Agentic Artificial Intelligence for Power Systems: Strategies to Identify and Close Capability Gaps

Eve Tsybina; Samim Konjicija; Slaven Peles

REVIEW 3 major objections 5 minor 21 references

Agentic Artificial Intelligence for Power Systems: Strategies to Identify and Close Capability Gaps

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Current agentic AI solves only the two simplest grid-planning prompts

desk verdict A candid capability-gap stress test with a useful failure-mode autopsy, but the 'solvable' label overstates what the agent actually got right — the abstract needs a ground-truth comparator or a rewording to 'workflow completion.' read the letter →

arxiv 2607.28822 v1 pith:2EB5KCCJ submitted 2026-07-30 eess.SY cs.SY

Eve Tsybina , Samim Konjicija , Slaven Peles This is my paper · ORCID

classification eess.SYcs.SY

keywords agenticartificialintelligencepowersystemplanninggridlargelanguagemodelscapabilitybenchmarkingnodalcontingencyanalysisscalability

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 sets out to test whether agentic AI—AI agents that can chain tool use, run simulations, and iterate—can do realistic power-grid connection studies. The authors replicate the common design of published grid-planning agents and subject their prototype to a structured ladder of six task-complexity levels and four grid sizes. They find that only the two lowest complexity levels are solvable, and only on 100- and 1,000-bus grids; levels 3–6 and grids of 10,000 or more buses produce no results. This matters because grid operators face a wave of new data-center and generation interconnections and had hoped agents could automate the repetitive screening work. The paper concludes that current agentic AI is not yet ready for realistic connection studies and identifies specific backend and retrieval improvements needed.

What carries the argument

The key machinery is a structured test suite built around a 'hop-count' difficulty ladder: each prompt's level is the number of discrete computational operations (constructing candidate sets, solving cases, evaluating feasibility, aggregating, selecting) required to go from the base case to the answer. This gives six levels, from single aggregation (Level 1) to N-1 contingency testing (Level 3) and N-2 contingency relief (Level 6). Grids are built by tiling a 100-bus block derived from a reduced IEEE 118-bus system and modified to introduce congestion and reactive-power constraints, then tested at 100, 1,000, 10,000, and 100,000 buses. The difficulty ladder and the tiled, congestible grids t

What would settle it

A reader could run the same six-level suite on the code released with the paper (or on the current in-house version mentioned in Section IV) and inspect whether Level 3–6 prompts ever produce solver output. The paper states that the OpenAI agent never sent contingencies into the solver and DeepSeek listed but did not evaluate them; if a run of the same agent shows a contingency set being solved and an answer returned, the central claim that these levels are unsolvable would be falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, its central discovery is that the current state of the art in agentic AI for power systems can handle at most two of six complexity levels on grids up to 1,000 buses. The authors built a generic agent incorporating storage, retrieval-augmented generation, a solver interface, an executor, a context module, and a validator—the elements common to all published grid-planning agents—and tested it on a suite of nodal planning prompts. Levels 1 and 2, which ask for bus hosting-capacity screening and reactive-adequacy reporting, were solved reliably by the OpenAI configuration (10/10 attempts) and partially by the DeepSeek configuration (5/10 on Level 1, 10/10 on Level 2).

Load-bearing premise

The conclusion that only the two simplest levels are solvable industry-wide rests on the assumption that the authors' self-built generic agent faithfully represents the actual state of the art in agentic AI for power systems; if the real agents have unreleased fixes or better backend engineering, they might perform higher on the ladder.

Editorial extensions

If this is right

Existing agentic AI cannot yet automate realistic connection studies; human engineers must still interpret solver outputs, since even the model's successful sweeps were not always correctly interpreted by the AI.
Grid size is a primary barrier: systems beyond 1,000 buses overflow context windows and face O(N²) sweep costs, so performance does not scale from small test grids to utility-scale systems.
Complexity beyond basic screening requires extensive, detailed retrieval (helper functions and dictionaries), not just a more powerful model or more tokens.
A reproducible test suite of this kind—six prompt levels, four grid scales—can serve as a benchmark for measuring genuine progress in agentic AI for power systems.
The specific failure modes identified (e.g., sweep-cache keys omitting the reactive-load parameter) imply that subtle implementation bugs can invalidate results even for easy prompts.

