What Single-Prompt Accuracy Misses: A Multi-Variant Reliability Audit of Language Models

2026-05-03

AI Summary

This study addresses a critical limitation in current single-prompt accuracy evaluations of language models, which overlook key reliability issues such as miscalibration, parsing failures, and prompt sensitivity. The authors propose a systematic reliability auditing framework driven by multiple prompt variants to assess 15 open-source models across dimensions including accuracy, probability calibration, linguistic confidence, parse success rate, and robustness to prompt perturbations. By introducing techniques such as label-normalized expected calibration error (ECE), chain-of-thought reasoning combined with first-token evaluators, confidence parsing, and perturbation propagation analysis, the work demonstrates how evaluation design profoundly influences conclusions. Findings reveal that assessment methodology can inflate accuracy estimates by up to 88%, linguistic confidence is consistently overestimated, and model size shows no stable correlation with prompt robustness (correlation coefficients ranging from –0.244 to 0.474), underscoring the necessity of transparent reporting of evaluation protocols.

Abstract

Single-prompt accuracy is the dominant way to benchmark language models, but it can miss reliability failures that matter. We evaluate a 15-model open-weight corpus, with the main reliability analyses focused on 10 instruct models across five classification and reasoning benchmarks under five prompt variants each, measuring accuracy, token-probability calibration, verbal-confidence calibration, verbal parse rate, and prompt-perturbation spread for every (model x dataset x variant) cell. We find three broad results. First, evaluation design can materially change the conclusion. Switching Expected Calibration Error (ECE) token from a raw to a label-set-normalised definition changes per-cell calibration by a mean absolute 0.149. More strikingly, pairing a chain-of-thought prompt with a first-character evaluator on ARC-Challenge reduces apparent accuracy by 72-88% across all five primary models; two independent repair procedures recover 93.8% and 102.7% of the lost performance, indicating an evaluator-side rather than model-side failure. Second, confidence signals are fragile. On MMLU-Pro, every primary model verbally reports confidence substantially above both its accuracy and its token-probability confidence on the same rows, and verbal parse rate can collapse for a single model on a single prompt variant. Third, prompt robustness does not track parameter count reliably. Across 10 instruct models, the correlation between model size and prompt-perturbation spread ranges from -0.244 to 0.474 across benchmarks. Taken together, these results show that reliability conclusions for small language models depend not only on the model being evaluated, but also on the evaluation pipeline used to measure it. We argue that calibration definitions, evaluator logic, verbal parseability, and prompt robustness should be reported explicitly when making reliability claims.

Problem

• language model reliability
• single-prompt accuracy
• prompt robustness
• calibration
• evaluation pipeline

Innovation

• multi-variant reliability audit
• prompt robustness
• calibration
• verbal confidence
• evaluation pipeline