openai/gpt-chat-latest → A (3:2)

Side A makes substantive parser and rendering improvements: it introduces typed deterministic block masking, a prose item-reference tokenizer that correctly handles raw URLs, punctuation, newlines, and code fences, enforces braced DSL item bodies, updates linkification accordingly, and adds focused tests for these behaviors. Side B adds valuable deployment infrastructure and a live audit dashboard (Docker, CI/CD, SSE status/UI, production config), but much of its impact is operational rather than improving the project's core parsing and data model, so its lasting contribution is somewhat smaller.