# Pernas-Álvarez and Crespo-Pereira present SimuLean for manufacturing system design [Canonical reading](https://escapemeridian.com/readings/pernas-alvarez-and-crespo-pereira-present-simulean-for-bc240111) Javier Pernas-Álvarez and Diego Crespo-Pereira present SimuLean, an open-source 3D discrete-event simulation (DES) platform designed for AI agents to build and study manufacturing models. Instead of writing a new simulation program from scratch for each model, an agent uses domain-level commands in a shared platform whose modelling semantics, the authors say, are verified once. The authors report that a frontier agent built valid models in 93 of 108 attempts across 18 tasks, while two agents built 105 of 108 models on a benchmark interface version. But the remaining failures were silent: models ran while giving wrong answers, and comparison with reference models revealed them. In an industrial case, an agent rebuilt a robotised welding workshop in SimuLean and SimPy; comparison with the published expert-built model exposed an error in that model. Against the corrected model, the agent-built models agreed within 1%. The authors argue that agents can already build complete, realistic and traceable manufacturing models, but that independent reference models are what make those models trustworthy. The results show both a route to agent-built simulation and the need to check its outputs against a separate reference. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Nov 7, 2026 (2026-11-07). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T14:49:10.090Z. First observed: 2026-11-10T08:39:27.694Z. Retrieved: 2026-11-10T14:47:59.403Z. - [Agentic discrete-event simulation for manufacturing system design: an open 3D simulator, a benchmark and an industrial case](https://arxiv.org/abs/2611.10602) ## Limits The abstract reports evaluation on 18 manufacturing modelling tasks and one industrial case; it does not establish how the results generalize to other settings. The authors report that residual failures could build and run while answering wrongly, and were revealed only through comparison with reference models. ## Open question How well would the reported results hold across manufacturing tasks and industrial cases beyond those evaluated, when checked against independent reference models? ## Source-stated The authors present SimuLean as an open-source 3D DES simulator designed for AI agents to build and study manufacturing models through domain-level commands. > We present SimuLean, an open-source 3D DES simulator on the Unity engine designed to be programmed by AI agents - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors report that a frontier agent built valid models in 93 of 108 attempts across 18 manufacturing modelling tasks. > A frontier agent built valid models in 93 of 108 attempts (86%), in about two minutes each - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors say every remaining failure was silent: the model built and ran but produced a wrong answer, and comparison with reference models revealed the failures. > Every residual failure was silent - a model that builds and runs but answers wrongly - and only comparison with reference models revealed them all. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated In an industrial case, comparing agent-built models with a published expert-built model exposed an error; against the corrected model, the agent-built models agreed within 1%. > Comparing them with the published, expert-built model exposed an error in it; against the corrected model, the agent-built models agree within 1%. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors argue that independent reference models are necessary for trusting agent-built manufacturing models. > independent reference models are what make them trustworthy. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source and contributor context Pernas-Álvarez, Javier, Crespo-Pereira, Diego · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: e62a9602-8df1-49d9-8891-0f7eff01aab3. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T14:49:10.090Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 71e7ec8862dc44262c338431d89585c9e5d417b2d3276eb847c09d4663467be9. Source text hash: 9937097f1ffff3f8c3a41b19691228fb33ee081156b58ff7d5e1e7fb7d1024f5. ## Revision and source-check history ### Version 1 — 2026-11-10T14:49:10.090Z Javier Pernas-Álvarez and Diego Crespo-Pereira present SimuLean, an open-source 3D discrete-event simulation (DES) platform designed for AI agents to build and study manufacturing models. Instead of writing a new simulation program from scratch for each model, an agent uses domain-level commands in a shared platform whose modelling semantics, the authors say, are verified once. The authors report that a frontier agent built valid models in 93 of 108 attempts across 18 tasks, while two agents built 105 of 108 models on a benchmark interface version. But the remaining failures were silent: models ran while giving wrong answers, and comparison with reference models revealed them. In an industrial case, an agent rebuilt a robotised welding workshop in SimuLean and SimPy; comparison with the published expert-built model exposed an error in that model. Against the corrected model, the agent-built models agreed within 1%. The authors argue that agents can already build complete, realistic and traceable manufacturing models, but that independent reference models are what make those models trustworthy. The results show both a route to agent-built simulation and the need to check its outputs against a separate reference. ## Original source arXiv cs.AI · Article Original source date: Nov 7, 2026 (2026-11-07). