openai/gpt-5.3-chat → A (3:2)

A introduces a new async settlement pipeline (mpsc worker, batching, and cached score computation) and shifts ranking reads to a cached, read-only path, improving correctness and performance under concurrency; it also removes unused demo-counter code. B adds richer DSL parsing/tokenization and linkification rules, but these are more feature-level enhancements rather than core architectural improvements.