openai/gpt-chat-latest → A (9:1)

Side A adds a new workspace tool (`sorterc`) plus a reusable `server::offline` module that compiles `.sorter` files, replays reducer state, scans `events.jsonl` for corrupt or unreplayable ingests, exposes a CLI, and includes documentation and tests. Side B adds a single regression test validating the ranking algorithm on a random spanning-tree case, which improves confidence but does not expand project functionality or architecture like the offline tooling does.