I work for a pharmaceutical company, and we need to validate all spreadsheets used to calculate results from raw data. I am reviewing the execution of such a protocol, and the design seems bass-ackwards to me.
Instead of generating an array of well-characterized data, and having the executor populate the spreadsheet with it to compare the actual results to the expected results, the author of the protocol (who is on vacation) has populated the protocol with results from the spreadsheet, and the executor is verifying the results with a calculator - the one built into Windows.
Does Excel use the same engine for arithmetic as the Windows calculator? Are they, in fact, proving anything other than “the FPU and Windows are not time-dependent”?
Excel does not use the same code for arithmetic as the windows calculator. Under normal conditions, both use the same Floating Point Hardware.
What are you trying to verify? The floating point hardware? If so, you could compare the values calculated using AMD hardware with the values calculated using Intel hardware. Do this by saving a copy of the spreadsheet as numbers, then recalculating the spreadsheet on different hardware.
Are you trying to verify Excel? If so, save a copy of the spreadsheet as numbers, then recalculate the original spreadsheet in Open Office.
Are you trying to verify Windows? If so, save a copy of the spreadsheet as numbers, then recalculate the original spreadsheet using Excel on a Mac.
Are you trying to verify the calculations used in the spreadsheet? If so, recalculate the values methodically, using any calculator (including Windows).
There have been errors in the hardware, in Excel, and in Windows (which provides rounding for some products, but not Excel or Windows Calculator last I knew), so those suggestions are not totally insane, but the problems and solutions have all received wide publicity. Except in exeptional circumstances, the hardware, Excel, and Windows are effectively verified by the large user base.
Far more common, and a far larger problem, are errors in the way users write the calculations in spreadsheets. These errors are tested for by a process like the one you describe, and Windows Calculator is valid for the test.
When I started, spreadsheets like Excel were used to find logic and calculation errors in manual spreadsheets, which were and are far more prone to errors. Later on, I was involved in a business replacing complex business functions from Excel, which often had logic and calculation errors – less often than the manual processes which they replaced, but often enough that we learned to expect our systematic process to get different results.
Because of my background as a programmer, I would use a programming approach to verifing your spreadsheets, but using a calculator is common when you need traceability rather than speed and volume.