no narrate self 😠
Summary: Abridge the diff, folding bulk package definitions and per-test bodies into anchored ellipsis rows while preserving behavior, error handling, and refactor structure.
ideally we'd make or find an actual test/benchmark/something harness that lets us run abridging repeatedly and concurrently with options to change one or many variables like inputs/outputs/models/etc. not sure how we'd grade them tho, to determine whether prompt or model changes improve or worsen summary/abridging results. requires some thonks.