openai/gpt-chat-latest → B (3:2)

Side B performs a substantial architectural refactor by replacing the cached `GroupState` with `ScopeVotes`, deriving edge weights and connected components on demand (`edges_from_scope`, `scope_components`, `item_index`) and updating persistence, ranking, pairing, UI, and tests to use the new model. Side A fixes an important correctness bug in Rank Centrality by switching to degree-based `d_max` and adds strong regression tests, but B's redesign simplifies state management, removes duplicated cached data, and establishes a more maintainable foundation across the project.