Reading between the lines

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

If the authors' generic agent is a fair proxy for the field, then published claims of agentic AI handling N-1 contingency management or grid control likely overstate maturity; but because the real agents' code and data are not public, this inference depends on that proxy's fidelity.
The hop-count difficulty ladder is a transferable profiling tool: the same methodology could be applied to other engineering AI applications (e.g., water networks, transportation) to locate the boundary between what agents can and cannot do.
The finding that native LLM reasoning with minimal retrieval fails on Level 3+ suggests that agent performance may be gated more by the quality of the tool/retrieval layer than by model intelligence, implying that investments in structured knowledge bases and sparse-data backends could yield larger gains than larger models.
If the cache-key bug is representative, testing protocols should randomize parameter permutations to expose silent result-reuse errors; the paper's proposed stricter protocols would be a good basis for such adversarial testing.

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.

Desk Editor's Note

read the letter

Colleague — worth a look. The real contribution is the stress-test itself: six hop-based prompt levels, four tiled grid sizes up to 100k buses, and a concrete failure-mode forensic (cache key omitting reactive load, O(N²) deep-copy sweeps, context-window overflow). That's not present in the cited prior work, and it's the right kind of challenge for the community. The authors are also refreshingly honest about their own prototype's limits.

The soft spots are real but addressable. First, the abstract claims to 'replicate the current state of the art,' but they built a single generic agent from AgentiGrid/ExaGO and infer the other agents' behavior from descriptions — then concede in Section IV that they can't predict how PowerAgent, GridAgent, or GridMind would do. So the headline finding applies to one prototype, not the literature. Second, and more seriously, the 'solvable' metric is under-defined. Table II counts attempts as solved, yet the same section says the AI didn't pick the correct numbers from solver outputs and its analytical recommendations didn't match solver results. That means 'solved' is at best workflow completion, not engineering-correct answers. Without a ground-truth ACOPF comparator, the central claim — that only Levels 1 and 2 are solvable — is ambiguous. It might be true, but the paper as written doesn't demonstrate it. The authors may have intended a distinction between completing steps and making extra recommendations; if so, they need to draw that line explicitly.

What holds up: the tiled grids and adjusted constraints are a reasonable way to test congestion and reactive issues, the hop ladder is a sensible complexity proxy, and the failure modes they document are instructive and specific. The citation pattern looks fine — they cite the five major agent papers and note the code-availability problem correctly.

Bottom line: the framework is reusable and the failure analysis is valuable, but the paper needs a ground-truth check or a rewording before the main claim stands. I'd send it to peer review, expecting a conditional accept after revision. Tempering the abstract and adding an explicit definition of 'solved' would fix most of my reservations.

Referee Report

3 major / 5 minor

Summary. The manuscript addresses whether current agentic AI for power systems can support nodal connection-planning studies at scale. The authors survey five grid-specific agentic frameworks, extract a common architecture, and build their own 'generic best practice' agent (AgentiGrid integrated with ExaGO). They evaluate it on four grid sizes (100, 1,000, 10,000, and 100,000 buses) and six prompt levels defined by the number of discrete 'hops' required. The reported results show that Level 1 and Level 2 prompts are completed on 100- and 1,000-bus grids (OpenAI 10/10 on both levels; DeepSeek 5/10 on Level 1 and 10/10 on Level 2), while Levels 3–6 and the 10,000- and 100,000-bus cases produced no results. The authors conclude that only the two lowest complexity levels are solvable on some grid sizes and identify backend and RAG bottlenecks as the main capability gaps.

