Testing Without Test Data?

[PeteCresswell]

This is a new situation to me.

I'm developing a fairly complicated (to my limited grey matter, *very*
complicated) mixture of Excel spreadsheets and, for the industrial-
strength number crunching) MS Access code.

The users have something I've never thought of before: a real, honest-
to-goodness "feel" for the numbers. This is truly impressive to one
who used to be lost once a month before Quicken came around. They can
just *look* at a column of numbers and tell within a few percent if
they're within reason. And they can run a bunch of Excel functions
on the numbers to prove it.


Therein lies the rub.


Typically, I ask the client for some input data and the accompanying
answers my application should arrive at using that data.

These guys are saying "No way, Jose'".... this stuff is just too
complex, too fuzzy, and too time-consuming for us to come up with test
cases.

I'm saying, "Well, how do you know the answers I'm giving you aren't
wrong?"

They're saying, "Don't worry.... we'll know.".

These folks are going to make decisions on hundreds of millions -
maybe *billions* of dollars - based on my crummy little
numbers.....

As much respect as I have for their "feel", it seems like letting them
go production without beaucoups test cases is madness. - and I am
faced with a due diligence dilemma.

Anybody been here?

How did you handle it?

 

[Arvin Meyer [MVP]]

I always work with a copy of real data when possible. It doesn't take too
many examples to get a feel for correct queries. One thing that I've found
though. Clients don't always know when their data is correct, despite their
protestations to the contrary. Always test and retest, then check with a
calculator to make sure.