Arbium releases HelixCove, a dataset of organizational work traces for evaluating frontier models

Arbium says HelixCove captures computer activity over time as "human traces" recording the actions, decisions, and context behind knowledge work. The company evaluated frontier models on 430 nontrivial questions about how an organization works, finding that models can answer such questions using human traces alone. The dataset was generated using patterns from hundreds of thousands of hours of real work captured by Arbium, in an environment that keeps people, documents, and decisions connected. A public playground and further reading are linked in the announcement.

✨ AI summary of the linked source — not editorial content.