Significance. If the reported results are verified, the paper provides a valuable cautionary empirical benchmark: it suggests that current agentic AI for power systems is not yet ready for realistic connection studies, and it offers a structured difficulty ladder and a reproducible test methodology. The hop-based prompt hierarchy and the constrained tiled grids are useful contributions, and the availability of code/data for the prototype is a strength. However, the headline claim depends critically on what is meant by 'solvable,' and the manuscript's own text indicates that workflow completion and correct engineering answers were conflated. The significance of the negative result therefore cannot be fully assessed until a correctness check against reference solutions is performed.

major comments (3)

[Section IV, Table II] The metric labeled 'solved' does not verify correctness. Immediately after Table II the text states: 'Even when the solver sweeps were implemented correctly, the AI did not pick the correct numbers from the solver outputs or did not reason correctly about the solutions,' and that analytical recommendations were 'not consistent with the numbers produced by the solver, for either vendor or any system.' Thus the 10/10 and 5/10 entries indicate that the agent completed the intended workflow, not that its outputs matched a reference solution. Without a ground-truth ACOPF comparator, the abstract's claim that 'only the two lowest complexity levels are solvable' is unsupported; if completed runs returned incorrect numbers, then no level was actually solved. Please add a per-attempt correctness check against a reference solution, or redefine 'solvable' as 'workflow completed' and report correctn
[Abstract; Sections III.A, V] The paper claims to 'replicate the current state of the art in agentic AI for power systems planning,' but the experiments run only the authors' own AgentiGrid prototype. Section III.A says the prototype 'closely reflects the state of the art described in the literature,' and Section V concedes that 'it is difficult to predict exactly how those agents would perform under the proposed framework.' Because PowerAgent, GridAgent, GridMind, LLM4OPF, and GAIA were not run, the headline conclusion is demonstrated only for this one prototype, not for the state of the art generally. Please narrow the abstract and conclusions to 'our prototype' or, where third-party code is public, run and compare the cited agents.
[Section III.B; Section IV; Table II] The 10,000- and 100,000-bus results and the Level 3–6 failures are described only qualitatively and are omitted from Table II. The failures are attributed to specific implementation limitations of one code snapshot (O(N^2) deep-copy sweeps and context-window overflow), which the authors state have already been addressed in a newer AgentiGrid version. As written, the central result is tied to a particular software version and cannot be independently reconstructed from the paper. A supplementary table listing all 480 runs (attempts, tokens, time, and failure mode) would let readers verify the reported coverage and would make the benchmark genuinely reproducible.

minor comments (5)

[Section III.B] The text introduces 'difficulty level from 0 to 6,' but Table II and the prompt list use only Levels 1–6. Clarify whether Level 0 exists and what it would be.
[Section III.A] The phrase 'The source code is available at [19]' may be ambiguous because reference [19] is the ExaGO v2 software DOI. Please specify whether the AgentiGrid agent code and the grid-generation scripts are included in that DOI or in the supplemental material.
[Section III.B] Using Claude Opus 4.8 for prompt validation and result aggregation is reasonable, but the statement that it 'cannot bias the measured performance of either model' is too strong. A third-party LLM could still misread or misreport solver outputs, which is a reporting risk rather than a model-performance bias.
[Table II caption] The caption says 'state of the art grids and prompts,' but the grids are newly constructed by tiling a 100-bus block and the prompts are newly defined. Rephrase to avoid implying these are the exact grids or prompts used by the cited agents.
[References] Reference [16] appears to have an unusual volume/issue/page format; verify it against the publisher's style and page range.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical benchmark with explicit scope limitations.

full rationale

The manuscript is an experimental evaluation rather than a derivation chain. Table II reports measured attempt/success counts for a specific agentic prototype, and the abstract's claim ('only the two lowest complexity levels are solvable on some of the test grid sizes') is a direct summary of those measurements, not a quantity derived from fitted parameters or from the definitions of the difficulty levels. The only self-citation is the ExaGO/AgentiGrid code release [19], which is code-reproduced and used as the test platform; per the reviewing rules, this is real evidence rather than a circularity. The paper explicitly limits generalization: 'Because code and data for existing agentic platforms are not publicly available, it is difficult to predict exactly how those agents would perform under the proposed framework' (Section V). That is a validity limitation, not a circular step. The results section also concedes that 'even when the solver sweeps were implemented correctly, the AI did not pick the correct numbers from the solver outputs or did not reason correctly about the solutions' (Section IV), which raises a measurement-validity concern about what 'solved' means, but it does not make the benchmark circular. No load-bearing step reduces to its own inputs by construction or by self-citation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The paper is an empirical benchmark, so there are no fitted physical constants or invented entities. The central claim rests on several domain assumptions about representativeness (the generic agent stands in for SOTA), task difficulty (hop-count ladder), grid realism (tiled blocks), and experimental protocol (10 attempts, prompt-preprocessing vendor). These choices are transparent but unvalidated against external ground truth.

