: Information integration requires manipulating data and metadata in ways that in general go beyond a single existing transformation formalism. As a result, a complete source-to-target mapping can only be expressed by combining different techniques like query languages, wrappers, scripting, etc., which are often speci c to a single integration platform or vendor. Such a mapping is not portable across different alternative deployment scenarios, thus limiting the mapping's reusability and putting the considerable investment required to create it at risk. To avoid this vendor lock-in, we de ne an integrated representation for operations on arbitrary data and metadata that is independent of any speci c metamodel or transformation language. , we can express mappings in an abstract, vendor-neutral form, improving the interoperability of integration tools and the exibility for the deployment of mappings.