openai/gpt-chat-latest → A (10:1)

Side A fundamentally redesigns ranking storage by replacing cached `GroupState` edge/index data with a minimal `ScopeVotes` model and rebuilding edges, item indexes, connected components, and rankings on demand. This touches ranking, persistence, reducers, UI, and tests to preserve behavior while simplifying state and eliminating duplicated cached data, whereas Side B mainly consolidates duplicated integration-test helpers into shared utilities (`test.common`/`test.oauth`) without changing core project behavior.