openai/gpt-chat-latest → A (4:1)

Side A fixes concrete failures that broke end-to-end OAuth login tests by correcting mock server behavior (request body reading, redirect responses, query parsing, null-safe token parsing, exception handling) and improving test synchronization and selector handling. Side B adds a substantial UI enhancement for the vote-compare page (fullscreen layout, preview morphing, history sorting, styling, and tests), but it is primarily feature work and presentation rather than restoring core functionality that had regressed.