CMAT 2026: Audit Logs for LLM Metacognitive Resilience

Audit logs and metrics for evaluating metacognitive resilience in frontier large language models. The dataset is hosted on Kaggle, but the author, organization, and specific collection date are unknown. The description suggests it is designed for auditing and evaluating AI model behavior.

Use Cases

• Benchmarking LLM resilience under adversarial prompting based on the described audit logs
• Analyzing latent space tightness and model consistency using the provided metrics
• Training models for improved metacognitive capabilities based on the evaluation framework

Strengths

• Focus on frontier LLM evaluation, a specialized and active research area.
• Data is structured as audit logs and metrics, which suggests a systematic evaluation approach.

Limitations

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

Provenance

Source

Kaggle