Query clarification is essential for retrieval‑augmented generation (RAG) systems when user queries are ambiguous. We study multi‑turn grounded conversations where a query may be ill‑formed in the dialogue context or, more intriguingly, well‑formed but leads to multiple relevant interpretations after document retrieval. The goal is to detect both situations and produce an appropriate clarification request.\ \ We introduce ClariEdge, a dataset of about 3,900 examples that contains ambiguous queries together with negative cases that do not require clarification. A lightweight LoRA‑fine‑tuned model reaches 96.6% classification accuracy and outperforms prompted baselines of larger models in both classification and clarification generation quality. Hard negatives prove crucial for separating truly ambiguous queries from those that appear ambiguous but can be resolved from context. The same LoRA performs well in pre‑retrieval and post‑retrieval stages, making it a viable checkpoint before answer generation in a RAG pipeline. The ClariEdge dataset is available at https://huggingface.co/datasets/ibm-research/ClariEdge.\ \ Review