The authors added eight hidden-knowledge question templates to a benchmark built around a generated company.
Enterprise agents struggle when facts are hidden across business systems
Eight new benchmark templates test whether agents can infer unstated facts from conflicting and related enterprise records.
Independent
Benjamin Gruenbaum · Doron Porat · Assaf Natanzon · Roy Zavida · Chen Dinachi · Or Itzahary
Research Digest··2 min read
The authors extend Era by Eon with questions whose answers cannot be found explicitly in any single record or instruction.
Why this paper
Independent · Part of Context Engineering for Agents, now 30 papers
In one line
The best agent answers 18 of 24 hidden knowledge questions; multi-record selection succeeds in only 1 of 84 attempts.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§