openai/gpt-chat-latest → A (3:2)

Side A fixes a correctness issue in vote/ranking presentation by changing rank gradient calculation to be per-group instead of global, aligns vote-history slider polarity with the live HUD, and adds focused unit tests covering the orientation and ranking invariants. Side B adds a useful UI capability by making pinned icons in ranked child groups clickable to unpin and verifies it with an end-to-end browser test, but it is a narrower feature enhancement rather than a broader correctness fix affecting core vote visualization and ranking behavior.