Whenever we build a fact table containing measurements of our business, we surround the fact table with everything we know to be true. We can extend this everything-we-know approach to our fact tables by including key pieces of metadata that are known to be true when an individual fact row is created. For instance, when we create a fact table row, we know the following:
What source system supplied the fact data
What version of the extract software created the row
What version of allocation logic, if any, was used to create the row
Whether a specific “Not Applicable” fact columnist unknown, impossible, corrupted, or not available yet
Whether a specific fact was altered after the initial load and, if so, why
Whether the row contains facts more than 2, 3, or 4 standard deviations from the mean or, equivalently, outside various bounds of confidence derived from some other statistical analysis