openai/gpt-5.2-chat → B (1:5)

Side B introduces a substantial architectural change: removing the demo counter feature across the stack, adding a new asynchronous settlement worker with batching, introducing cached ranking computation, refactoring state management to use a SettlementClient, adjusting locking strategy (write → read where possible), updating ranking APIs, and modifying multiple integration tests. It adds a new module and significantly changes core request handling and persistence flow. In contrast, Side A focuses on fixing and hardening OAuth test mocks and related E2E test utilities, which, while valuable, are limited to test infrastructure and bug fixes. The scope and impact of Side B are considerably larger.