openai/gpt-chat-latest → A (3:2)

Side A changes application behavior by treating each unranked sibling as its own navigation group instead of aggregating all unranked items together, matching the documented grouping semantics, and it adds a regression test covering the new behavior. Side B improves test configuration by auto-discovering all test namespaces, reducing maintenance overhead, but it does not change the project's runtime behavior.