# Default-omission model fidelity experiment

Read [the report](../../text-default-omission-fidelity-20260913.md).
Both gates retain raw streams, source manifests/ZIPs, start/result receipts and
the original runner. The independent static review identifies pre-format files;
the executed input captures identify final formatted source. No semantic edits
followed review. Run `python -B verify.py` here for read-only archive integrity,
exact source membership/hashes and recorded Cargo outcome checks. The runner's
producer paths need adaptation for new captures; use the report's Cargo commands.
Binary hashes are historical execution receipts; binaries are not included.
This experiment proves bounded model fidelity, not production editor correctness,
multi-file formatting, source8, GUI/API, Linux or release acceptance.