openai/gpt-chat-latest → A (4:1)

Side A changes the rank-history thread index from a silent fallback (`unwrap_or(0)` after converting to 1-based) to an enforced invariant with `expect`, aligns the value with the project's 0-based `/t/tag/N` URLs, updates documentation, always renders the link, and adds integration tests verifying the behavior. Side B is a localized UI improvement that groups each unranked sibling separately and adds a corresponding unit test, but it affects presentation rather than correcting a broader API/data consistency issue.