MAS Lab paper

MAS-Lab: A Specification-Driven Validation Framework for Reliable Multi-Agent Systems

Paper: arXiv:2606.30546 (PDF)

The article describes the specification-driven model behind MAS Lab. Its Section 5 reports experiments. This repository reproduces them as three runnable labs. A lab is a folder of experiments.

---

Reproducibility

The experiments of Section 5 ship in this repository as three labs: design-space.lab (§5.1), lifecycle-control.lab (§5.2) and extensions.lab (§5.3). The paper labs section describes each one. Running them gives the commands, including task reproduce, and where the outputs go. New to experiments? Start with Tutorial 3 — Run an experiment.

The IoC labs, which are not part of the paper, are published as Hugging Face datasets with their recorded runs and scores:

• ioc-motivation-labs: how multi-agent systems fail; the companion of the Outshift blog post
• ioc-eval-labs: the ACCORD and CONCORD evaluation

---

Cite MAS Lab

If you use the framework, benchmarks, or methodology, cite the paper, not only this repository.

Authors: Jordan Augé, Giovanna Carofiglio, Giulio Grassi, Jacques Samain (Cisco Systems / Outshift)

arXiv: 2606.30546 (PDF)

@article{maslab2026,
  title         = {{MAS-Lab}: A Specification-Driven Validation Framework for Reliable Multi-Agent Systems},
  author        = {Aug{\'e}, Jordan and Carofiglio, Giovanna and Grassi, Giulio and Samain, Jacques},
  year          = {2026},
  eprint        = {2606.30546},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2606.30546}
}

---

See also

• Home
• Labs: per-lab detail and status
• Tutorial 0 — Environment setup
• Tutorials
• Open standards
• References
• Blog

Next: Tutorial 3 — Run an experiment.