Validation is always a problem. In one early example, I came across, the core data on which the model did not match what was really there. The data was assumed

Validation is always a problem. In one early example, I came across, the core data on which the model did not match what was really there. The data was assumed to be correct and, of course, nobody actually did the checking. The previous year's data was used for the following year's calculations. I knew the data for the spreadsheet was wrong only because I had been involved in looking at the actual data sources in weeks previously. But there can also be formulae errors. They seem to produce the correct information, but they will have subtle errors that miss necessary edge cases or use set values when they shouldn't.