openai/gpt-5.3-chat → B (3:5)

Commit B restores broken OAuth mock infrastructure that multiple end-to-end auth tests depend on, fixing request parsing, response handling, null safety, and adding error isolation. These changes unblock entire test flows and improve reliability across the suite. Commit A improves URL correctness and adds a strong, exhaustive test for ranking logic, but its impact is more localized compared to the broad test infrastructure repair in B.