DecisionOS_v2: A Multi-Agent Workflow for Evidence-Based Business Decisions

This paper presents DecisionOS v2, a prototype workflow for evidence-based business decisions that distributes tasks across six software roles: Orchestrator, Data, Analysis, Risk, Recommendation, and Report. The workflow combines document retrieval, predictive models, risk review, a rule-based confidence indicator, and human approval. Existing project materials report preliminary results for customer-churn classification and sales forecasting. These predictive metrics do not establish the decision quality of the complete workflow. The paper defines a reproducible evaluation plan to compare the system with simpler alternatives, examine individual components, and assess evidence support, decision quality, latency, and cost.