free parameters (3)

Nearest-neighbor count used in Level 3-6 prompts = 3
Hand-chosen constant in the contingency-testing prompts. Fewer or more neighbors would change the difficulty of Levels 3-6 and could shift the solvability boundary.
Grid scale multipliers = 1, 10, 100, 1000 (x 100-bus block)
Selected grid sizes (100, 1,000, 10,000, 100,000 buses). The scaling conclusions ('larger grids fail') are defined relative to this hand-picked ladder.
Attempts per condition = 10
Number of repeated LLM runs per prompt-level combination. Small sample; success rates in Table II are counts out of 10 without confidence intervals.

assumptions (5)

domain assumption The self-built generic agent faithfully represents the current state of the art in agentic AI for power systems.
Section III.A: 'It incorporates all of the agentic AI elements discussed in section II and closely reflects the state of the art described in the literature.' This is the load-bearing bridge from the prototype's failures to conclusions about the whole field; no external SOTA agent was run.
domain assumption Hop-count levels (0-6) are a valid measure of task complexity for connection studies.
Section III.B defines difficulty as the number of discrete operations. The headline claim about 'two lowest complexity levels solvable' is entirely relative to this operationalization.
domain assumption Tiled copies of a 100-bus adjusted IEEE-118 block produce representative large congested grids.
Section III.B constructs 1k-100k bus grids by tiling identical blocks. Repeated identical structure may not capture real heterogeneity and could misrepresent scaling behavior.
domain assumption Ten attempts per condition adequately characterize stochastic LLM success rates.
Section III.A justifies 10 attempts by LLM nondeterminism, but no statistical analysis is provided; small cell sizes make the 5/10 vs 10/10 distinction potentially noisy.
domain assumption Claude Opus 4.8 used for prompt validation and result aggregation does not bias either vendor's measured performance.
Section III.B assumes cross-vendor independence removes bias. The validation/aggregation steps could still impose systematic preferences (e.g., prompt rewording favoring one model).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Agentic Artificial Intelligence for Power Systems: Strategies to Identify and Close Capability Gaps." pith.science (2026). https://pith.science/paper/2EB5KCCJ

@misc{pith2026260728822,
  author       = {Pith},
  title        = {Pith review of: Agentic Artificial Intelligence for Power Systems: Strategies to Identify and Close Capability Gaps},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2EB5KCCJ}},
  note         = {Machine review of arXiv:2607.28822}
}

read the original abstract

The rapid expansion of AI-driven information infrastructure, particularly data centers, is placing unprecedented pressure on power systems and accelerating the pace at which new assets must interconnect with the grid. As bulk transmission expansion rolls out slowly, new loads and generation are increasingly deployed within existing network constraints. Agentic AI is urgently needed to automate the numerous and repetitive connection processes, but its maturity has not been systematically validated on complex tasks and large-scale systems. We replicate the current state of the art in agentic AI for power systems planning and evaluate it against a structured suite of nodal planning problems spanning six levels of task complexity and four grid scales. We find that only the two lowest complexity levels are solvable on some of the test grid sizes, and identify the specific capability upgrades required to close this gap. Adopting stricter testing protocols and reproducible evaluation benchmarks is essential for assessing both genuine progress and the operational readiness of agentic AI.

