cff-version: 1.2.0
message: "If you use this benchmark, please cite it as below."
title: "mssense Evaluation Benchmark for Closed-Vocabulary Action Trace Generation"
version: "1.1-eval"
date-released: "2026-07-01"
license: CC-BY-4.0
type: dataset
authors:
  - family-names: "Yahaya Alassan"
    given-names: "Mahaman Sanoussi"
    affiliation: "Novelis, Paris, France"
  - family-names: "Ettifouri"
    given-names: "El Hassane"
    affiliation: "Novelis, Paris, France"
  - family-names: "Dahhane"
    given-names: "Walid"
    affiliation: "Novelis, Paris, France"
keywords:
  - "Robotic Process Automation"
  - "Domain-Specific Languages"
  - "Structured Generation"
  - "Closed-Vocabulary Generation"
  - "Benchmark"
abstract: >-
  An evaluation-only benchmark (1865 samples, 9 task families) for
  closed-vocabulary action trace generation in conversational RPA authoring:
  clarification policy, LAT audit, semantic judgment, workflow creation,
  business-rule extraction, visual grounding/governance and modification intent.
doi: 10.5281/zenodo.21105006
identifiers:
  - type: doi
    value: 10.5281/zenodo.21105006
    description: "Zenodo record for this release"