Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

Develop AI Agents for System Engineering in Factorio

T0 review · 3 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read The paper argues that AI agents should train and be evaluated for system engineering inside Factorio, the factory-building sandbox game.

desk verdict A clearly argued position paper proposing Factorio as an agent-benchmark testbed; the 'ideal' claim outruns the evidence, but it deserves a serious referee. read the letter →

arxiv 2502.01492 v1 pith:OGBZHYHH submitted 2025-02-03 cs.AI

classification cs.AI
keywords AIagentssystemengineeringFactoriosandboxgamesagentevaluationlong-horizonplanningviablemodeldynamicenvironments
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

System engineering — designing, building, maintaining, and adapting large interconnected systems — is high-value and hard, and the paper argues that AI agents should be trained and measured for it in automation-oriented sandbox games. Current agent benchmarks, like coding-bug-fix tasks, are static: they never force an agent to keep a system viable as requirements, resource supplies, and failures change over time. The paper's concrete proposal is Factorio, whose core loop is turning raw materials into ever more automated production lines, because it naturally tests long-horizon planning, trade-offs between efficiency and flexibility, and adaptation to disruptions. If correct, this redirects agent evaluation from single-shot questions to open-ended simulations, and makes a game the proving ground for future AI infrastructure designers.

What carries the argument

The viable system model (VSM) — a cybernetic account in which viable systems have five nested levels (front-line operations, coordination, internal optimization, future planning, and ultimate policy) — carries the argument. The paper maps Factorio's mechanics onto those levels: assemblers and miners as System 1, belts and splitters as System 2, balancing and throughput management as System 3, tech-tree planning and expansion as System 4, and mission policy as System 5. The law of requisite variety, another cybernetic idea, supplies the evaluation requirement: a system can remain viable only if its internal range of responses at least matches the variety of disturbances it may face over time. Factorio's modding system and its science-per-minute metric provide the concrete way to generate that variety and to measure an agent's long-horizon performance.

What would settle it

Train an agent to sustain high science per minute in Factorio under modded disruptions, then run the same agent on a realistic logistics or power-grid simulator with partial observations and shifting demand. If its performance is not better than an agent trained on static question-answering or coding benchmarks, the claim that Factorio is the right training ground is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper's contribution is a reframing: evaluate AI agents not on static, closed benchmarks but in dynamic, moddable, open-ended system-building environments. It claims Factorio is the best available instance, because its mechanics require the agent to operate at all five levels of the viable system model — from individual machines that turn ore into plates up to the strategic decisions about what the factory should ultimately achieve — and because its mods, headless server, and 2D efficiency make it practical to instrument. The paper maps every major game system (belts, trains, construction robots, science packs, pollution-driven attacks) onto a systems-engineering concern, and proposes an evaluator-agent setup in which a controller injects shortages, failures, or new objectives while the agent rebuilds. The result it is trying to establish is that success in Factorio is a meaningful signal of system-engineering capability.

Load-bearing premise

The load-bearing premise is that engineering skills learned in a video game will transfer to real-world system-engineering work, such as running an energy grid or a supply chain; the paper asserts this transfer but does not measure it.

Editorial extensions

If this is right

  • Agent evaluation shifts from one-shot question-answering and code-fix benchmarks to long-horizon, dynamic simulations where success is measured in sustained throughput and resilience rather than a pass/fail test.
  • A control API for Factorio, on the model of existing Minecraft bot interfaces, would let frontier agents use native mouse-and-keyboard and GUI inputs, forcing advances in multimodal perception, long-context memory, and real-time action under partial information.
  • Modded variants can recreate energy-grid expansion, supply-chain shocks, and market negotiation, so a single sandbox can cover multiple system-engineering subdomains without building new simulators from scratch.
  • Multi-agent and human-agent coordination can be studied in the same environment, with evaluator agents injecting failures and shifting objectives to test whether the overall system keeps operating.

