<p>Mimir adds support for uncertain data to Apache Spark by using lightweight annotations called "Caveats" that are propagated through queries. Caveated data and records can carry additional metadata (e.g., textual descriptions of uncertainty, or details about the scope of the uncertainty), but this information is not propagated during normal query evaluation. By only propagating the presence of a caveat in a query result, the performance impact on query evaluation is minimal, and the content of the resulting caveats can be subsequently derived through a form of program slicing applied to queries.</p>