A geologist can often tell at once if a collection of information is unusual, whether it is an exception which could indicate an interesting phenomenon or is simply due to bad data. By contrast, the automated repositories we have built for our information are usually limited to simple validation checks, such as "don't accept wells greater than 15,000 ft. total depth", with irritating consequences for deep sea drilling programs. The task is well-suited to computers; the difficulty is explaining the task to the computer. More specifically, our ability to explain what information we use to reach each geological conclusion, and our ability to identify the information we use to do it, places unnecessary limitations on the ability of our databases to manage our data efficiently. The original promise of database designers, going back over twenty years, was that collecting together all available information in one place would enable new connections to be made. A measure of how far short of this goal we have fallen is that "data warehouses" are now being sold on exactly the same promise. A more realistic approach uses software agents (semi-autonomous pieces of software capable of operating independently and reporting on their findings) to help control and monitor the data storage systems we already have. The application of agents, or "data moles", to the identification, quantification and resolution of errors and inconsistencies in geological data offers a promising new approach to data quality. Data moles provide valuable assistance to the task of giving the explorationist the best possible information on which to make interpretations and judgements, and convey competitive advantage by improving the quality of our strategic information resource.