openai/gpt-chat-latest → A (3:2)

Side A introduces a substantial architectural improvement by moving vote persistence and ranking recomputation into a dedicated settlement worker, adding cached ranking reads (`ranked_items_cached`) so HTTP handlers can use read locks instead of recomputing under write locks. It also removes an obsolete demo counter and simplifies the UI around the real ranking workflow. Side B adds useful DSL parsing and linkification improvements (typed block masking, prose item reference tokenization, braced body enforcement, and URL linkification), but these are more feature-focused and narrower in long-term impact than the concurrency and performance redesign in Side A.