openai/gpt-chat-latest → A (4:1)

Side A adds a targeted regression test that exercises a key algorithmic property: with only 25 spanning-tree comparisons using perfect strength ratios, the ranking implementation recovers the full alphabetic order. Side B is primarily a structural refactor that extracts forum code into new modules and inlines/removes helper functions, plus adds a macOS profiling script; these improve organization and tooling but introduce little new core behavior.