Figures

Figures reproduced from arXiv: 2607.28822 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 2 canonical work pages

[1]

Datacenters and energy report,

S&P Global Market Intelligence, “Datacenters and energy report,” S&P Global Market Intelligence, Tech. Rep., 2025, subscription based

2025
[2]

[Online]

Entso-e|planning the future grid - TYNDP. [Online]. Available: https://tyndp.entsoe.eu/
[3]

The asian super grid in northeast asia and china’s belt and road initiative,

G. Christoffersen, “The asian super grid in northeast asia and china’s belt and road initiative,” German Institute for International and Security Affairs, Tech. Rep., 2018

2018
[4]

FERC technical conference regarding the challenge of resource ade- quacy in RTO and ISO regions,

“FERC technical conference regarding the challenge of resource ade- quacy in RTO and ISO regions,” Federal Energy Regulatory Commis- sion, Jun. 2025

2025
[5]

National Association of Manufacturers

(2025, Nov.) Tech firms are building their own power plants. National Association of Manufacturers. Accessed 11/14/2025. [Online]. Available: https://nam.org/ tech-firms-are-building-their-own-power-plants-34960/

2025
[6]

Poweragent: A road map toward agentic intel- ligence in power systems: Foundation model, model context protocol, and workflow,

Q. Zhang and L. Xie, “Poweragent: A road map toward agentic intel- ligence in power systems: Foundation model, model context protocol, and workflow,”IEEE Power and Energy Magazine, vol. 23, no. 5, pp. 93–101, 2025

2025
[7]

Grid- agent: An llm-powered multi-agent system for power grid control,

Y . Zhang, A. M. Saber, A. Youssef, and D. Kundur, “Grid- agent: An llm-powered multi-agent system for power grid control,” arXiv preprint arXiv:2508.05702v3, 2025. [Online]. Available: https: //arxiv.org/html/2508.05702v3

arXiv 2025
[8]

Gridmind: Llms-powered agents for power system analysis and operations,

H. Jin, K. Kim, and J. Kwon, “Gridmind: Llms-powered agents for power system analysis and operations,” ser. SC Workshops ’25. New York, NY , USA: Association for Computing Machinery, 2025, p. 560–568. [Online]. Available: https://doi.org/10.1145/3731599.3767409

arXiv 2025

Show all 21 references

[9]

Large foundation models for power systems,

C. Huang, S. Li, R. Liu, H. Wang, and Y . Chen, “Large foundation models for power systems,”arXiv, 2023. [Online]. Available: https://arxiv.org/abs/2312.07044

2023 arXiv
[10]

A large language model for advanced power dispatch,

Y . Cheng, H. Zhao, X. Zhou, J. Zhao, Y . Cao, C. Yang, and X. Cai, “A large language model for advanced power dispatch,”Scientific Reports, vol. 15, no. 1, p. 8925, 2025

2025
[11]

Grid copilot: A large language model (llm) based framework for transforming long-term planning analyses,

S. Chaturvedi, S. Jin, S. Abhyankar, T. Thurber, K. Oikonomou, and N. V oisin, “Grid copilot: A large language model (llm) based framework for transforming long-term planning analyses,” in2025 IEEE Power & Energy Society General Meeting (PESGM), 2025, pp. 1–5. [Online]. Availa...

2025
[12]

Generative ai for power grid operations,

S. L. Choi, R. Jain, C. Feng, P. Emami, H. Zhang, J. Hong, T. Kim, S. Park, F. Ding, M. Bagguet al., “Generative ai for power grid operations,” National Renewable Energy Laboratory (NREL), Golden, CO (United States), Tech. Rep. NREL/TP-5D00-91176, 2024. [Online]. Available: ht...

2024
[13]

Ai for the grid and ai on the grid: Implications, challenges and opportunities,

B. Cheng, A. Botterud, T. Levin, S. Nadarajah, D. Zhao, and J. Kwon, “Ai for the grid and ai on the grid: Implications, challenges and opportunities,” Argonne National Laboratory, Tech. Rep., February
[14]

Review of llms applications in electrical power and energy systems,

F. Amjad, T. Kor ˜otko, and A. Rosin, “Review of llms applications in electrical power and energy systems,”IEEE Access, vol. 13, pp. 150 951– 150 969, 2025

2025
[15]

A review of large language models for energy systems: Applications, challenges, and future prospects,

H. Mirshekali, M. Reza Shadi, F. Ghanadi Ladani, and H. Reza Shaker, “A review of large language models for energy systems: Applications, challenges, and future prospects,”IEEE Access, vol. 13, pp. 163 162– 163 188, 2025