Reading between the lines

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

  • Beyond the paper, if Factorio becomes a standard testbed, the field gains a common, parameterizable hard task, but the paper's transfer assumption would still need its own benchmark before real-world claims can be made.
  • Beyond the paper, the evaluator-agent framework suggests an adversarial curriculum: an evaluator that learns to generate the most disruptive failures could create a self-hardening training loop, analogous to self-play.
  • Beyond the paper, a natural next experiment is to measure whether high science-per-minute in Factorio correlates with performance on established human system-design interview tasks or on small physical supply-chain simulators.
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

3 major / 4 minor

Summary. This position paper argues that system engineering is a uniquely high-leverage capability for AI agents, that current static benchmarks do not capture the dynamic trade-offs it requires, and that automation-oriented sandbox games—especially Factorio—should be adopted as training and evaluation environments. The argument draws on cybernetic frameworks (Ashby's law of requisite variety and Beer's viable system model), maps VSM levels to Factorio mechanics in Table 1, and outlines technical advantages such as 2D rendering efficiency, Lua modding, headless multiplayer support, and an Agent-Evaluator framework for injecting perturbations. An appendix provides a visual introduction to Factorio for unfamiliar readers. The paper contains no experiments or empirical evaluations.

Significance. The proposed direction is timely and the paper is clearly organized. Its concrete assets include the SPM metric as a community-standard summary score, the explicit VSM-to-Factorio mapping, and the realistic technical advantages of Factorio for AI experimentation (headless server, modding API, cross-platform support). The paper is also appropriately cautious in Section 5 by acknowledging the realism objection. However, the central claim that Factorio is the ideal testbed for developing real-world system engineering capability rests on unvalidated assumptions about transfer, and the 'ideal' wording is not supported by systematic comparison with alternative games. As a position paper, this is a useful hypothesis-generating proposal, but it does not yet establish the strength of its main claim.

major comments (3)
  1. [Section 5] The third paragraph of Section 5 acknowledges the realism objection but replies that 'unrealistic' environments 'highlight the essence of system engineering... without the noise associated with realistic physics simulations.' This is an empirical claim about which features are essential, and it is load-bearing for the entire proposal. No evidence or protocol is given for how skills learned in a deterministic, fully observable, single-player factory simulator would transfer to systems with stochastic component failures, partial observability, safety and regulatory constraints, or human and organizational factors. The manuscript should either add a falsifiable transfer-evaluation design (for example, train an agent in Factorio and measure its performance on a high-fidelity simulator or real-world task against baselines) or explicitly reframe the contribution as a hypothesis about transfer rather than a demonstrated capability.
  2. [Section 3.4 and Section 4] The claim that Factorio 'stands out as the ideal sandbox game' is asserted rather than established. Section 3.4 lists several alternative games (Cities: Skylines, Satisfactory, etc.) but does not evaluate them against the five core properties it proposes, and Section 4 immediately moves to Factorio's advantages. A systematic comparison with at least one or two direct alternatives, such as Satisfactory, Dyson Sphere Program, or Shapez 2, using the paper's own criteria would be needed to justify the word 'ideal.' Adding a criteria-by-game comparison table would directly address the paper's strongest claim.
  3. [Section 3.2 and Table 1] The VSM/LRV framework is presented as if it operationalizes system-engineering capability, but no measurable definition of 'variety' is given for Factorio, and the mapping from VSM levels to game mechanics is an analogy. Since the paper builds its evaluation argument on this framework, it should define proxies for the framework's constructs (for example, how to measure environmental variety VE and response variety VR in a Factorio episode) or state explicitly that the framework is used only as a heuristic. As written, the theoretical support is suggestive rather than load-bearing.
minor comments (4)
  1. [Section 3.4] The paragraph beginning 'We deduce from the ar, sandbox games...' contains an incomplete phrase and should be corrected to 'We deduce from the above that sandbox games...'.
  2. [Title and running header] The title and running header display 'F actorio' with a spurious space; this formatting artifact should be fixed in the camera-ready version.
  3. [Section 4.2] The claim that 'realistically the game only needs to be played at around 5 FPS' is unsupported; either remove the claim or provide a reference or measurement justifying the frame-rate requirement.
  4. [References] Several gameplay figures are cited from Reddit, Imgur, and Steam community posts; these are acceptable for screenshots but should be explicitly labeled as community sources rather than standard research citations.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the Factorio-as-testbed claim is an argued position rather than a derivation from its own inputs; the sole self-citation is illustrative and not load-bearing.

