pith. sign in

arxiv: 2605.24573 · v1 · pith:XRXBNPR3new · submitted 2026-05-23 · 💻 cs.CL

AstroMind: A High-Fidelity Benchmark for Spacecraft Behavior Reasoning Based on Large Language Models

Hao Liu , Siyuan Yang , Qinglei Hu , Dongyu Li This is my paper

Pith reviewed 2026-06-30 13:08 UTC · model grok-4.3

classification 💻 cs.CL
keywords spacecraft behavior reasoninglarge language modelsbenchmarkastrodynamics simulationintent inferencemaneuver estimationthreat assessmentspace domain awareness
0
0 comments X

The pith

AstroMind turns high-fidelity spacecraft simulations into verifiable LLM reasoning tasks for intent, maneuvers, and threats.

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

The paper presents AstroMind as a benchmark that converts astrodynamics simulations and observational constraints into three types of reasoning problems: inferring spacecraft intent, estimating maneuver parameters, and assessing threats. Each scenario adds realistic sensing noise and textual intelligence of varying reliability, then scores answers for both semantic accuracy and consistency with physical laws. This setup tests whether current language models can combine tactical reading with quantitative physics checks, a gap the authors say existing detection pipelines leave open. Results across open-weight models show task-specific strengths rather than uniform superiority, with prompt structure and data mix affecting outcomes as much as scale.

Core claim

AstroMind is a physics-grounded benchmark designed to close that gap. It draws on high-fidelity astrodynamics simulations and real observational constraints, converting them into verifiable reasoning problems across three task types: intent inference, maneuver parameter estimation, and threat assessment. Each scenario includes realistic sensing noise and multi-source textual intelligence at varying reliability levels. Evaluation metrics capture both semantic correctness and quantitative consistency under physical constraints.

What carries the argument

Conversion of high-fidelity astrodynamics simulations plus sensing noise and multi-source intelligence into three task families (intent inference, maneuver parameter estimation, threat assessment) scored on semantic and physical-consistency metrics.

If this is right

  • No single open-weight model leads on every task; Qwen3-32B leads intent inference while QwQ-32B leads threat assessment and error metrics.
  • Structured reasoning prompts improve results consistently, with larger gains for models already able to track physical constraints.
  • Training data composition and reasoning style affect performance at least as much as raw model size.
  • The benchmark supplies a shared test where both physics fidelity and tactical interpretation must hold for success.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same simulation-to-text pipeline could be adapted to test LLMs on other constrained physical domains such as air-traffic or maritime traffic analysis.
  • If models improve on AstroMind, they may become usable for initial triage of real orbital events before human analysts review the raw data.
  • The finding that prompt structure helps suggests targeted fine-tuning on physics-consistent reasoning traces could close remaining gaps.

Load-bearing premise

High-fidelity simulations with added noise and mixed-reliability text can be turned into problems that actually measure an LLM's ability to reason under physical constraints.

What would settle it

Models achieve near-random accuracy or produce answers that violate basic orbital mechanics on a large fraction of the benchmark items even when given the full scenario data.

Figures

Figures reproduced from arXiv: 2605.24573 by Dongyu Li, Hao Liu, Qinglei Hu, Siyuan Yang.

