cff-version: 1.2.0
message: If you use this dataset, please cite this dataset release.
title: "RoleRouteBench: A Reproducible Multi-Domain Environment for Evaluating Dynamic Multi-Agent Routing"
type: dataset
version: 4.0.0-beta.2
doi: 10.57967/hf/11430
authors:
  - family-names: Bui
    given-names: Trang
date-released: 2026-11-04
repository-code: https://github.com/BuiThiThanhTrang/RACO_MAS
url: https://huggingface.co/datasets/TrangBui/RoleRouteBench
abstract: >-
  RoleRouteBench is a benchmark for dynamic role routing and multi-agent
  orchestration on multi-step tasks. Version 4.0.0-beta.2 publishes 82 live
  trajectories across database migration and integrity, code debugging, formal
  proof, and structured data processing, together with frozen role pools,
  packet artifacts, evaluator-only references, provenance, and reference JEV
  results.
keywords:
  - multi-agent systems
  - agent routing
  - dynamic routing
  - agent orchestration
  - tool use
  - language model evaluation
  - small language models
  - benchmark