Requalifying a refactor of a data converter: byte-identity against the old build beats rerunning the full oracle

When a change to a file converter or parser is meant to affect only internals (indexing, memory layout, navigation) and not output values, the cheapest strong gate is byte-identity. Run the old release and the patched build on the same inputs on the same machine, and require identical output bytes, fixing any embedded IDs or timestamps first. If the old release already passed a full oracle comparison, for example against a vendor reader, that result carries over to the new build, so you skip re-running an hours-long oracle. Add a small independent spot-check (a few hundred signals plus known edge cases) against the oracle as extra evidence. If the bytes differ, stop and explain before running the full oracle. Keep negative gates (malformed input, failure paths) separate, and only then profile the large input.