full rationale

This is a position paper with no equations, fitted parameters, or derived quantitative predictions, so the main circularity patterns (self-definitional fits, fitted inputs renamed as predictions, results forced by construction) do not apply. The central claim, that Factorio is an ideal environment for developing AI system-engineering capability, is argued qualitatively in Sections 3 and 4 from feature-based considerations such as automation mechanics, modding support, scalability, and the mapping of game elements to the Viable System Model. These arguments are not presented as a formal derivation from an input assumption that already contains the conclusion. The only self-citation is Roy et al. (2021), PrefixRL, which includes the author Neel Kant; it appears in Section 2.3 merely as one example of AI outperforming human engineers in a design task. That citation is not load-bearing for the paper's central recommendation, so it does not constitute circularity. Section 5's response to the realism objection, dismissing detailed physics and regulation as 'noise', is an asserted empirical claim rather than a demonstrated result, and the paper itself acknowledges that skills may need pairing with domain-specific testing. These are evidentiary or correctness weaknesses, not circularity, because the conclusion does not reduce to the paper's own inputs. Accordingly, no specific circular step can be exhibited under the stated hard rules, and the honest finding is a low circularity score.

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

The paper's argument rests on the unverified transfer from game environments to real-world engineering, and on the applicability of cybernetic frameworks to AI evaluation. No free parameters or invented entities are introduced because the paper is a qualitative proposal.

assumptions (3)
  • domain assumption Sandbox games can train and evaluate real-world system engineering skills.
    Invoked throughout, especially in Section 3.4 and Section 5, to justify the use of Factorio as a proxy for real engineering challenges.
  • domain assumption Ashby's Law of Requisite Variety and Beer's Viable System Model are appropriate frameworks for evaluating AI agents.
    Section 3.2 builds the evaluation criteria on these cybernetic theories, but their suitability for AI agents is not empirically validated.
  • domain assumption Factorio's mechanics sufficiently capture the essence of real-world systems (resource flows, bottlenecks, scalability, adaptability).
    Section 4.1 asserts that Factorio mirrors real-world engineering dilemmas, but no evidence is provided that this mapping holds beyond analogy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Develop AI Agents for System Engineering in Factorio." pith.science (2026). https://pith.science/paper/OGBZHYHH

@misc{pith2026250201492,
  author       = {Pith},
  title        = {Pith review of: Develop AI Agents for System Engineering in Factorio},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OGBZHYHH}},
  note         = {Machine review of arXiv:2502.01492}
}
read the original abstract

Continuing advances in frontier model research are paving the way for widespread deployment of AI agents. Meanwhile, global interest in building large, complex systems in software, manufacturing, energy and logistics has never been greater. Although AI driven system engineering holds tremendous promise, the static benchmarks dominating agent evaluations today fail to capture the crucial skills required for implementing dynamic systems, such as managing uncertain trade-offs and ensuring proactive adaptability. This position paper advocates for training and evaluating AI agents' system engineering abilities through automation-oriented sandbox games-particularly Factorio. By directing research efforts in this direction, we can equip AI agents with the specialized reasoning and long-horizon planning necessary to design, maintain, and optimize tomorrow's most demanding engineering projects.

Figures

Figures reproduced from arXiv: 2502.01492 by the authors.

