Which Metrics Save the Most Human Annotation? Prediction-Powered Evaluation and Meta-Evaluation

arXiv:2608.26638 · cs.CL, stat.ML · Submitted 2026-08-27 · Read on arXiv

cs.CL, stat.ML

Submitted: 2026-08-27

Updated: 2026-08-30

Comments: Accepted at EMNLP 2026 (Main). Code available: https://github.com/CHATS-lab/ppi-eval

Code: https://github.com/CHATS-lab/ppi-eval

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Empirical Coverage — 15× in this paper · explained in 2 episodes
• Accuracy — 6× in this paper · explained in 1 episode
• Pairwise Accuracy — 5× in this paper
• Consistency — 4× in this paper · explained in 3 episodes
• Calibration — 3× in this paper · explained in 7 episodes
• Prediction-Powered Saving Ratio — 3× in this paper
• LLM-as-a-Judge — 2× in this paper · explained in 3 episodes
• Adam — 1× in this paper · explained in 2 episodes
• Aggregation — 1× in this paper · explained in 1 episode
• Interpretability — 1× in this paper · explained in 15 episodes
• Verifiable Rewards — 1× in this paper · explained in 2 episodes

Sources

• PPI++: Efficient Prediction-Powered Inference
• Power Analysis for Prediction-Powered Inference
• Efficient Inference for Noisy LLM-as-a-Judge Evaluation
• Noisy but Valid: Robust Statistical Evaluation of LLMs with Imperfect Judges
• Doubly-Robust LLM-as-a-Judge: Externally Valid Estimation with Imperfect Personas
• Predictions as Surrogates: Revisiting Surrogate Outcomes in the Age of AI
• Writing-Zero: Bridge the Gap Between Non-verifiable Tasks and Verifiable Rewards
• How to Correctly Report LLM-as-a-Judge Evaluations
• No Free Lunch: Non-Asymptotic Analysis of Prediction-Powered Inference
• A Note on the Prediction-Powered Bootstrap

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering