openai/gpt-5.2-chat → B (1:4)

Side B fixes vote highlighting and polarity end-to-end: it corrects slider-to-ratio mapping (`slider_value_from_ratios`), introduces `winner_side`/`winner_text`, updates the history UI to use a consistent disabled slider, adjusts CSS gradients for correct visual emphasis, and adds multiple tests (including a ranking invariant). Side A merely removes a zero-ratio guard in `apply_vote` and updates a test comment/expectations, a small cleanup with limited impact compared to B’s cross-layer correctness fix.