openai/gpt-chat-latest → B (4:1)

Side B introduces a substantial architectural improvement by removing the demo counter, adding a dedicated settlement worker that batches vote persistence and ranking recomputation, separating cached ranking reads from recomputation (`ranked_items_cached`), and updating state management to use this flow. Side A is a valuable test infrastructure bugfix—repairing OAuth mock request handling (`getRequestBody`, query parsing, null checks, redirect/state handling, exception handling) and stabilizing Playwright auth tests—but its impact is primarily confined to test reliability rather than the project's core runtime design.