CogEval-Bench: Structural Diagnostic for Agent Memory

CogEval-Bench is a structural diagnostic benchmark for evaluating cognitive structure in agent memory. It is part of the two-layer evaluation in CogniFold: Always-On Proactive Memory via Cognitive Folding, authored by Wang et al. in 2026. The dataset includes 6 scenarios, such as academic_research and customer_support.

Use Cases

• Benchmarking agent memory structure formation based on continuous event streams.
• Evaluating retrieval capabilities beyond simple fact lookup, as described in the diagnostic framework.
• Testing AI agents across diverse scenarios like academic_research and customer_support mentioned in the description.

Strengths

• Designed for a specific, advanced evaluation task: assessing cognitive structure in memory.
• Includes 6 distinct scenarios for testing, providing scenario diversity.
• Part of a published research framework (CogniFold) with an associated paper and code.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file formats are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

OpenNorve

Collection Method

Likely generated as part of the CogniFold research project for structural evaluation.

Freshness

Last updated 2026-05-19 15:11:33; freshness should be verified.