openai/gpt-5.3-chat → A (3:2)

A changes production code to use `display_path()` in vote URLs instead of full storage URLs, aligning links with user-visible paths, and adds a comprehensive test that exercises all 45 pairwise votes and validates final ranking correctness. B primarily fixes test infrastructure (query parsing, request body reading, null handling, and mock server robustness) to restore OAuth E2E tests, which is valuable but less impactful than A’s user-facing behavior change plus stronger correctness guarantees.