openai/gpt-chat-latest → A (9:1)

Side A makes a targeted functional improvement by changing rank-row coloring from list position to per-group score normalization, adds a dedicated `score_gradient_t` helper, updates rendering to use min/max scores, and includes focused regression tests for edge cases like tied scores. Side B is a massive seed commit containing speculative designs and many new files, but it also includes obvious accidental terminal transcript dumps in source files (`forms.rs`, `ranking.rs`), making much of the patch noisy and of questionable lasting value compared with A's precise, test-backed bug fix.