Does it need to be done at each step, though? Couldn't the test data be added at just one point (where the schema is known) and just let it run through all the subsequent migrations to see if it goes boom?
I have also found this kind of approach very valuable. With defensive code, where the code itself does a bit of the kind of correctness checking that one might conventionally put into tests, it works even better. Pre/post- conditions in functions for example.
Not if you could have data that could only have been inserted at some stage to behind with and then a subsequent migration only breaks that case (eg create table in step 10 and step 11 breaks that new table - if you only had data that you inserted in step 1, it would still pass all migrations. In other words, you need to have sample data that exercises the entire DDL that is added or you risk missing something.
