A pipeline revision changes a field type and transformation logic used by downstream consumers. The team wants to reduce regressions before replacing the production version. Which process best fits?

1. Test only that the script starts, without checking transformed output or downstream schema expectations.
2. Version the changes, run representative transformation and schema-compatibility checks, and validate a controlled deployment with a recovery plan.
3. Edit production code in place and discard the previous version immediately.
4. Change the schema and every consumer simultaneously without coordinating versions or preserving evidence.