openai/gpt-chat-latest → B (4:1)

Side B makes a targeted algorithmic improvement to pair selection by introducing component-aware bridge priorities and rank-based refinement order, then backs it with focused tests covering attachment, spanning-tree completion, and adjacent-rank selection. Side A removes a large autocomplete/parser graph in favor of a much simpler paste-and-go flow, which simplifies the codebase but also drops substantial functionality (autocomplete, incremental parsing, race handling) while replacing it with basic URL extraction and redirect behavior.