TraceMAS Offers Transparent Multi-Agent Forecasting with Causal Loops

Junhyeok Kang, Sangjun Han, Hyeokjun Choe, Soonyoung Lee· August 5, 2026 View original

Key takeaways

  • Autonomous forecasting agents can lack transparency, hindering trust and inspection.
  • TraceMAS uses Causal Loop Diagrams to make multi-agent forecasting traceable.
  • It links textual evidence, data, and model revisions explicitly.
  • Traceability allows users to understand the "why" behind forecast changes.

Who benefits

FinanceEnergySupply ChainBusiness Consulting

Summary

TraceMAS is an interactive demo system for traceable multi-agent forecasting that uses causal loop diagrams to make agent decisions transparent. It links textual evidence, data choices, and model revisions, allowing users to inspect why forecasts change and what evidence supports them.

Enterprise forecasting is increasingly leveraging autonomous agents for tasks like document interpretation, data search, and model revision. While this enhances adaptability, it often obscures the reasoning behind forecast changes. TraceMAS, a new interactive demo system, addresses this by providing transparency. It organizes agent outputs around two types of Causal Loop Diagrams (CLDs): an Ideal CLD derived from domain documents and a Data-Grounded CLD that links these factors to actual data and model choices. This framework allows users to inspect the entire evidence-to-forecast process, including agent revisions, causal maps, and data mappings.

Why it matters

For professionals relying on AI-driven forecasts, understanding the underlying reasoning and evidence is crucial for trust, validation, and strategic decision-making.

How to implement this in your domain

  1. 1Evaluate current forecasting pipelines for transparency and explainability gaps.
  2. 2Explore integrating causal-loop diagramming into existing analytical workflows to map factors and relationships.
  3. 3Pilot a multi-agent system for a specific forecasting task, focusing on logging and visualizing agent decisions and data usage.
  4. 4Develop an interface that allows stakeholders to interactively explore the evidence, data, and model revisions behind a forecast.
  5. 5Train analysts and decision-makers on how to interpret and leverage traceable forecasting outputs for better insights.

Original post by Junhyeok Kang, Sangjun Han, Hyeokjun Choe, Soonyoung Lee

"arXiv:2608.03339v1 Announce Type: new Abstract: Enterprise forecasting increasingly relies on autonomous agents that interpret documents, search for data, generate code, and revise models. While this autonomy helps build adaptive forecasting pipelines, it also makes it difficult…"

View on X

Originally posted by Junhyeok Kang, Sangjun Han, Hyeokjun Choe, Soonyoung Lee on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses