Narrative generation algorithms are often based on generic rules that are applied for all instances of a given resource. However, in many cases it is useful to have narrative whose appearance is specific to a narrower subset of resources. This extension allows defining a narrative generation algorithm that applies to all resources that are instantiations of a particular definition (Questionnaire, ObservationDefinition, PlanDefinition, etc.) This allows the organization of the narrative to be tied to the specific elements and structure associated with the definition.

This extension points to a Library instance. The Library instance should have a type of FHIRPath Liquid and a content.contentType of "text/html". The Library.content.data will be a base64 encoded Liquid template that uses the FHIRPath Liquid variant. This allows the use of FHIRPath expressions as part of the Liquid logic. When this logic is executed with `%resource` context being set to the resource instantiated from the definition using this extension, it will generate HTML narrative based on the data present in the resource instance.

While the Liquid can be hand-coded, in most cases it will be generated algorithmically based on core elements, and potentially custom extensions, found in the definition resource. However, once the Liquid is generated, it will be generic and can be executed by any engine that supports FHIRPath Liquid