Era by Eon agents solve 18 of 24 hidden-fact tasks

The new Era by Eon benchmark’s best agent answered 18 of 24 hidden-fact attempts correctly, while all systems together got only 2 of 84 hardest questions right. The arXiv paper evaluates 12 agents on enterprise records where answers must be inferred across contradictory data rather than retrieved from one document. Builders should add cross-record and hidden-state cases before granting agents write access; Business Arena’s long-horizon reliability results show why tool completion does not equal capital-safe operation.