Figure 1
Figure 1. The Law of Requisite Variety. TV : A → E is a trajec￾tory where a system stays viable through adaptation. TU : A → D′ shows an alternate trajectory where the system does not adapt and becomes unviable. A system is viable within the total state space S when the variety of the environment at that time VE re￾mains a subset of variety the system can handle VR. Systems must adapt proactively (A → B) to ensure this condit… view at source ↗
Figure 2
Figure 2. The Viable System Model. Systems are organized into five levels concisely given as: 1. operational units, 2. coordination, 3. present optimization, 4. future planning, and 5. ultimate policy. See [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. An example of early-game resource extraction and smelt￾ing in Factorio. Box A shows mining drills extracting iron ore, Box B highlights stone furnaces which take ore and fuel and create plates, and Box C highlights belt routing and inserter mechanics. (s6pgbu) We first refer to [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figures from the paper (11 more)
Figure 5
Figure 5. Figure 5: Recipe for logistic (green) science packs. Automating intermediate goods significantly reduces total crafting time from 8.7 seconds (raw) to 6 seconds. A.3. Science Packs and the Tech Tree The science packs produced in [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 4
Figure 4. Figure 4: Dependency graph for red and green science packs. In￾puts include both raw materials and intermediates, reflecting the growing complexity of production chains. There are thus many points to consider when designing assembly lines for these finished goods. The throughput…
Figure 7
Figure 7. Figure 7: Nuclear power generation is actually quite realistic in Factorio. Uranium ore is mined, the vast majority of which (99.3%) is inert U-238. The more valuable U-235 is needed for energy￾intensive applications. There is an enrichment process by which U-238 can be refined …
Figure 8
Figure 8. Figure 8: Biters are alien residents of the planet where the agent has crash landed. They are docile initially but become aggravated by air pollution from the factory’s hydrocarbon-powered operations. Thinking about biters is thus a core trade-off of expanding systems in Factori…
Figure 9
Figure 9. Figure 9: Defense against Biters is essentially a resource sink in Factorio. Settings and mods can be used to dramatically change the difficulty associated with defending bases from Biters. Early defenses rely on a combination of walls and gun tur￾rets, as seen in [PITH_FULL_IM…
Figure 10
Figure 10. Figure 10: A red and green science production setup. All belts are running from left to right. Iron and copper plates enter on the bottom-most belt (Box A). There are assemblers throughout the line which have certain recipes selected. For example Box B has the assembler responsi…
Figure 11
Figure 11. Figure 11: The main bus design is a common choice for mid-game scaling. Branches for individual component assembly fork off the main bus using belt splitters and underground belts. (Steam) showcasing how thoughtful belt organization and transporta￾tion planning are essential for…
Figure 13
Figure 13. Figure 13: Trains can move large quantities of resources long dis￾tances much faster than belts while reusing the same underlying infrastructure, making them crucial for any scalable build. (Gy￾dron) and ensure smooth throughput. Multiple stations can be linked along a track net…
Figure 14
Figure 14. Figure 14: The variety and customization associated with building rail networks is vast in Factorio. (sparr) Track layouts, particularly intersections and junctions, are another crucial element [PITH_FULL_IMAGE:figures/full_fig_p016_14.png]
Figure 16
Figure 16. Figure 16: Construction robots automate the placement of arbitrar￾ily complex player-made blueprints. Here the blueprint has been partially constructed by robots and needs to be completed and connected to a source of power. (lmit1) 16 [PITH_FULL_IMAGE:figures/full_fig_p016_16.png]
Figure 17
Figure 17. Figure 17: Robots can also be used to efficiently clear out a factory and reclaim the resources. Here the section of the factory has been marked for clearance and robots will swarm it when the player finalizes the selection. (Argoya) Moreover, these same construction robots faci…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI

    cs.SE 2025-05 conditional novelty 3.0 of 10

    A qualitative taxonomy positions vibe coding and agentic coding as complementary paradigms rather than rivals in AI-assisted software development.

Reference graph

Works this paper leans on

56 extracted references · 37 canonical work pages · cited by 1 Pith paper

  1. [1]

    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 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    How to cancel deconstruction

    Argoya, V. How to cancel deconstruction. https://gamerant.com/factorio-how-cancel-deconstruction/. Accessed on 2025-01-29

  3. [3]

    Ashby, R. W. An Introduction to Cybernetics. Chapman and Hall, 1956

  4. [4]

    Z., Xiong, Z., Zintgraf, L., Finn, C., and Whiteson, S

    Beck, J., Vuorio, R., Liu, E. Z., Xiong, Z., Zintgraf, L., Finn, C., and Whiteson, S. A survey of meta-reinforcement learning, 2024. URL https://arxiv.org/abs/2301.08028

  5. [5]

    Cybernetics and Management

    Beer, S. Cybernetics and Management. John Wiley & Sons, 1959

  6. [6]

    Brain of the Firm

    Beer, S. Brain of the Firm. Allen Lane, 1972

  7. [7]

    Blanchard, B. S. and Fabrycky, W. J. Systems Engineering and Analysis. Prentice Hall, 5th edition, 2010

  8. [8]

    Buede, D. M. and Miller, W. D. The Engineering Design of Systems: Models and Methods. Wiley, 3rd edition, 2016

Show all 56 references
  1. [9]