Figure 1
Figure 1. Figure 1: The AstroMind Framework. Four-tier pipeline from astrodynamics to scored LLM tasks. Tier 1 produces physical ground truth via the Poliastro dynamics core; Tier 2 adds sensor noise and simulated intelligence reports; Tier 3 formats the result into structured tasks (Intent Inference, Parameter Es￾timation, Threat Assessment); Tier 4 scores responses with multi-dimensional metrics and the Reasoning Loop. imme… view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchical Taxonomy of Spacecraft Behaviors in AstroMind. Eight primary categories and 29 subcategories, derived from historical incidents and operational doctrine. The range runs from routine behaviors (Station-keeping, Deployment) to high-stakes non-cooperative activities (RPO, Kinetic ASAT)— each subcategory maps to a distinct, observable motion signature, so intent labels are physically anchored rath… view at source ↗
Figure 3
Figure 3. Figure 3: AstroMind Benchmark Scenario Distribution. Horizontal bar chart showing scenario counts across the 8 primary categories (133 scenarios, 399 questions total). Mission Phases and Non-cooperative Activities are jointly the largest categories (25 scenarios each, 18.8%)—these are the behaviors where misclassification carries the highest operational cost, so their weight in the benchmark is deliberate. Each bar … view at source ↗
Figure 4
Figure 4. Figure 4: The “Reasoning Loop” Architecture. An iterative inference scaffold that enforces physical consistency through four phases: (1) Condition Parsing— extracting explicit and implicit constraints; (2) Strategy Planning—deriving applicable formulas before computation; (3) Step-wise Execution with per-step unit/dimensionality checks; and (4) Reflection and Refinement, where a physical-plausibility check can trigg… view at source ↗
Figure 5
Figure 5. Figure 5: Intent Inference Accuracy vs. Model Parameter Size. Three patterns from Section IV are visible: training data composition matters (GPT-OSS at 20B is competitive with larger general-purpose models); a threshold-like scaling pattern is observed within the families tested here (Gemma and Qwen scaling trends); and reasoning-oriented models vary widely, with QwQ (32B) well ahead of smaller reasoning baselines. … view at source ↗
Figure 7
Figure 7. Figure 7: Task Accuracy and Judged Reasoning Quality Among Reasoning￾Oriented Models. Horizontal axis: average judged score (logic, physics, completeness). Vertical axis: intent inference accuracy. Bubble size: threat assessment accuracy. Text labels: Median RE, computed on model-specific parsed subsets—read alongside “Parsed N” in Table III. Judged reasoning quality and task accuracy are correlated but not the same… view at source ↗
Figure 8
Figure 8. Figure 8: Mixed Effects of the Reasoning Loop on 8B Models. Raw-value trajectories across four metrics (Intent Accuracy, Threat Accuracy, Median RE, Logic Score) with and without the Reasoning Loop. Qwen3 (•), DeepSeek-R1 (■), and Llama-3 (▲) show mixed effects: most metrics improve, but Llama-3’s Median RE rises slightly (0.9931→0.9966) and Qwen3’s Threat Accuracy dips marginally (45.45%→44.71%). Gains for DeepSeek… view at source ↗
read the original abstract

Understanding why a spacecraft maneuvers -- rather than simply that it did -- is an increasingly important problem for space domain awareness as Earth orbits grow crowded and contested. Current analysis pipelines are built for detection: they are good at picking up that something happened, less good at reasoning about what it means. AstroMind is a physics-grounded benchmark designed to close that gap. It draws on high-fidelity astrodynamics simulations and real observational constraints, converting them into verifiable reasoning problems across three task types: intent inference, maneuver parameter estimation, and threat assessment. Each scenario includes realistic sensing noise and multi-source textual intelligence at varying reliability levels. Evaluation metrics capture both semantic correctness and quantitative consistency under physical constraints. Benchmarking a suite of open-weight models shows no single model dominates every axis: Qwen3 (32B) leads on intent inference accuracy; QwQ (32B) leads on threat assessment and achieves the lowest median relative error on parsed items; GPT-OSS (20B) produces the strongest judged reasoning quality and extracts the most scalar values for parameter estimation (136 of 241 parsed items). Training data composition and reasoning style matter as much as model size. Structured reasoning prompts help consistently across tested 8B models, with larger gains for those that can already track physical constraints. AstroMind gives the field a shared test for a problem where getting the physics right and reading the tactical situation correctly are both required -- neither is sufficient on its own.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper introduces AstroMind, a benchmark for LLM-based reasoning on spacecraft behavior in space domain awareness. It converts high-fidelity astrodynamics simulations and real observational constraints into three task types—intent inference, maneuver parameter estimation, and threat assessment—incorporating realistic sensing noise and multi-source textual intelligence of varying reliability. Evaluation uses metrics for semantic correctness and quantitative physical consistency. Benchmarking open-weight models (e.g., Qwen3-32B, QwQ-32B, GPT-OSS-20B) shows no single model dominates; performance depends on training data composition and reasoning style, with structured prompts helping smaller models track constraints.

