We introduce Scope-then-Traverse, a novel framework for multi-hop question answering that resolves the fundamental tension between evidence breadth and reasoning depth. The framework decouples the QA process into two synergistic stages. The first stage, Knowledge Scoping, generates a complete logical blueprint by decomposing the complex query into independent sub-questions and dependent templates. Subsequently, it performs parallel retrieval using only the high-certainty, low-ambiguity independent sub-questions to build a high-recall evidence pool. This strategy ensures comprehensive evidence coverage while avoiding premature commitment to a single, potentially flawed reasoning path. The second stage, Path-Aware Traversal, constructs a coherent reasoning path within this pre-scoped pool, guided by the blueprint from Stage 1. At each step, a fused scoring mechanism selects the next evidence node by jointly optimizing for local relevance and global path cohesion. Experiments on HotpotQA, 2WikiMultiHopQA, and MuSiQue demonstrate that our approach achieves strong performance by grounding answers in explicit and verifiable reasoning paths.