What is Jevolution and what does it study?

An instrumented spatial predator-prey framework that makes the dependencies of remote decision-making agent experiments explicit: the world may change during inference, the simulator decides which actions are available, and births and immigration alter who can act.

What does the reference implementation separate?

The evolving world, asynchronous request scheduling, controller adapters, and experiment records. The model combines local observations, energy and resource constraints, predefined signals, and inherited scalar traits, and the mechanisms are described with the ODD protocol.

What did the communication case study find?

Across six complete habitat-seed pairs, the mean signals-on minus signals-off difference in normalized rabbit abundance was -0.0095, with a nominal 95% pair-bootstrap interval of [-0.0772, 0.0602]. The authors say the small study does not resolve the direction of the effect.

Does Jevolution claim emergent language or predictive ecological validity?

No. The paper positions its contribution as an inspectable experimental system and its evidence boundaries, not a new policy-learning algorithm, a demonstration of emergent language or adaptive selection, or biological predictive validity.

How is Jev used in the framework?

In the completed communication case study, Jev controls both species in the predator-prey world, illustrating the end-to-end experiment workflow from world evolution through controller adapters to recorded outcomes.

Where can I find the project and what are its artifacts?

The paper is on arXiv, and the project website is jevolution.world. The preprint runs to 29 pages with 5 figures and 12 tables describing the framework, its implemented controls, and what it explicitly does not establish.

CommunityJev Tools & CLIs

Jevolution: A Framework for Studying Communication and Collective Dynamics in AI Agents (arXiv 2610.14449)

Q: What do the synthetic fixtures actually validate?

They exercise scheduling, signaling, and inheritance against the pinned implementation, and validate specified software behavior rather than model intelligence or ecological realism. The fixtures are free, so specified behavior can be re-checked without the full simulator setup.

An instrumented spatial predator-prey framework for studying communication and collective dynamics in AI agents, separating the evolving world, request scheduling, controller adapters, and experiment records. A case study reports a signals-on minus signals-off rabbit-abundance difference of -0.0095.

See highlights

Open arxiv.org

Format: Article
Published by: Community
Last verified: Oct 2026

Tagscommunity research arxiv multi-agent simulation

Works forJev for Realtime control Jev for Evaluation

Used inJev for Software & DevTools Jev for Data & Analytics

Highlights

Reference implementation separates the evolving world, request scheduling, controller adapters, and records.
Agents combine local observations, energy and resource constraints, predefined signals, and inherited traits.
Mechanisms are described with the ODD protocol, marking implemented controls versus open research questions.
Communication case study controls both species with Jev across six complete habitat-seed pairs.
Mean signals-on minus signals-off rabbit abundance was -0.0095, with a nominal 95% interval [-0.0772, 0.0602].

Watch out

arXiv preprint posted October 10, 2026 (cs.MA, cs.AI, cs.LG) by Xirui Huang, Nikhil Hooda, and Utkarsh Gupta; 29 pages, 5 figures, 12 tables. The six-pair case study is small and does not resolve the direction of the signal-system effect, and the synthetic fixtures validate software behavior rather than model intelligence or ecological realism.

Questions about this entry

Short answers drawn from the linked project or page.

Related terms

Glossary definitions related to this entry.

Autoresearch

More Jev resources like this

KITE: Scaling Jev Population Experiments with Sparse Flagship Calibration (arXiv 2609.27535)

KITE queries a typed behavioral kernel once per unique state, then runs populations of any size from the table, reserving an expensive flagship model for sparse paired anchors. On Epstein experiments with 9,070 participants, anchors covering 1.7% of states cut effect error by 41%.