We have introduced the concept of a SNAT (Source Navigation Annotated Target tree) as a document which specifies the transformation of source data sets of a particular kind into instances of a particular document model. A SNAT document is a location tree obtained from the target document model and augmented by metadata. The names, semantics and scope of the metadata items are defined by a metadata model (see for example table "Metadata items SNAT"). Relying on such a metadata model, a SNAT-based code generator transforms a SNAT document into executable code implementing primitive operations and composing them into the transformation as a whole.

Defined in such general terms, the concept of a SNAT-based code generator becomes an abstract model which may guide the design of concrete code generators, distinguished by a particular metadata model and expecting source data with a particular media type. To illustrate this possibility, we introduce a SNAT-based code generator for constructing XML documents from RDF data.