Significance. If the benchmark construction and metrics are rigorously validated, AstroMind could fill a gap by providing a shared, physics-grounded test for LLMs that must combine semantic understanding with quantitative consistency under physical constraints. This is relevant for applications in contested orbital environments where both tactical interpretation and physical feasibility matter.

major comments (1)
  1. [Abstract / Methods] The abstract and high-level description provide no details on simulation generation, data splits, metric computation formulas, or error analysis (e.g., how relative error on parsed items or physical consistency is quantified). This makes it impossible to verify the central claim that the tasks meaningfully test physics-constrained reasoning; a methods section with explicit generation pipeline and validation against real constraints is required.
minor comments (2)
  1. [Evaluation] Clarify the exact number of scenarios per task type and how textual intelligence reliability levels are parameterized.
  2. [Experiments] Specify the open-weight model versions and prompting templates used for the structured reasoning experiments.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for highlighting the need for greater methodological transparency. We agree that the abstract is high-level by design and that explicit details on benchmark construction are essential for verifying the physics-constrained reasoning claims. We will revise the manuscript with an expanded Methods section.

read point-by-point responses

  1. Referee: [Abstract / Methods] The abstract and high-level description provide no details on simulation generation, data splits, metric computation formulas, or error analysis (e.g., how relative error on parsed items or physical consistency is quantified). This makes it impossible to verify the central claim that the tasks meaningfully test physics-constrained reasoning; a methods section with explicit generation pipeline and validation against real constraints is required.

    Authors: We acknowledge the concern. The current manuscript includes a Methods section (Section 3) outlining the overall pipeline, but it does not provide the level of explicit formulas, split statistics, or validation steps requested. In revision we will add: (1) the precise simulation generation process (astrodynamics integrator, orbital element sampling ranges, and noise models drawn from real sensor specifications); (2) data split details (scenario counts and stratification criteria); (3) exact metric definitions, including the relative error formula for parsed parameters and the physical consistency scoring procedure; and (4) validation steps comparing generated scenarios against public orbital catalogs. These additions will directly support the central claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper introduces AstroMind as a benchmark constructed from astrodynamics simulations, observational constraints, sensing noise, and textual intelligence. No equations, derivations, fitted parameters, or predictions appear in the abstract or described structure. The work reports evaluation results on existing models rather than deriving new quantities from prior fitted values or self-citations. No load-bearing steps reduce to inputs by construction, making the central claim self-contained as a benchmark definition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No details available from the abstract to identify specific free parameters, axioms, or invented entities; review is abstract-only.

pith-pipeline@v0.9.1-grok · 5800 in / 1032 out tokens · 26338 ms · 2026-06-30T13:08:09.850203+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 20 canonical work pages · 4 internal anchors

  1. [1]

    Esa space environment report 2025,

    European Space Agency, “Esa space environment report 2025,” ESA Space Debris Office, Darmstadt, Germany, Tech. Rep. GEN-DB-LOG- 00288-OPS-SD, 2025. [Online]. Available: https://www.sdo.esoc.esa. int/environment report/Space Environment Report latest.pdf

    2025

  2. [2]

    The low earth orbit satellite population and impacts of the spacex starlink constellation,

    J. C. McDowell, “The low earth orbit satellite population and impacts of the spacex starlink constellation,”The Astrophysical Journal Letters, vol. 892, no. 2, p. L36, 2020. [Online]. Available: https://doi.org/10.3847/2041-8213/ab8016

    work page doi:10.3847/2041-8213/ab8016 2020

  3. [3]

