I disagree that you can't reach 100% accuracy of the known data. If the bottom line of the data warehouse doesn't match the inputing systems, then the data warehouse creator needs to be able to explain why. Every possible permutation can't be checked, but a good sampling can be, and should be.
We had a consultant who built us a warehouse and said that the data shouldn't be expected to match completely. After some investigation, however, it was discovered that he had violated the "level of granularity" rule which is crucial. In this case the data didn't match because of a flaw in the data warehouse.
Data warehouse designers should consider stored procedure groups to self-document the steps. They can only be created and edited in the analyzer (at least in 2K and earlier) but hopefully if someone is building a dw using the analyzer is not a big issue.