openai/gpt-chat-latest → A (3:2)

Side A makes several durable security and correctness improvements: voting now fails closed instead of silently falling back to an anonymous actor, auth cookies gain the Secure flag when appropriate, OAuth mocking is gated behind an environment flag, open-redirect sanitization is strengthened with tests, and the durable dependency is pinned by revision for reproducible builds. Side B fixes a real multi-user bug by removing shared browser delegate sentinels and updating pending-session handling and tests, but its scope is narrower than the broader authentication hardening and security improvements in Side A.