    Collision frequency of artificial satellites: The creation of a debris belt,

    D. J. Kessler and B. G. Cour-Palais, “Collision frequency of artificial satellites: The creation of a debris belt,”Journal of Geophysical Research: Space Physics, vol. 83, no. A6, pp. 2637–2646, 1978. [Online]. Available: https://doi.org/10.1029/JA083iA06p02637

    work page doi:10.1029/ja083ia06p02637 1978

  4. [4]

    Global counterspace capabilities: An open source assessment,

    Secure World Foundation, “Global counterspace capabilities: An open source assessment,” Secure World Foundation (SWF), Tech. Rep., April 2025. [Online]. Available: https://www.swfound.org/ publications-and-reports/2025-global-counterspace-capabilities-report

    2025

  5. [5]

    Space threat assessment 2025,

    C. Swope, K. A. Bingen, M. Young, and K. LaFave, “Space threat assessment 2025,” Center for Strategic and International Studies (CSIS), Tech. Rep., April 2025. [Online]. Available: https: //www.csis.org/analysis/space-threat-assessment-2025

    2025

  6. [6]

    Artificial intelligence and machine learning for space domain awareness: Characterizing the impact on mission effectiveness,

    L. A. Zhang, K. Langeland, J. Tran, J. Logue, P. Puri, G. Nacouzi, A. Jacques, and G. J. Briggs, “Artificial intelligence and machine learning for space domain awareness: Characterizing the impact on mission effectiveness,” RAND Corporation, Tech. Rep., 2024. [Online]. Available: https://www.rand.org/pubs/research reports/RRA2318-1.html 12

    2024

  7. [7]

    Inferring spacecraft maneuver intention via inverse optimal control,

    M. R. Goulet, T. Goulet, K. A. LeGrand, and S. Mou, “Inferring spacecraft maneuver intention via inverse optimal control,” inAIAA SCITECH 2025 Forum. American Institute of Aeronautics and Astronautics, January 2025. [Online]. Available: https://doi.org/10. 2514/6.2025-0982

    2025

  8. [8]

    Survey mode: A review of machine learning in resident space object detection and characterization,

    K. Tsaprailis, G. Choumos, V . Lappas, and C. Kontoes, “Survey mode: A review of machine learning in resident space object detection and characterization,” inAIAA SCITECH 2024 Forum, January 2024. [Online]. Available: https://doi.org/10.2514/6.2024-2065

    work page doi:10.2514/6.2024-2065 2024

  9. [9]

    Time-llm: Time series forecasting by reprogramming large language models,

    M. Jin, S. Wang, L. Ma, Z. Chu, J. Y . Zhang, X. Shi, P.-Y . Chen, Y . Liang, Y .-F. Li, S. Pan, and Q. Wen, “Time-llm: Time series forecasting by reprogramming large language models,” inInternational Conference on Learning Representations (ICLR), 2024. [Online]. Available: https://openreview.net/forum?id=Unb5CVPtae

    2024

  10. [10]

    Measuring massive multitask language understanding,

    D. Hendrycks, C. Burns, S. Basart, A. Zou, M. Mazeika, D. Song, and J. Steinhardt, “Measuring massive multitask language understanding,” inProceedings of the International Conference on Learning Representations (ICLR), 2021. [Online]. Available: https://openreview.net/forum?id=d7KBjmI3GmQ

    2021

  11. [11]

    What disease does this patient have? a large-scale open domain question answering dataset from medical exams,

    D. Jin, E. Pan, N. Oufattole, W.-H. Weng, H. Fang, and P. Szolovits, “What disease does this patient have? a large-scale open domain question answering dataset from medical exams,”Applied Sciences, vol. 11, no. 14, p. 6421, 2021. [Online]. Available: https://doi.org/10. 3390/app11146421

    2021

  12. [12]

    Legalbench: A collaboratively built benchmark for measuring legal reasoning in large language models,

    N. Guha, J. Nyarko, D. E. Ho, C. R ´e, A. Chilton, A. Narayana, A. Chohlas-Wood, A. Peters, B. Waldonet al., “Legalbench: A collaboratively built benchmark for measuring legal reasoning in large language models,” inAdvances in Neural Information Processing Systems 36 (NeurIPS 2023), Datasets and Benchmarks Track, 2023. [Online]. Available: https://openrev...

