I just presumed splitting out the roles would make everything clearer and debugging easier. For the concept to work, the rules engine had to interact with the workbooks inputs and outputs, and that seemed like something that could have subtle mistakes and hard to debug by the consultants if everything were on a single complicated sheet.

Consultants are less familiar conceptually with functions, and this way I could get them to label variable inputs and label outputs. I expect consultants wouldn’t structure things unless given a format. The “formulas” sheet could be pretty wild and complicated: many consultants are more results focused than OCD organised. I didn’t want to have formulas etcetera mixed up with parsing the inputs and outputs. Also that way the spreadsheet can easily and unambiguously define the inputs and outputs for the engine to grab the names/definitions from.

I can think of other ways to do it, but it was just what seemed the most likely to work for multiple reasons.

One goal was to avoid recalculations by tracing the dependencies (yay spreadsheets) and using memoisation for the outputs, and versioning the results, and versioning the consultants’ formulas (i.e. the workbook was the unit of programming change, one reason to not put too much in each workbook). Auditing where values came from, and why values changed if formulas were modified, was one desire for the engine.