# CSV fix plan

Status: example plan, human plan gate not executed by these scripts.
Input: ticket.md, fixture.json, exporter_broken.py, test_exporter.py.

1. Reproduce the faulty exporter with the acceptance tests.
2. Keep test expectations fixed. Delegate quoting to csv.writer in exporter.py.
3. Run the tests again and retain output and source hashes.
4. Supply the relevant files and evidence to a reviewer.

Changed application file: exporter.py.
Verification: test_exporter.py and test_artifacts.py.
Risk: these checks cover the selected cell cases, not all consumers or encodings.
Decision required on real work: a human approves scope, criteria and this plan
before authorizing a sandboxed agent to build. Source edits or criteria changes
outside this plan go back for review.