How to prove new software matches the original spreadsheet
A cleaner interface does not make a replacement system correct. When a quote, engineering result, assessment, or report depends on an established workbook, the new calculation needs evidence that it matches the original across the cases the business actually encounters.
01
Define what parity means
List the outputs that must match, the permitted numeric tolerance, the rounding stage, the unit conventions, and any outputs that are expected to change. Without that agreement, a comparison can reveal differences without telling the team whether they are defects.
Exact equality is appropriate for discrete decisions and many commercial calculations. Floating-point engineering calculations may need a documented absolute or relative tolerance.
02
Build a representative case set
Historical cases are useful because they include combinations people actually selected. They should be supplemented with boundary cases: minimums, maximums, values just above and below a threshold, missing optional inputs, and known exceptions.
Keep cases that exposed past defects. They become regression tests that protect the logic every time the application changes.
03
Compare intermediate values when outputs differ
A final-price mismatch only says something diverged. Capturing important intermediate values narrows the cause to an input transformation, lookup, branch, rounding point, or final aggregation.
This is also where undocumented workbook behavior becomes visible. A cell may rely on coercion, an implicit default, or the order in which a macro updates values. The goal is to understand that behavior before deciding whether to reproduce or correct it.
04
Keep parity tests after launch
The comparison suite should become part of the application's automated tests. When rates change or a new option is added, the system can prove that unrelated historical cases remain unchanged.
This turns the original spreadsheet from a fragile production tool into a documented reference model—and turns confidence in the new software into something repeatable rather than anecdotal.