The Era by Eon Benchmark generates a whole fake company — Salesforce, Zendesk, Slack, Gong simulators — with exact answer keys
arXiv 2609.09853 (2026-09-09) addresses the fact that enterprise agents cannot be evaluated on customer production data and no substitute carries ground truth. Era by Eon seeds an entity graph from industry, company size, business model and application portfolio, then drives product simulators for Salesforce, Zendesk, Slack and Gong plus a question-conditioned generator that builds internal database schemas and records from that same graph, so every expected answer is computed from final records and grading is exact. Across 23 generated companies the mean realism score rose from 61.8 to 97.0 with zero records flagged by the adversarial synthetic detector. In the simulator track, nine models answered the same 33 questions three times each for accuracy between 42.4% and 76.8%, though only three of 36 pairwise differences survived correction — a useful reminder of how few questions most enterprise agent claims rest on.
Source
↳ Follow the thread