openai/gpt-chat-latest → A (9:1)

Side A makes substantive parser and rendering improvements: it introduces deterministic typed block masking, a prose item-reference tokenizer that correctly handles raw URLs, punctuation, newlines, and code fences, enforces braced DSL item bodies instead of ambiguous fenced bodies, fixes external breadcrumb root detection, and updates linkification plus extensive tests. Side B is primarily a UI/layout cleanup and CSS styling change (removing a wrapper element and improving ranking-number presentation), with little impact on core functionality or long-term correctness.