# How Reusable Are Benchmarks with Richer Feedback?

arXiv: https://arxiv.org/abs/2609.32109 · HTML: https://mlsanity.com/paper/2609.32109/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 28 references that are in the map.
- First posted on arXiv 26 Sep 2026.

## Builds on (most cited first, 19 of 19)

- [Do ImageNet Classifiers Generalize to ImageNet?](https://mlsanity.com/paper/1902.10811/) (2019)
- [Concentrated Differential Privacy: Simplifications, Extensions, and Lower Bounds](https://mlsanity.com/paper/1605.02065/) (2016)
- [LiveBench: A Challenging, Contamination-Limited LLM Benchmark](https://mlsanity.com/paper/2406.19314/) (2024)
- [Preserving Statistical Validity in Adaptive Data Analysis](https://mlsanity.com/paper/1411.2664/) (2014)
- [Algorithmic Stability for Adaptive Data Analysis](https://mlsanity.com/paper/1511.02513/) (2015)
- [The Ladder: A Reliable Leaderboard for Machine Learning Competitions](https://mlsanity.com/paper/1502.04585/) (2015)
- [The Leaderboard Illusion](https://mlsanity.com/paper/2504.20879/) (2025)
- [Dynaboard: An Evaluation-As-A-Service Platform for Holistic Next-Generation Benchmarking](https://mlsanity.com/paper/2106.06052/) (2021)
- [Model Similarity Mitigates Test Set Overuse](https://mlsanity.com/paper/1905.12580/) (2019)
- [Interactive Fingerprinting Codes and the Hardness of Preventing False Discovery](https://mlsanity.com/paper/1410.1228/) (2014)
- [Preventing False Discovery in Interactive Data Analysis is Hard](https://mlsanity.com/paper/1408.1655/) (2014)
- [Inherent Trade-Offs between Diversity and Stability in Multi-Task Benchmarks](https://mlsanity.com/paper/2405.01719/) (2024)
- [A bounded-noise mechanism for differential privacy](https://mlsanity.com/paper/2012.03817/) (2020)
- [The advantages of multiple classes for reducing overfitting from test set reuse](https://mlsanity.com/paper/1905.10360/) (2019)
- [Climbing a shaky ladder: Better adaptive risk estimation](https://mlsanity.com/paper/1706.02733/) (2017)
- [Natural Analysts in Adaptive Data Analysis](https://mlsanity.com/paper/1901.11143/) (2019)
- [Fingerprinting Codes Meet Geometry: Improved Lower Bounds for Private Query Release and Adaptive Data Analysis](https://mlsanity.com/paper/2412.14396/) (2024)
- [Approximate Convex Hull of Data Streams](https://mlsanity.com/paper/1712.04564/) (2017)
- [Who Defines "Best"? Towards Interactive, User-Defined Evaluation of LLM Leaderboards](https://mlsanity.com/paper/2604.21769/) (2026)

## Abstract (arXiv)

We study whether benchmarks reliably guide model selection as developers adapt to evaluation feedback across multiple criteria. We find that the worst-case test-set size needed to estimate the best score among k adaptively chosen models, under any convex combination of the criteria, grows exponentially with the number of criteria, reaching the Theta(√k) cost of answering k adaptive statistical queries with only O(log k) criteria, at fixed accuracy and confidence. In attacks on multi-task large language model benchmarks with five to ten criteria, feedback restricted to nondominated task profiles produces large reused-to-held-out score gaps and frequent false winners. These results challenge a prominent explanation for prior observed reliable benchmark reuse---that developers mainly respond to convincing improvements over the current best---in rich-feedback settings, while leaving open how often ordinary model development encounters this vulnerability.