openai/gpt-chat-latest → A (5:1)

Side A fixes a real UI correctness issue by changing rank row styling to be computed per ranking group instead of globally, removes obsolete offset logic, and adds regression tests for the gradient behavior. It also aligns vote history visualization with slider semantics by introducing consistent winner/slider mapping, updating the UI, and adding multiple polarity tests, whereas Side B is a small but isolated fix that simply defines `GITHUB_API_BASE_URL` with a default to prevent a missing-variable failure.