    2023

  13. [13]

    Orbit estimation of a continuously thrusting spacecraft using variable dimension filters,

    G. M. Goff, J. T. Black, and J. A. Beck, “Orbit estimation of a continuously thrusting spacecraft using variable dimension filters,” Journal of Guidance, Control, and Dynamics, vol. 38, no. 12, pp. 2390– 2404, 2015. [Online]. Available: https://doi.org/10.2514/1.G001091

    work page doi:10.2514/1.g001091 2015

  14. [14]

    B. D. Tapley, B. E. Schutz, and G. H. Born,Statistical Orbit Determi- nation. Elsevier, 2004, the standard textbook for orbit determination and Kalman filters. [Online]. Available: https://shop.elsevier.com/books/ statistical-orbit-determination/tapley/978-0-12-683630-1

    2004

  15. [15]

    Satellite maneuver detection and estimation with optical survey observations,

    A. Pastor, G. Escribano Bl ´azquez, M. Sanjurjo-Rivo, and D. Escobar, “Satellite maneuver detection and estimation with optical survey observations,”The Journal of the Astronautical Sciences, vol. 69, pp. 879–917, 2022. [Online]. Available: https://doi.org/10.1007/s40295-022-00311-5

    work page doi:10.1007/s40295-022-00311-5 2022

  16. [16]

    Why do many resource-rich countries have negative genuine saving? Anticipation of better times or rapacious rent seeking.Resource and Energy Economics, 32(1):28–44, 2010

    L. Porcelli, A. Pastor, A. Cano, G. Escribano Bl ´azquez, M. Sanjurjo- Rivo, D. Escobar, and P. Di Lizia, “Satellite maneuver detection and estimation with radar survey observations,”Acta Astronautica, vol. 201, pp. 274–287, 2022. [Online]. Available: https://doi.org/10.1016/j. actaastro.2022.08.021

    work page doi:10.1016/j 2022

  17. [17]

    Advances in Space Research72(4), 907–921 (2023) https://doi.org/10.1016/j.asr

    G. Escribano Bl ´azquez, M. Sanjurjo-Rivo, J. A. Siminski, A. Pastor, and D. Escobar, “Automatic maneuver detection and tracking of space objects in optical survey scenarios based on stochastic hybrid systems formulation,”Advances in Space Research, vol. 69, no. 9, pp. 3460–3477, 2022. [Online]. Available: https://doi.org/10.1016/j.asr. 2022.02.034

    work page doi:10.1016/j.asr 2022

  18. [18]

    A review of anomaly detection in spacecraft telemetry data,

    A. Fejjari, A. Delavault, R. Camilleri, and G. Valentino, “A review of anomaly detection in spacecraft telemetry data,”Applied Sciences, vol. 15, no. 10, p. 5653, 2025. [Online]. Available: https://doi.org/10.3390/app15105653

    work page doi:10.3390/app15105653 2025

  19. [19]

    Spaceseg: A high-precision intelligent perception segmentation method for multi- spacecraft on-orbit targets,

    H. Liu, P. Guo, S. Yang, Z. Jiang, Q. Hu, and D. Li, “Spaceseg: A high-precision intelligent perception segmentation method for multi- spacecraft on-orbit targets,”arXiv preprint arXiv:2503.11133, 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2503.11133

    work page doi:10.48550/arxiv.2503.11133 2025

  20. [20]

