How do you regression-test a tiny COBOL change without full UAT?

One-line fix in a calculate paragraph. Change board wants evidence. Full UAT cycle is two weeks.

What is the lightest thing that still counts as due diligence in your shop?

Capture before/after on a known input file. Same JCL, compare SYSOUT and key output datasets. Attach the diff to the ticket. Not glamorous. Auditors accept it for low-risk calcs.

We keep a "golden" small file per program in a test HLQ. If you cannot hit the paragraph with that file, your golden set is incomplete.