Tag
This paper studies temporal failure modes in LLM-based statutory question answering, including post-cutoff staleness and recency bias. It introduces a benchmark of 312 expert-validated German statutory QA pairs and evaluates LLMs under various inference settings.