TraceMAS: A Traceable Multi-Agent System for Knowledge-Based Forecasting
TraceMAS, a novel interactive demo system, has been launched to tackle the issue of traceability within multi-agent forecasting frameworks. As outlined in a paper on arXiv (2608.03339), this system organizes outputs from agents using two causal-loop models: the Ideal Causal Loop Diagram (Ideal CLD) and the Data-Grounded Causal Loop Diagram (Data-Grounded CLD). The Ideal CLD illustrates essential factors and their causal relationships derived from domain documents, while the Data-Grounded CLD connects these factors to internal variables, external data, or documented proxies. This framework aids in feature development and model design, ensuring a link between textual evidence, data selections, and model updates. It is designed to assist practitioners in understanding forecast changes, the evidence behind those changes, and adjustments made to data and modeling choices. The paper is classified as a new announcement and can be accessed at https://arxiv.org/abs/2608.03339.
Key facts
- TraceMAS is an interactive demo system for traceable multi-agent forecasting.
- It uses two causal-loop representations: Ideal CLD and Data-Grounded CLD.
- Ideal CLD captures key factors and causal relations from domain documents.
- Data-Grounded CLD links factors to internal variables, external data, or documented proxies.
- The system guides feature construction and model design.
- It preserves the connection between textual evidence, data choices, and model revisions.
- The paper is available on arXiv with ID 2608.03339.
- The system addresses the difficulty of inspecting forecast changes and evidence.
Entities
Institutions
- arXiv