# Towards Reliable LLM Evaluation: Correcting the Winner's Curse in Adaptive Benchmarking

arXiv: https://arxiv.org/abs/2605.05973 · HTML: https://mlsanity.com/paper/2605.05973/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- Builds on 28 of its 37 references that are in the map.
- First posted on arXiv 7 May 2026.

## Builds on (most cited first, 28 of 28)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [PaLM 2 Technical Report](https://mlsanity.com/paper/2305.10403/) (2023)
- [Large Language Models Are Human-Level Prompt Engineers](https://mlsanity.com/paper/2211.01910/) (2022)
- [Large Language Models as Optimizers](https://mlsanity.com/paper/2309.03409/) (2023)
- [DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines](https://mlsanity.com/paper/2310.03714/) (2023)
- [State of What Art? A Call for Multi-Prompt LLM Evaluation](https://mlsanity.com/paper/2401.00595/) (2023)
- [tinyBenchmarks: evaluating LLMs with fewer examples](https://mlsanity.com/paper/2402.14992/) (2024)
- [Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs](https://mlsanity.com/paper/2406.11695/) (2024)
- [Preserving Statistical Validity in Adaptive Data Analysis](https://mlsanity.com/paper/1411.2664/) (2014)
- [Exact post-selection inference, with application to the lasso](https://mlsanity.com/paper/1311.6238/) (2013)
- [Anchor Points: Benchmarking Models with Much Fewer Examples](https://mlsanity.com/paper/2309.08638/) (2023)
- [Efficient multi-prompt evaluation of LLMs](https://mlsanity.com/paper/2405.17202/) (2024)
- [The Ladder: A Reliable Leaderboard for Machine Learning Competitions](https://mlsanity.com/paper/1502.04585/) (2015)
- [Optimal Inference After Model Selection](https://mlsanity.com/paper/1410.2597/) (2014)
- [Active Statistical Inference](https://mlsanity.com/paper/2403.03208/) (2024)
- [Can Unconfident LLM Annotations Be Used for Confident Conclusions?](https://mlsanity.com/paper/2408.15204/) (2024)
- [Efficient Prompt Optimization Through the Lens of Best Arm Identification](https://mlsanity.com/paper/2402.09723/) (2024)
- [How Benchmark Prediction from Fewer Data Misses the Mark](https://mlsanity.com/paper/2506.07673/) (2025)
- [Cost-Optimal Active AI Model Evaluation](https://mlsanity.com/paper/2506.07949/) (2025)
- [Efficient Evaluation of LLM Performance with Statistical Guarantees](https://mlsanity.com/paper/2601.20251/) (2026)
- [Prompt Risk Control: A Rigorous Framework for Responsible Deployment of Large Language Models](https://mlsanity.com/paper/2311.13628/) (2023)
- [On Speeding Up Language Model Evaluation](https://mlsanity.com/paper/2407.06172/) (2024)
- [A Framework for Efficient Model Evaluation through Stratification, Sampling, and Estimation](https://mlsanity.com/paper/2406.07320/) (2024)
- [ReliableEval: A Recipe for Stochastic LLM Evaluation via Method of Moments](https://mlsanity.com/paper/2505.22169/) (2025)
- [Bandit-Based Prompt Design Strategy Selection Improves Prompt Optimizers](https://mlsanity.com/paper/2503.01163/) (2025)
- [Efficient Evaluation of Large Language Models via Collaborative Filtering](https://mlsanity.com/paper/2504.08781/) (2025)

## Built on by (most cited first, 7 of 7)

- [Sampling Luck Masquerades as Allocation Gain: Auditing Test-Time Budget Allocation for Neural Combinatorial Optimization](https://mlsanity.com/paper/2608.13087/) (2026)
- [Selection-Aware Stress Testing for Interactive Agents](https://mlsanity.com/paper/2608.30916/) (2026)
- [How Sensitive Are LLM Leaderboard Claims to Hidden Model Selection?](https://mlsanity.com/paper/2609.28177/) (2026)
- [Which Self-Improvements Should We Trust? Reliable Self-Improvement When Agents Reuse Their Benchmarks](https://mlsanity.com/paper/2609.33180/) (2026)
- [Suppressing Pressure, Amplifying Evidence: Self-Guided Attention Steering to Mitigate Sycophancy and Stubbornness](https://mlsanity.com/paper/2610.04329/) (2026)
- [The Winner's Curse in LLM Self-Improvement Loops: Selection Noise, Lock-in, and Acceptance Rules](https://mlsanity.com/paper/2610.09239/) (2026)
- [Harness Evolution Hits a Ceiling: When Weight Training Should Begin](https://mlsanity.com/paper/2610.11655/) (2026)

## Abstract (arXiv)

Adaptive prompt and program search makes LLM evaluation selection-sensitive. Once benchmark items are reused inside tuning, the observed winner's score need not estimate the fresh-data performance of the full tune-then-deploy procedure. We study inference for this procedure-level target under explicit tuning budgets. We propose SIREN, a selection-aware repeated-split reporting protocol that freezes the post-search shortlist, separates splitwise selection from held-out evaluation, and uses an item-level Gaussian multiplier bootstrap for uncertainty quantification. In a fixed-shortlist regime with smooth stabilized selection, the estimator admits a first-order item-level representation, and the bootstrap yields valid simultaneous inference on a finite budget grid. This supports confidence intervals for procedure-performance curves and pre-specified equal-budget and cross-budget comparisons. Controlled simulations and MMLU-Pro tuning experiments show that winner-based reporting can be optimistic and can change deployment conclusions, while SIREN remains close to the finite-sample reporting target.