Hi-Q is an evidence-conditioned framework for multi-hop question answering that dynamically refines queries into hierarchical trees by testing whether retrieved evidence supports each query node. A resolution operator determines when nodes are satisfied, and a dependency-preserving operator expands unresolved ones, letting the query structure emerge from corpus support rather than fixed templates. The authors report gains of +15.1 EM over iterative retrieval and +11.5 EM over graph-based RAG in full-corpus settings.