openai/gpt-chat-latest → B (9:1)

Side B adds a substantial end-to-end browser test that seeds a pool, exercises the vote→next-pair workflow, and verifies key invariants such as votes appearing in edge history and all presented pairs belonging to the requested pool. Side A only deletes the JVM test runner entrypoint (`test/runner.clj`), removing functionality without introducing a replacement in the shown patch.