openai/gpt-chat-latest → B (3:2)

Side B adds substantial tooling improvements with lasting developer value: it introduces `compile --ingest` to replay and compile a single ingest from an event log, refactors replay logic, makes `scan` dramatically faster by using a parse-only pass instead of full replay, and surfaces detailed `parse_error` information with updated CLI output and tests. Side A improves the vote-compare UI with a fullscreen layout, preview morphing, reordered vote history, and related CSS/tests, but these are primarily presentation and workflow enhancements rather than foundational tooling and diagnostics.