Measuring LLM Trust Allocation Across Conflicting Software Artifacts

Researchers developed TRACE, a framework to evaluate how LLMs allocate trust between conflicting software artifacts like code, documentation, and tests. The study found that current LLMs are better at identifying natural-language specification issues than detecting subtle code-level problems, with models showing systematic blind spots when implementations drift while documentation remains plausible.

• →LLMs show asymmetric sensitivity to different artifact types, with documentation bugs creating larger quality gaps than implementation faults.
• →Models detect explicit documentation bugs well (67-94%) but struggle when only implementation drifts while documentation stays plausible.
• →Six of seven tested models showed poorly calibrated confidence levels in their trust allocation decisions.
• →Current LLMs are more effective at auditing natural-language specifications than detecting subtle code-level drift.
• →The research suggests explicit artifact-level trust reasoning is needed before using LLMs for correctness-critical applications.