# How Reproducible Are Evaluation Conclusions? A Self-Audit of LLM-Inferred Prompt Structure

arXiv: https://arxiv.org/abs/2609.30074 · HTML: https://mlsanity.com/paper/2609.30074/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 15 of its 20 references that are in the map.
- First posted on arXiv 24 Sep 2026.

## Builds on (most cited first, 15 of 15)

- [Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity](https://mlsanity.com/paper/2104.08786/) (2021)
- [Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting](https://mlsanity.com/paper/2310.11324/) (2023)
- [How is ChatGPT's behavior changing over time?](https://mlsanity.com/paper/2307.09009/) (2023)
- [Adding Error Bars to Evals: A Statistical Approach to Language Model Evaluations](https://mlsanity.com/paper/2411.00640/) (2024)
- [An Empirical Study of the Non-determinism of ChatGPT in Code Generation](https://mlsanity.com/paper/2308.02828/) (2023)
- [Prompting Is Programming: A Query Language for Large Language Models](https://mlsanity.com/paper/2212.06094/) (2022)
- [(Why) Is My Prompt Getting Worse? Rethinking Regression Testing for Evolving LLM APIs](https://mlsanity.com/paper/2311.11123/) (2023)
- [Guidelines for Empirical Studies in Software Engineering involving Large Language Models](https://mlsanity.com/paper/2508.15503/) (2025)
- [Reflections on the Reproducibility of Commercial LLM Performance in Empirical Software Engineering Studies](https://mlsanity.com/paper/2510.25506/) (2025)
- [Hidden Measurement Error in LLM Pipelines Distorts Annotation, Evaluation, and Benchmarking](https://mlsanity.com/paper/2604.11581/) (2026)
- [Real Money, Fake Models: Deceptive Model Claims in Shadow APIs](https://mlsanity.com/paper/2603.01919/) (2026)
- [Skill Coverage: A Test Adequacy Metric for Agent Skills](https://mlsanity.com/paper/2606.20659/) (2026)
- [Rank Intervals for Leaderboards: A Hierarchical Framework for Model Evaluation](https://mlsanity.com/paper/2606.08679/) (2026)
- [MILE: A Mutation Testing Framework of In-Context Learning Systems](https://mlsanity.com/paper/2409.04831/) (2024)
- [A Unified Perturbation Framework for Analyzing Leaderboard Stability and Manipulation](https://mlsanity.com/paper/2605.15761/) (2026)

## Abstract (arXiv)

Evaluations of LLM systems routinely average over small prompt sets and report models as a ranked table. We ask how much confidence such a table deserves, using LLM-based prompt-structure inference as the case study: eight open model variants across five families and 8B to 675B parameters, caching disabled, 293 raw intermediate representations persisted. The measured phenomenon is unstable to begin with. Identical calls do not reliably recover identical structure, with mean node-set Jaccard from 0.39 to 0.96 and 72% of prompt-model cells never node-set-perfect. Auditing the evaluation weakens its conclusions further, and this is our main contribution. Under a joint cluster bootstrap over prompts, only the bottom of the ranking is firm: the two least reproducible models hold rank in 99% and 86% of replicates, the middle four in 27% to 48%, and the top two in 68% each, so the table identifies the worst model reliably but does not reliably identify the best. Two equally defensible rules for merging repeated campaigns change four of eight rows and move the study-wide headline by 7 percentage points. Checking the inferred structure against ground-truth annotations shows reproducibility cannot be read as accuracy. And four of the eight endpoints were withdrawn within ten weeks of measurement, so the study as specified can no longer be run. Small-sample LLM evaluations can therefore look far more definitive than their evidence supports. We recommend reporting rank stability, per-cell provenance, executed sensitivity comparisons, raw per-run outputs, and a measurement date alongside any ranking.