# MIRROR: A Hierarchical Benchmark for Metacognitive Calibration in Large Language Models

arXiv: https://arxiv.org/abs/2604.19809 · HTML: https://mlsanity.com/paper/2604.19809/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 40 references that are in the map.
- First posted on arXiv 15 Apr 2026.

## Builds on (most cited first, 21 of 21)

- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Datasheets for Datasets](https://mlsanity.com/paper/1803.09010/) (2018)
- [OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments](https://mlsanity.com/paper/2404.07972/) (2024)
- [GAIA: a benchmark for General AI Assistants](https://mlsanity.com/paper/2311.12983/) (2023)
- [Towards Understanding Sycophancy in Language Models](https://mlsanity.com/paper/2310.13548/) (2023)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Discovering Language Model Behaviors with Model-Written Evaluations](https://mlsanity.com/paper/2212.09251/) (2022)
- [SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models](https://mlsanity.com/paper/2303.08896/) (2023)
- [Augmented Language Models: a Survey](https://mlsanity.com/paper/2302.07842/) (2023)
- [Chain-of-Verification Reduces Hallucination in Large Language Models](https://mlsanity.com/paper/2309.11495/) (2023)
- [Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models](https://mlsanity.com/paper/2305.19187/) (2023)
- [Large Language Models Must Be Taught to Know What They Don't Know](https://mlsanity.com/paper/2406.08391/) (2024)
- [Metacognitive Prompting Improves Understanding in Large Language Models](https://mlsanity.com/paper/2308.05342/) (2023)
- [LLMs are Greedy Agents: Effects of RL Fine-tuning on Decision-Making Abilities](https://mlsanity.com/paper/2504.16078/) (2025)
- [Agentic Knowledgeable Self-awareness](https://mlsanity.com/paper/2504.03553/) (2025)
- [Learning When to Act or Refuse: Guarding Agentic Reasoning Models for Safe Multi-Step Tool Use](https://mlsanity.com/paper/2603.03205/) (2026)
- [Metacognitive Myopia in Large Language Models](https://mlsanity.com/paper/2408.05568/) (2024)
- [Monitor-Generate-Verify (MGV): Formalising Metacognitive Theory for Language Model Reasoning](https://mlsanity.com/paper/2511.04341/) (2025)
- [MetaCLASS: Metacognitive Coaching for Learning with Adaptive Self-regulation Support](https://mlsanity.com/paper/2602.02457/) (2026)
- [MedCoG: Maximizing LLM Inference Density in Medical Reasoning via Meta-Cognitive Regulation](https://mlsanity.com/paper/2602.07905/) (2026)
- [AI, Metacognition, and the Verification Bottleneck: A Three-Wave Longitudinal Study of Human Problem-Solving](https://mlsanity.com/paper/2601.17055/) (2026)

## Built on by (most cited first, 6 of 6)

- [Metacognition in LLMs: Foundations, Progress, and Opportunities](https://mlsanity.com/paper/2607.11881/) (2026)
- [EvalResearchBench: Can AI Agents Design Their Own Evaluations?](https://mlsanity.com/paper/2610.04184/) (2026)
- [Reality Monitoring in Large Language Models: Self-Knowledge That Transforms with Conversation Memory](https://mlsanity.com/paper/2607.23927/) (2026)
- [Agentic Cognitive Depth: Operational Criteria for Evaluating LLM Agents](https://mlsanity.com/paper/2610.04168/) (2026)
- [Not Self-Decidable: LLMs Cannot Draw the Boundary of What an Agent Verifier Can Check](https://mlsanity.com/paper/2610.04699/) (2026)
- [Agentic Self-Awareness: Aligning Knowing, Acting, and Holding](https://mlsanity.com/paper/2610.19226/) (2026)

## Abstract (arXiv)

We introduce MIRROR, a benchmark comprising eight experiments across four metacognitive levels that evaluates whether large language models can use self-knowledge to make better decisions. We evaluate 16 models from 8 labs across approximately 250,000 evaluation instances using five independent behavioral measurement channels. Core experiments are run across the full model roster; experiments with specialized infrastructure requirements report explicitly marked model subsets. We find two phenomena with direct implications for agentic deployment: (1) compositional self-prediction fails universally -- the Compositional Calibration Error ranges from 0.500 to 0.943 on the original 15-model Exp3-v1 set (and 0.434 to 0.758 on the balanced 16-model Exp3-v2 expansion), indicating that models cannot predict their own performance on multi-domain tasks, and (2) models exhibit above-chance but imperfect domain-specific self-knowledge yet systematically fail to translate even this partial awareness into appropriate agentic action-selection -- external metacognitive control reduces the Confident Failure Rate from 0.600 to 0.143 (76% reduction at temperature 0; mean 70% at temperature 0.7 across 5 models from 4 labs). Providing models with their own calibration scores produces no significant improvement (p > 0.05); only architectural constraint is effective. This suggests that external metacognitive scaffolding -- not improved self-knowledge -- is the path to safer autonomous AI systems. Code, data, and Croissant metadata will be released publicly with the benchmark.