openai/gpt-chat-latest → B (4:1)

Side B fixes a concrete security issue by sanitizing untrusted Reddit `body_html` with the `ammonia` library before passing it to `maud::PreEscaped`, adds focused tests verifying scripts and event handlers are removed, and applies the protection consistently in both rendering paths. Side A primarily removes legacy GitDiscovery projection code and makes the application Evidence-only, simplifying behavior and updating tests/schema requirements, but it largely deletes compatibility logic rather than adding a broadly protective capability.