    Orbitzoo: Real orbital systems challenges for reinforcement learning,

    A. Oliveira, K. Dyreby, F. M. Caldas, and C. Soares, “Orbitzoo: Real orbital systems challenges for reinforcement learning,” inAdvances in Neural Information Processing Systems (NeurIPS), 2025. [Online]. Available: https://openreview.net/forum?id=oElWLpkOux

    2025

  21. [21]

    Superglue: A stickier benchmark for general-purpose language understanding systems,

    A. Wang, Y . Pruksachatkun, N. Nangia, A. Singh, J. Michael, F. Hill, O. Levy, and S. R. Bowman, “Superglue: A stickier benchmark for general-purpose language understanding systems,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 32, 2019. [Online]. Available: https://papers.nips.cc/paper/8589-superglue

    2019

  22. [22]

    Nature , author =

    K. Singhal, S. Azizi, T. Tu, S. S. Mahdavi, J. Wei, H. W. Chung, N. Scales, A. Tanwani, H. Cole-Lewis, S. Pfohlet al., “Large language models encode clinical knowledge,”Nature, vol. 620, no. 7972, pp. 172–180, 2023. [Online]. Available: https: //doi.org/10.1038/s41586-023-06291-2

    work page doi:10.1038/s41586-023-06291-2 2023

  23. [23]

    Evaluating Large Language Models Trained on Code

    M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. d. O. Pinto, J. Kaplan, H. Edwards, Y . Burda, N. Joseph, G. Brockmanet al., “Evaluating large language models trained on code,”arXiv preprint arXiv:2107.03374,

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    Evaluating Large Language Models Trained on Code

    [Online]. Available: https://doi.org/10.48550/arXiv.2107.03374

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2107.03374

  25. [25]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. Chi, Q. V . Le, and D. Zhou, “Chain-of-thought prompting elicits reasoning in large language models,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 35, 2022, pp. 24 824–24 837. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2022/hash/ 9d5609613524e...

    2022

  26. [26]

    Tree of thoughts: Deliberate problem solving with large language models,

    S. Yao, D. Yu, J. Zhao, I. Shafran, T. Griffiths, Y . Cao, and K. Narasimhan, “Tree of thoughts: Deliberate problem solving with large language models,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2023/hash/ 271db9922b8d1f4dd7aaef84ed5ac703-Abstract-Con...

    2023

  27. [27]

    Solving quantitative reasoning problems with language models,

    A. Lewkowycz, A. Andreassen, D. Dohan, E. Dyer, H. Michalewski, V . Ramasesh, A. Slone, C. Anil, S. Imanol, T. Gutman- Soloet al., “Solving quantitative reasoning problems with language models,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 35, 2022, pp. 3843–3857. [Online]. Available: https://proceedings.neurips.cc/paper files/paper...

    2022

  28. [28]

    Nature Reviews Physics , author =

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, “Physics-informed machine learning,”Nature Reviews Physics, vol. 3, no. 6, pp. 422–440, 2021. [Online]. Available: https://doi.org/10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  29. [29]

    ”general intelligence,

    C. Spearman, “”general intelligence,” objectively determined and measured,”The American Journal of Psychology, vol. 15, no. 2, pp. 201–292, 1904. [Online]. Available: https://doi.org/10.2307/1412107

    work page doi:10.2307/1412107 1904

  30. [30]

    Gardner,Frames of mind: The theory of multiple intelligences

    H. Gardner,Frames of mind: The theory of multiple intelligences. Basic Books, 1983. [Online]. Available: https://books.google.com/ books/about/Frames of Mind.html?id=ObgOAAAAQAAJ

    1983

  31. [31]

    Joint Publication 3-14: Space Operations,

    Joint Chiefs of Staff, “Joint Publication 3-14: Space Operations,” 2020, original publication dated April 10, 2018; incorporating Change 1 dated October 26, 2020

    2020

  32. [32]

    Orbit Data Messages,

    Consultative Committee for Space Data Systems, “Orbit Data Messages,” CCSDS, Blue Book CCSDS 502.0-B-3, 2023, recommended Standard. [Online]. Available: https://public.ccsds.org/Pubs/502x0b3e1. pdf