2025
[16]

Agentic artificial intelligence for smart grids: A comprehensive review of autonomous, safe, and explainable control frameworks,

M. Kiasari and H. Aly, “Agentic artificial intelligence for smart grids: A comprehensive review of autonomous, safe, and explainable control frameworks,”Energies, vol. 19, no. 3, p. 617, 2026. [Online]. Available: https://www.mdpi.com/1996-1073/19/3/617

2026
[17]

IEEE 118-bus three-area test system,

“IEEE 118-bus three-area test system,” University of Washington. [Online]. Available: https://labs.ece.uw.edu/pstca/
[18]

Real-time optimal power flow with linguistic stipulations: Integrating gpt-agent and deep reinforcement learning,

Z. Yan and Y . Xu, “Real-time optimal power flow with linguistic stipulations: Integrating gpt-agent and deep reinforcement learning,” IEEE Transactions on Power Systems, vol. 39, no. 2, pp. 4747–4750, 2024

2024
[19]

Exago v2,

S. Peles, N. Koukpaizan, M. Alam, J. Hambrick, and E. Tsybina, “Exago v2,” [Computer Software] https://doi.org/10.11578/dc.20260223.2, 2026

2026 doi
[20]

A real-time operations manual for the ieee 118 bus transmission model,

A. Anderson, S. Kincic, B. Jefferson, B. Mcgary, C. Fallon, D. Ciesiel- ski, J. Wenskovitch, and Y . Chen, “A real-time operations manual for the ieee 118 bus transmission model,” Pacific Northwest National Laboratory, Richland, Washington 99354, Tech. Rep. PNNL-334996, 2022

2022
[2026]

Available: http://dx.doi.org/10.2139/ssrn.6315298

[Online]. Available: http://dx.doi.org/10.2139/ssrn.6315298

doi

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

Datacenters and energy report,

S&P Global Market Intelligence, “Datacenters and energy report,” S&P Global Market Intelligence, Tech. Rep., 2025, subscription based

2025

[2] [2]

[Online]

Entso-e|planning the future grid - TYNDP. [Online]. Available: https://tyndp.entsoe.eu/

[3] [3]

The asian super grid in northeast asia and china’s belt and road initiative,

G. Christoffersen, “The asian super grid in northeast asia and china’s belt and road initiative,” German Institute for International and Security Affairs, Tech. Rep., 2018

2018

[4] [4]

FERC technical conference regarding the challenge of resource ade- quacy in RTO and ISO regions,

“FERC technical conference regarding the challenge of resource ade- quacy in RTO and ISO regions,” Federal Energy Regulatory Commis- sion, Jun. 2025

2025

[5] [5]

National Association of Manufacturers

(2025, Nov.) Tech firms are building their own power plants. National Association of Manufacturers. Accessed 11/14/2025. [Online]. Available: https://nam.org/ tech-firms-are-building-their-own-power-plants-34960/

2025

[6] [6]

Poweragent: A road map toward agentic intel- ligence in power systems: Foundation model, model context protocol, and workflow,

Q. Zhang and L. Xie, “Poweragent: A road map toward agentic intel- ligence in power systems: Foundation model, model context protocol, and workflow,”IEEE Power and Energy Magazine, vol. 23, no. 5, pp. 93–101, 2025

2025

[7] [7]

Grid- agent: An llm-powered multi-agent system for power grid control,

Y . Zhang, A. M. Saber, A. Youssef, and D. Kundur, “Grid- agent: An llm-powered multi-agent system for power grid control,” arXiv preprint arXiv:2508.05702v3, 2025. [Online]. Available: https: //arxiv.org/html/2508.05702v3

arXiv 2025

[8] [8]

Gridmind: Llms-powered agents for power system analysis and operations,

H. Jin, K. Kim, and J. Kwon, “Gridmind: Llms-powered agents for power system analysis and operations,” ser. SC Workshops ’25. New York, NY , USA: Association for Computing Machinery, 2025, p. 560–568. [Online]. Available: https://doi.org/10.1145/3731599.3767409

arXiv 2025