    Industrial revolution 3

    Deadlock989. Industrial revolution 3. https://mods.factorio.com/mod/IndustrialRevolution3. Accessed on 2025-01-29

  2. [10]

    2023 Manufacturing Industry Outlook

    Deloitte. 2023 Manufacturing Industry Outlook . https://www2.deloitte.com/, 2023. [Accessed December 30, 2024]

  3. [11]

    A survey on in-context learning, 2024

    Dong, Q., Li, L., Dai, D., Zheng, C., Ma, J., Li, R., Xia, H., Xu, J., Wu, Z., Liu, T., Chang, B., Sun, X., Li, L., and Sui, Z. A survey on in-context learning, 2024. URL https://arxiv.org/abs/2301.00234

  4. [12]

    Space exploration

    Earendel. Space exploration. https://mods.factorio.com/mod/space-exploration. Accessed on 2025-01-29

  5. [13]

    Nuclear power

    FactorioWiki. Nuclear power. https://wiki.factorio.com/File:Nuclear_setup.png. Accessed on 2025-01-29

  6. [14]

    Minedojo: Building open-ended embodied agents with internet-scale knowledge, 2022

    Fan, L., Wang, G., Jiang, Y., Mandlekar, A., Yang, Y., Zhu, H., Tang, A., Huang, D.-A., Zhu, Y., and Anandkumar, A. Minedojo: Building open-ended embodied agents with internet-scale knowledge, 2022. URL https://arxiv.org/abs/2206.08853

  7. [15]

    Forrester, J. W. Industrial Dynamics. MIT Press, Cambridge, Massachusetts, 1961

  8. [16]

    and Nelson, S

    Glebova, D. and Nelson, S. Trump announces \ 500b ai infrastructure project — with biz titans saying it can cure cancer. New York Post, 2025. URL https://nypost.com/2025/01/21/us-news/trump-announces-500b-ai-project-with-biz-titans-saying-it-can-cure-cancer/

  9. [17]

    Introducing gemini 2.0: our new ai model for the agentic era

    Google DeepMind . Introducing gemini 2.0: our new ai model for the agentic era. Google Blog, 2024. URL https://blog.google/technology/google-deepmind/google-gemini-ai-update-december-2024/

  10. [18]

    Elon musk unveils surprise 'robovan' at flashy tesla event, says optimus humanoid robots will cost less than a car

    Griffin, A. Elon musk unveils surprise 'robovan' at flashy tesla event, says optimus humanoid robots will cost less than a car. New York Post, 2024. URL https://nypost.com/2024/10/11/business/elon-musk-unveils-surprise-robovan-at-flashy-tesla-event-says-optimus-humanoid-robots...

  11. [19]

    train_unloading

    Gydron. train_unloading . https://www.reddit.com/r/factorio/comments/j0ftu0/consuming_a_full_blue_belt_with_3_stack_inserters/. Accessed on 2025-01-29

  12. [20]

    Webvoyager: Building an end-to-end web agent with large multimodal models, 2024

    He, H., Yao, W., Ma, K., Yu, W., Dai, Y., Zhang, H., Lan, Z., and Yu, D. Webvoyager: Building an end-to-end web agent with large multimodal models, 2024. URL https://arxiv.org/abs/2401.13919

  13. [21]

    Measuring massive multitask language understanding, 2021

    Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., and Steinhardt, J. Measuring massive multitask language understanding, 2021. URL https://arxiv.org/abs/2009.03300

  14. [22]

    Holling, C. S. Resilience and stability of ecological systems. Annual Review of Ecology and Systematics, 4 0 (1): 0 1--23, 1973. doi:10.1146/annurev.es.04.110173.000245

  15. [23]

    City blocks

    Imgur. City blocks. https://imgur.com/a/welcome-to-grid-YORiJ. Accessed on 2025-01-29

  16. [24]

    E., Yang, J., Wettig, A., Yao, S., Pei, K., Press, O., and Narasimhan, K

    Jimenez, C. E., Yang, J., Wettig, A., Yao, S., Pei, K., Press, O., and Narasimhan, K. Swe-bench: Can language models resolve real-world github issues?, 2024. URL https://arxiv.org/abs/2310.06770

  17. [25]

