openai/gpt-chat-latest → B (3:2)

Side B corrects and standardizes the thread post index semantics across RPC, HTML, types, and integration tests: it switches rank history to the intended 0-based `/t/tag/N` indexing, removes the silent `unwrap_or(0)` fallback in favor of asserting the expected invariant with `expect(...)`, and updates rendering and documentation accordingly. Side A is a worthwhile cleanup that removes a redundant zero-ratio guard and updates tests to reflect that zero-weight edges are skipped elsewhere, but it is primarily dead-code removal relying on existing validation.