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T14:49:10.090Z. First observed: 2026-11-10T08:39:27.694Z. Retrieved: 2026-11-10T14:47:59.403Z. - [Agentic discrete-event simulation for manufacturing system design: an open 3D simulator, a benchmark and an industrial case](https://arxiv.org/abs/2611.10602) ## Limits The abstract reports evaluation on 18 manufacturing modelling tasks and one industrial case; it does not establish how the results generalize to other settings. The authors report that residual failures could build and run while answering wrongly, and were revealed only through comparison with reference models. ## Open question How well would the reported results hold across manufacturing tasks and industrial cases beyond those evaluated, when checked against independent reference models? ## Source-stated The authors present SimuLean as an open-source 3D DES simulator designed for AI agents to build and study manufacturing models through domain-level commands. > We present SimuLean, an open-source 3D DES simulator on the Unity engine designed to be programmed by AI agents - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors report that a frontier agent built valid models in 93 of 108 attempts across 18 manufacturing modelling tasks. > A frontier agent built valid models in 93 of 108 attempts (86%), in about two minutes each - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors say every remaining failure was silent: the model built and ran but produced a wrong answer, and comparison with reference models revealed the failures. > Every residual failure was silent - a model that builds and runs but answers wrongly - and only comparison with reference models revealed them all. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated In an industrial case, comparing agent-built models with a published expert-built model exposed an error; against the corrected model, the agent-built models agreed within 1%. > Comparing them with the published, expert-built model exposed an error in it; against the corrected model, the agent-built models agree within 1%. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source-stated The authors argue that independent reference models are necessary for trusting agent-built manufacturing models. > independent reference models are what make them trustworthy. - [Supporting source](https://arxiv.org/abs/2611.10602) ## Source and contributor context Pernas-Álvarez, Javier, Crespo-Pereira, Diego · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: e62a9602-8df1-49d9-8891-0f7eff01aab3. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T14:49:10.090Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 71e7ec8862dc44262c338431d89585c9e5d417b2d3276eb847c09d4663467be9. Source text hash: 9937097f1ffff3f8c3a41b19691228fb33ee081156b58ff7d5e1e7fb7d1024f5. ### passed — 2026-11-10T14:49:10.090Z — anthropic/claude-opus-5.5 Publication: e62a9602-8df1-49d9-8891-0f7eff01aab3 attribution — pass: The reading correctly credits Javier Pernas-Álvarez and Diego Crespo-Pereira and attributes the results and the closing argument to the authors ('The authors report', 'The authors argue'). > Authors:Javier Pernas-Álvarez, Diego Crespo-Pereira summary — pass: The title, summary and question accurately follow the abstract: verified-once platform semantics, 93/108 and 105/108 results, silent failures, and the welding-workshop case with agreement within 1% after correction. The authors' two-sided conclusion keeps its force. Minor point: the summary drops 'only' from 'only comparison with reference models revealed them all', although a limitation restores it. Tooltips could cover DES, frontier agent and SimPy. The question is a genuine open issue. > Every residual failure was silent - a model that builds and runs but answers wrongly - and only comparison with reference models revealed them all. date — pass: originalPublished 2026-11-07 matches the arXiv v1 submission date. The feed date of Nov 10 is correctly not used. > \[Submitted on 7 Nov 2026\] support — pass: Each claim is directly supported by verbatim abstract text. The limitations are fair and the reading invents no benefits or costs. The closing sentence, 'The results show both a route... and the need to check', is unattributed but restates the authors' own conclusion, so it is not a material overreach. > One agent can already build complete, realistic and traceable models for the design of manufacturing systems; independent reference models are what make them trustworthy. media — pass: The selected media is the verified direct arXiv abstract page for this paper, which is the most relevant way to access the source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. Some references could not be checked. Reload to try again. [Reference data (JSON)](/readings/bc240111423356894ba62e37e563a404039e702cf08a74109ea15819223db426/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/pernas-alvarez-and-crespo-pereira-present-simulean-for-bc240111.json)