    2023

  33. [33]

    Recommended Design and Opera- tional Practices,

    Consortium for Execution of Rendezvous and Ser- vicing Operations, “Recommended Design and Opera- tional Practices,” CONFERS, Tech. Rep., 2022. [Online]. Available: https://cdn.ymaws.com/satelliteconfers.org/resource/resmgr/ confers publications/confers operating practices .pdf

    2022

  34. [34]

    Bhattarai, I

    K. Bhattarai, I. Y . Oh, J. M. Sierra, J. Tang, P. R. O. Payne, Z. Abrams, and A. M. Lai, “Leveraging gpt-4 for identifying cancer phenotypes in electronic health records: a performance comparison between gpt-4, gpt-3.5-turbo, flan-t5, llama-3-8b, and spacy’s rule-based and machine learning-based methods,”JAMIA Open, vol. 7, no. 3, p. ooae060, 2024. [Onli...

    work page doi:10.1093/jamiaopen/ooae060 2024

  35. [35]

    A Survey on LLM-as-a-Judge

    J. Gu, X. Jiang, Z. Shi, H. Tan, X. Zhai, C. Xu, W. Li, Y . Shen, S. Ma, H. Liu, S. Wang, K. Zhang, Y . Wang, W. Gao, L. Ni, and J. Guo, “A survey on llm-as-a-judge,”The Innovation, p. 101253, 2026, originally released as arXiv:2411.15594. [Online]. Available: https://doi.org/10.1016/j.xinn.2025.101253

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.xinn.2025.101253 2026

  36. [36]

    P ´olya,How to Solve It: A New Aspect of Mathematical Method, ser

    G. P ´olya,How to Solve It: A New Aspect of Mathematical Method, ser. Princeton Science Library. Princeton University Press, 2014

    2014

  37. [37]

    Review on hardware devices and software techniques enabling neural network inference onboard satellites,

    L. Diana and P. Dini, “Review on hardware devices and software techniques enabling neural network inference onboard satellites,” Remote Sensing, vol. 16, no. 21, p. 3957, 2024. [Online]. Available: https://doi.org/10.3390/rs16213957

    work page doi:10.3390/rs16213957 2024

  38. [38]

    A comprehensive survey of orbital edge computing: Systems, applications, and algorithms,

    Z. Yin, C. Wu, C. Guo, Y . Li, M. Xu, W. Gao, and C. Chi, “A comprehensive survey of orbital edge computing: Systems, applications, and algorithms,”Chinese Journal of Aeronautics, vol. 38, no. 7, 2025. [Online]. Available: https://doi.org/10.1016/j.cja.2024.11.026

    work page doi:10.1016/j.cja.2024.11.026 2025

  39. [39]

    Fractionated spacecraft: The new sprout in distributed space systems,

    J. Guo, D. C. Maessen, and E. Gill, “Fractionated spacecraft: The new sprout in distributed space systems,” inProceedings of the 60th International Astronautical Congress, 2009, iAC Paper IAC-09-D1.1.4

    2009

  40. [40]

    Satellite edge artificial intelligence with large models: Architectures and technologies,

    Y . Shi, J. Zhu, C. Jiang, L. Kuang, and K. B. Letaief, “Satellite edge artificial intelligence with large models: Architectures and technologies,”Science China Information Sciences, vol. 68, no. 7,

  41. [41]

    Available: https://doi.org/10.1007/s11432-024-4425-y

    [Online]. Available: https://doi.org/10.1007/s11432-024-4425-y

    work page doi:10.1007/s11432-024-4425-y

  42. [42]

    gpt-oss-120b & gpt-oss-20b Model Card

    OpenAI, “gpt-oss-120b & gpt-oss-20b model card,” 2025. [Online]. Available: https://arxiv.org/abs/2508.10925

    work page internal anchor Pith review Pith/arXiv arXiv 2025