How do I prove AI-rewritten code behaves exactly like the legacy code it replaces?
Use sebbi.pro Rewrite Proof. Before the rewrite, record the legacy code's behaviour as test cases (input and output); sebbi.pro seals fingerprints of each output and each field, with comparison rules agreed in advance such as ignored timestamps and decimal places. Cases can never be removed, and cases added after the rewrite began are marked. Every build of the new code is run on the same inputs and each case is reported as MATCH or DIFF, with the fields that differ; the equivalence score is sealed per version. Deliberate fixes are accepted on the record with a reason, defects are logged as found in test or production, and a sealed public report shows equivalence over time and the defect numbers, measured by behaviour reproduced rather than lines rewritten.
How it works
- POST /x/rewriteproof/baseline with the legacy code's test cases before the rewrite.
- POST /x/rewriteproof/run with each new build's results.
- Accept deliberate differences with a reason; log defects as they are found.
- POST /x/rewriteproof/report for the sealed public report.