonglet, Zeyu Zhang, Tao Cheng, Meng Fang, Iryna Gurevych, and Xiuying Chen. 2025. https://arxiv.org/abs/2502.13759 Geolocation with real human gameplay data: A large-scale dataset and human-like reasoning framework . Preprint, arX...

    2025

  10. [18]

    Claude 3.7 Sonnet. 2025. Claude 3.7 sonnet documentation. https://docs.anthropic.com/en/docs/overview. Anthropic AI Model

    2025

  11. [19]

    Gemini Team, Petko Georgiev, Ving Ian Lei, Ryan Burnell, Libin Bai, Anmol Gulati, Garrett Tanzer, Damien Vincent, Zhufeng Pan, Shibo Wang, Soroosh Mariooryad, Yifan Ding, Xinyang Geng, Fred Alcober, Roy Frostig, Mark Omernick, Lexi Walker, Cosmin Paduraru, Christina Sorokin, a...

    2024 arXiv

  12. [20]

    Yicong Tian, Chen Chen, and Mubarak Shah. 2017. https://doi.org/10.1109/CVPR.2017.216 Cross-view image matching for geo-localization in urban environments . In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 1998--2006

    2017 doi

  13. [21]

    Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024. https://arxiv.org/abs/2409.12191 Qwen2-v...

    2024 arXiv

  14. [22]

    Frederik Warburg, S ren Hauberg, Gregory D. D. Funke, and Yoko Yuki. 2020. Mapillary street-level sequences: A dataset for lifelong place recognition. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition ( CVPR ) Workshops , pages 931--940. IEEE

    2020

  15. [23]

    Tobias Weyand, Ilya Kostrikov, and James Philbin. 2016. https://doi.org/10.1007/978-3-319-46475-6_3 PlaNet - photo geolocation with convolutional neural networks . In Computer Vision -- ECCV 2016, Proceedings, Part II , volume 9906 of Lecture Notes in Computer Science, pages 3...

    2016 doi

  16. [24]

    Ruixiang Yang, Cheng Zhang, Lingxi Meng, He Wang, Xiaoyan Li, Yuke Li, Shuo Wang, Haoran Wei, Yiyang Li, Wentao Qu, Pengchuan Zhang, Jiazheng Xu, Bihan Wen, Diyi Yang, Kangkang Lu, Saurabh Gupta, Guanzhong Wang, Zhiqiang Shen, Baining Guo, and 3 others. 2024. https://arxiv.org...

    2024 arXiv

  17. [25]

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. 2017. Scene parsing through ADE20K dataset. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition ( CVPR ) , pages 633--641. IEEE

    2017

  18. [26]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  19. [27]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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