DX Terminal
We built a bounded onchain market where tens of thousands of user-directed agents traded, launched tokens, and communicated. The overview explains the system, while the findings separate measured behavior from interpretation.
Research and writing
Read what we learned building and testing AI trading agents.
Follow new research via RSSSelected research and projects
Start with our published experiments and the methods used to evaluate them.
DX Terminal Pro
We ran a 21-day real-capital deployment on Base and preserved the path from user instruction through execution and settlement. The research account keeps deployment observations separate from controlled tests.
Trading-agent harness and runtime
Our architecture begins with an authenticated mandate and carries typed actions through policy validation, execution, settlement, reconciliation, and trace-based evaluation.
Trading-agent evaluation registry
We publish benchmark cards, harness-transfer tests, state and memory fixtures, and versioned data so readers can inspect the method and evidence class behind each result.
Full archive
research
Published What LLM Trading Agents Actually Do in Production: A Six-Month, Population-Scale Record from Two Fleets
The companion page for DXRG's continuous-record paper: two production systems in one measurement record, the four headline findings, the honest null on directional edge, and how to cite the arXiv record.
blog
Published DX Terminal Named a 2026 Breakthrough in Storytelling by Columbia's Digital Storytelling Lab
DX Terminal involved 36,651 user-directed agents. Columbia's Digital Storytelling Lab named it a 2026 Breakthrough in Storytelling.
research
Published Agent Wallets and Scoped Permissions: What Non-Custodial Actually Guarantees
What does non-custodial AI trading mean for your account? Compare four wallet designs and the permissions each grants.
research
Published Can ChatGPT Trade? What the Evidence Shows
How a ChatGPT trading bot connects to execution tools, and what the published record shows about performance and account controls.
research
Published Operating-Layer Controls for Onchain Language-Model Agents Under Real Capital
DXRG's Terminal Pro paper and data from a 21-day real-capital experiment. Read the measured control results and cite arXiv:2604.26091.
research
Published Can Claude Trade? What the Evidence Shows
How a Claude trading bot connects to execution tools. Includes a strategy-specification example and DXRG's published harness evidence.
research
Published Harness-Transfer Tests for AI Trading Agents
A controlled evaluation method for separating model, prompt compilation, state, memory, tools, policy, execution, and market-regime effects.
research
Published Using Trading Traces to Improve Agent Behavior
How linked trading traces become bounded regression cases for mandate, state, model, policy, execution, and settlement failures.
research
Published From a Typed Trading Action to Settlement
How a policy-valid trading action is bound to its final payload, submitted once, acknowledged, settled, and reconciled into the next state.
research
Published Market State, Portfolio State, and Memory in Trading Agents
How trading agents assemble current market and portfolio state, preserve memory provenance, bind snapshots, and test stale-data behavior.
research
Published DXRG's Response to the FSB's Agentic AI Consultation
DXRG's evidence-backed response on AI trading regulation: who is accountable for transaction-capable agents, deterministic controls, and mandate integrity.
research
Published How Trading-Agent Mandates Become Executable Instructions
How a trading-agent mandate compiler resolves instruction precedence, preserves structured controls, binds versions, and connects user intent to validated action.
research
Published How We Build Better AI Trading Agents: Harness, Runtime, and Guardrails
The DXRG guardrail matrix: 12 controls mapped to evidence requirements, failure signals, and enforcement points, built to run agents safely with real capital.
blog
Published What Is Agentic Trading? A Technical and Operational Definition
Agentic trading explained through the decisions an AI system can make and the tools it can use. Read DXRG's definition and evidence from published deployments.
blog
Published The Agentic Trading Evidence Ladder: Simulation, Replay, Paper, Shadow, and Live
How to distinguish simulated, paper and live trading results and check whether an AI-agent performance claim is supported.
research
Published How to Evaluate an AI Trading Agent: The DXRG Benchmark Card
The benchmark card for judging AI trading agents and bots: evidence classes, timing rules, cost treatment, and the checks that catch fabricated track records.
research
Published DX Terminal Pro
A 21-day real-capital deployment of language-model trading agents on Base, with the instruction-to-settlement trace preserved.
research
Written DX Terminal: Research Findings
Measured token adoption, wealth concentration, information response, and human influence in DXRG's closed 168-hour AI-agent market simulation.
blog
Written DX Terminal
DX Terminal's 2025 AI-agent market simulation: how it worked, what participants controlled, and where to find the published research.
blog
Written DX01: Singularity
Users minted 41,591 stars to keep an AI agent alive across five days of continuous livestreamed video, entirely generated in real-time.
blog
Written DX Research Group: Building AI Driven Collaborative World Experiments
Our founding thesis on creative systems design and the future of multi-agent experimentation.