    Highly accurate protein structure prediction with alphafold

    Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., et al. Highly accurate protein structure prediction with alphafold. Nature, 596 0 (7873): 0 583--589, 2021. doi:10.1038/s41586-021-03819-2. URL https://www.nature.com/articles/s41586-021-03819-2

  18. [26]

    N., Seymour, S

    Kossiakoff, A., Sweet, W. N., Seymour, S. J., and Biemer, S. M. Systems Engineering Principles and Practice. Wiley, 2nd edition, 2011

  19. [27]

    Towards out-of-distribution generalization: A survey, 2023

    Liu, J., Shen, Z., He, Y., Zhang, X., Xu, R., Yu, H., and Cui, P. Towards out-of-distribution generalization: A survey, 2023. URL https://arxiv.org/abs/2108.13624

  20. [28]

    Construction bots are not building

    lmit1. Construction bots are not building. https://www.reddit.com/r/factorio/comments/b0errp/why_are_the_construction_robots_not_building_the/. Accessed on 2025-01-29

  21. [29]

    Madni, A. M. and Madni, C. C. Architecting the Future of Systems and Software: A Computational Intelligence Perspective. Springer, 2018

  22. [30]

    Building Resilient Semiconductor Supply Chains

    McKinsey & Company . Building Resilient Semiconductor Supply Chains . https://www.mckinsey.com, 2022. [Accessed December 30, 2024]

  23. [31]

    Nvidia announces project gr00t foundation model for humanoid robots and major isaac robotics platform update

    NVIDIA . Nvidia announces project gr00t foundation model for humanoid robots and major isaac robotics platform update. NVIDIA Newsroom, 2024. URL https://nvidianews.nvidia.com/news/foundation-model-isaac-robotics-platform

  24. [32]

    Introducing operator research preview

    OpenAI . Introducing operator research preview. OpenAI, 2025. URL https://openai.com/index/introducing-operator/

  25. [33]

    train_guide

    piper.spirit. train_guide . https://steamcommunity.com/sharedfiles/filedetails/?id=2737259470. Accessed on 2025-01-29

  26. [34]

    and Willson, M

    Price, I. and Willson, M. Gencast predicts weather and the risks of extreme conditions with state-of-the-art accuracy. Google DeepMind Blog, 2024. URL https://deepmind.com/discover/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/

  27. [35]

    Mineflayer: Create minecraft bots with node.js

    PrismarineJS . Mineflayer: Create minecraft bots with node.js. https://github.com/PrismarineJS/mineflayer, 2024. Accessed: 2024-01-28

  28. [36]

    A survey on domain adaptation theory: learning bounds and theoretical guarantees, 2022

    Redko, I., Morvant, E., Habrard, A., Sebban, M., and Bennani, Y. A survey on domain adaptation theory: learning bounds and theoretical guarantees, 2022. URL https://arxiv.org/abs/2004.11829

  29. [37]

    2022 Data Report: Reshoring & FDI Announcements

    Reshoring Initiative . 2022 Data Report: Reshoring & FDI Announcements . https://reshorenow.org, 2023. [Accessed December 30, 2024]

  30. [38]

    Prefixrl: Optimization of parallel prefix circuits using deep reinforcement learning

    Roy, R., Raiman, J., Kant, N., Elkin, I., Kirby, R., Siu, M., Oberman, S., Godil, S., and Catanzaro, B. Prefixrl: Optimization of parallel prefix circuits using deep reinforcement learning. In 2021 58th ACM/IEEE Design Automation Conference (DAC), pp.\ 853–858. IEEE, December ...

  31. [39]

    Example of factorio setup: Basic resource extraction and smelting

    s6pgbu. Example of factorio setup: Basic resource extraction and smelting. https://www.reddit.com/r/factorio/comments/s6pgbu/new_to_game_this_is_my_first_real_attempt_at_a/. Accessed on 2025-01-29

  32. [40]

    Mastering chess and shogi by self-play with a general reinforcement learning algorithm, 2017

    Silver, D., Hubert, T., Schrittwieser, J., Antonoglou, I., Lai, M., Guez, A., Lanctot, M., Sifre, L., Kumaran, D., Graepel, T., Lillicrap, T., Simonyan, K., and Hassabis, D. Mastering chess and shogi by self-play with a general reinforcement learning algorithm, 2017. URL https...

