Thus why you should deploy any solution to a test environment first, to try to catch those unknowns. In a normal line of work, an ETL gathering data from an outside source should have (I say should though it probably does not happen much in the real world) a document describing in technical detail the data rows, columns and formats that you should expect. It is like a service agreement. An ETL may fail but if you can point to the document and state the reason was non-adherence by the outside data, then the reason out of your scope and you are in the clear.