TraceArena is now positioned as an open-source Physical World OS

[tonyhyworld]

TraceArena: from agent demos to runnable AI Worlds

TraceArena is the open-source Physical World OS for multi-agent decision worlds. The project is not primarily a training-data repository. Its core job is to make a physical-world problem runnable: load resources, goals, rules, tools, information boundaries, and authoritative outcomes into a world contract.

Multiple agents can then compete or cooperate under the same constraints. The world simulates consequences, rejects invalid actions, settles outcomes, and compares paths under the declared objective, constraints, and evidence. The result is a decision surface that people can inspect before controlled execution.

What is available today

• no-key deterministic Market Replay
• Incident Response World and benchmark card
• full scenario-pack contract and contribution guide
• replayable evidence, settlement records, and watchable runs
• local Vue frontend and self-hosted runtime

Why this matters

The long-term goal is to let operations, logistics, governance, research, market, and other physical-world problems load into one extensible AI World OS instead of rebuilding a one-off agent demo for every domain. Replay traces are evidence for inspection and improvement; they are not the primary product promise.

Join the next layer

• Run the live demo: https://tonyworld888-tracearena-demo.static.hf.space/index.html
• Follow the Quickstart: https://github.com/tonyhyworld/TraceArena/blob/main/docs/quickstart.md
• Propose a scenario pack in Issue Call for scenario-pack contributors / 首批场景包贡献者招募 #2: Call for scenario-pack contributors / 首批场景包贡献者招募 #2
• Tell us which physical-world problem should be loaded next.

中文：TraceArena 现在正式定位为“面向物理世界问题的开源 AI World OS”。我们不是把 Agent 训练数据作为第一卖点，而是把资源、目标、规则、工具和结果权威装载成可运行的世界，让多智能体在共同约束下博弈、比较路径、接受后果并持续执行。欢迎贡献新的场景包，把下一个真实问题加载进 AI 世界。

— 张诺亚 / Noah Zhang

[tonyhyworld]

v0.1.9 is live: make the world contract the product surface

The public preview is now aligned around the Physical World OS direction:

• load goals, resources, rules, tools, and settlement authority;
• compare competing agent paths under the same constraints;
• replay evidence, actions, recovery, and outcomes instead of judging only the final answer.

The v0.1.9 release includes the updated demo narrative and a direct scenario-pack contribution path.

What is the first physical-world problem you would want to load into this runtime? A small synthetic fixture is enough to start.