Testing data-intensive systems is paramount to increase our reliance on e-governance services. An incorrectly computed \emph{tax} can have catastrophic consequences in terms of public image. Testers at Norwegian Customs and Excise reveal that faults occur from interactions between database features such as \emph{field values}. Taxation rules, for example, are triggered due to an interaction between 10,000 items, 88 country groups, and 934 tax codes. There are about 12.9 trillion 3-wise interactions. Finding interactions to uncover specific faults is like finding a needle in a haystack. Can we surgically generate a test database for interactions that interest testers? We address this question with a methodology and tool {\tool} to automatically populate a test database that covers all T-wise interactions for \emph{selected features}. {\tool} generates a constraint model of interactions in {\Alloy} and solves it using a divide-and-combine strategy. Our experiments demonstrate scalability of our methodology and we project its industrial applications.