A development account describes converting a synthetic 20-sheet analytical workbook created by Claude Opus 5.5 into Python. Grouping 78,766 formulas by relative references yielded 469 distinct pieces of logic, and extraction errors were corrected until tests matched 63,309 cells calculated by Excel. Runtime fell from 3.4 seconds to 0.6 in the author’s environment. However, the original workbook and intended defects were generated by the same AI, and matching existing values does not prove correct business logic. Preserving originals, checking input extraction, and independently verifying calculations are central to migration.