I have done this type of testing, and I only know of two approaches. One is to keep a special database for testing. The other is to continuously fix the tests as the database changes.

Depending on the nature of your tests, you many need to do some of each. To test new data, keep the code constant. To test new code, keep the database constant.

The trick is to test one dimension at a time. Don't test new code and new data at the same time.

If it is impractical to keep a test database, you may be able to separate the code changes from the data changes using statistics. I use R for this type of statistics.

It should work perfectly the first time! - toma