openai/gpt-chat-latest → A (9:2)

Side A introduces a substantial architectural shift from flat scope-based rankings to a hierarchical `ItemId`/`GlobalTree` model, adds canonical URL parsing, breadcrumb navigation, node persistence, and a journal worker while updating UI and state handling to support tree-structured rankings. Side B mainly removes legacy compatibility paths and requires evidence-only records in the ledger/UI, simplifying code and tests but primarily deleting fallback behavior rather than adding new core capabilities.