When Corrective Hints Hurt: Prompt Design in Reasoner-Guided Repair of LLM Overcaution on Entailed Negations under OWL~2~DL

TL;DR

This paper investigates how different prompt designs affect GPT-5.4's ability to correctly answer OWL~2~DL compliance queries, revealing that certain hints can worsen performance and emphasizing the importance of prompt framing.

Contribution

It provides a systematic comparison of interaction modes, showing that reasoner-verdict prompts significantly improve accuracy over generic retries and that prompt framing impacts correctness.

Findings

Direct faithfulness is 43.9%

Reasoner-verdict repair reaches 97.8% accuracy

Prompt framing can influence model correctness significantly

Abstract

We report a reproducible error pattern in GPT-5.4 on OWL~2~DL compliance queries: the model frequently answers ``unknown'' when the reasoner-entailed answer is ``no'' under \emph{FunctionalProperty} closure or class \emph{disjointness}. Using 180 reasoner-audited queries from a procedural expansion of the observed pattern plus 18 hand-authored held-out queries in two unrelated domains (insurance and clinical), we compare four interaction modes under matched query budget: single-shot, three rounds of generic ``you-are-wrong'' retry, three rounds of reasoner-verdict repair with an open-world-assumption (OWA) hint, and the same repair without the hint. Direct faithfulness is 43.9\,\% (Wilson 95\,\% CI $[36.8,51.2]$); generic retry reaches 81.7\,\% ($[75.4,86.6]$); the verdict-with-hint variant is \emph{worse} at 67.2\,\% ($[60.1,73.7]$); the verdict-only variant reaches 97.8\,\% ($[94.4,99.1]$). All pairwise comparisons remain significant under McNemar's exact test with Bonferroni correction ($\alpha = 0.01$; all $p < 10^{-5}$). The same fingerprint accounts for 4/4 errors on the held-out queries. Our interpretation is bounded: prompt framing can matter more than corrective content, and reasoner-guided wrappers should be ablated explicitly.