  33. [41]

    train_patterns

    sparr. train_patterns . https://www.reddit.com/r/factorio/comments/8bappn/chunk_aligned_rhd_rail_blueprints_mostly_for_141/. Accessed on 2025-01-29

  34. [42]

    Main bus

    Steam. Main bus. https://steamuserimages-a.akamaihd.net/ugc/492403625774777066/667C0316C478068CEEC1FE3D4196191F86C4849E/. Accessed on 2025-01-29

  35. [43]

    Sterman, J. D. Business Dynamics: Systems Thinking and Modeling for a Complex World. McGraw Hill, Boston, Massachusetts, 2000

  36. [44]

    Congress

    U.S. Congress . CHIPS and Science Act (Public Law 117--167) . https://www.congress.gov/bill/117th-congress/house-bill/4346, 2022. [Accessed December 30, 2024]

  37. [45]

    M., et al

    Vinyals, O., Babuschkin, I., Czarnecki, W. M., et al. Grandmaster level in starcraft ii using multi-agent reinforcement learning. Nature, 575: 0 350--354, 2019. doi:10.1038/s41586-019-1724-z. URL https://doi.org/10.1038/s41586-019-1724-z

  38. [46]

    D., Roedler, G

    Walden, D. D., Roedler, G. J., Forsberg, K. J., Hamelin, R. D., and Shortell, T. M. (eds.). INCOSE Systems Engineering Handbook: A Guide for System Life Cycle Processes and Activities. Wiley, 4th edition, 2015

  39. [47]

    Voyager: An open-ended embodied agent with large language models

    Wang, G., Zheng, L., Hou, L., Xu, P., Liu, L., Park, J., Wang, X., and Wu, Y. Voyager: An open-ended embodied agent with large language models. https://arxiv.org/abs/2305.16291, 2023

  40. [48]

    A comprehensive survey of continual learning: Theory, method and application, 2024

    Wang, L., Zhang, X., Su, H., and Zhu, J. A comprehensive survey of continual learning: Theory, method and application, 2024. URL https://arxiv.org/abs/2302.00487

  41. [49]

    Cybernetics: Or Control and Communication in the Animal and the Machine

    Wiener, N. Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press, 1948

  42. [50]

    Introducing devin, the first ai software engineer

    Wu, S. Introducing devin, the first ai software engineer. Cognition Blog, 2024. URL https://www.cognition.ai/blog/introducing-devin

  43. [51]

    E., Zhang, A

    Yang, J., Jimenez, C. E., Zhang, A. L., Lieret, K., Yang, J., Wu, X., Press, O., Muennighoff, N., Synnaeve, G., Narasimhan, K. R., Yang, D., Wang, S. I., and Press, O. Swe-bench multimodal: Do ai systems generalize to visual software domains?, 2024. URL https://arxiv.org/abs/2...

  44. [52]

    Mmmu: A massive multi-discipline multimodal understanding and reasoning benchmark for expert agi, 2024

    Yue, X., Ni, Y., Zhang, K., Zheng, T., Liu, R., Zhang, G., Stevens, S., Jiang, D., Ren, W., Sun, Y., Wei, C., Yu, B., Yuan, R., Sun, R., Yin, M., Zheng, B., Yang, Z., Liu, Y., Huang, W., Sun, H., Su, Y., and Chen, W. Mmmu: A massive multi-discipline multimodal understanding an...

  45. [53]

    Megablackmarket mod

    ZeroAinz. Megablackmarket mod. https://mods.factorio.com/mod/MegaBlackMarket. Accessed on 2025-01-29

  46. [54]

    P., Zhang, H., Gonzalez, J

    Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., and Stoica, I. Judging llm-as-a-judge with mt-bench and chatbot arena, 2023. URL https://arxiv.org/abs/2306.05685

  47. [55]

    Zhou, C., Liu, G., Liu, P., Du, W., Qiao, Y., Su, H., Deng, Z., and Xing, E. P. Agentbench: Evaluating llms as agents. https://arxiv.org/abs/2308.07490, 2023

  48. [56]

    Diplomacy mod

    ZwerOxotnik. Diplomacy mod. https://mods.factorio.com/mod/diplomacy. Accessed on 2025-01-29

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.