BA Agent RL Environment and Benchmark

Pipeline

Feature Brief

No task loaded. Choose a feature from the sidebar.

Input Documents

(no documents)

Action Runner

BA Agent Bench Leaderboard — 8 tasks · 6-metric composite

Frontier LLMs and Centific BA Toolkit on enterprise requirements generation. Composite = Alignment 35 · Coherence 24 · Completeness 18 · Compliance 10 · Testability 9 · Spec Quality 4.

#	Model	Composite	Algn	Coh	Cmpl	Cmpl-ce	Test	Spec

Metric Weights

Notes

Centific BA Toolkit = the multi-agent pipeline whose reward shape this RL env mirrors.
Full leaderboard, harness, and dataset details: centific.com/benchmark/agentic-rl/1

About this environment

BA Agent RL Environment turns the BA Agent Bench into a trainable RL setting. Agents run a four-stage requirements-engineering pipeline over real enterprise work items and earn dense per-step rewards plus a benchmark-aligned terminal score on the generated user stories.

Dataset: CentificAIResearch/BA-Agent-Bench · 8-feature stratified subset.