Traverse

Traverse is a research lab that builds reinforcement learning environments and training data for frontier AI labs. It is a service for model developers rather than a tool for end users. Its stated focus is subjective and long horizon work, such as tasks in law, healthcare, sales, writing and strategic decision making, where many outputs can be valid and quality depends on judgment, taste and context. The company's argument is that reinforcement learning produced superhuman models in math and coding because those domains are largely deterministic, and that most economically valuable work is not. It holds that ambiguous tasks become verifiable when enough information is captured about the situation, the constraints and the reasoning behind expert decisions. It therefore aims to observe real experts working inside real environments and preserve the reasoning that leads to their decisions, rather than relying on synthetic data from experts answering contrived prompts. The site names the long term goal as giving frontier models the foundations to perform and eventually surpass human white collar work. The company lists its backers and angel investors on the home page. Pricing is not published, and labs interested in working together are asked to get in touch through the contact link.