Watch the introduction to this project and the broader research program:

🎬 https://youtu.be/dcG4GsKA-uo

A research program investigating whether continuously refined, evidence-backed engineering knowledge can improve AI agent behavior in measurable ways.

This repository is a research program whose current implementation happens to be software. The code exists to answer research questions, not to prove the research correct. Start with RESEARCH.md.

• RESEARCH.md: research charter, principles, and methodological standards.
• ASSUMPTIONS.md: assumptions tracked separately from established knowledge.
• ROADMAP.md: research milestones and their evidence gates.
• ARCHITECTURE.md: conceptual description of each component and why it exists.
• GLOSSARY.md: fixed definitions of key terms.
• REPLICATION.md: how an external person reproduces and audits the experiment.

One action class, an agent initiating or modifying a funds transfer, tested through Milestone 0 and Milestone 1. See ROADMAP.md.

This is a research artifact. The governance rules in this repository are illustrative material for an experiment. They are not validated compliance controls and not legal, regulatory, or financial advice. Nothing here should be treated as authoritative guidance for real financial systems.

MIT. See LICENSE.