iiRDS Rendition: the file behind an information unit

Example

A packaging machine manufacturer publishes the topic "Changing the sealing jaw." The authoring system exports it three times: as the DITA-based XML source, as HTML for the service portal and as a PDF for printing. In the iiRDS package all three files sit in content subfolders, and the one topic carries three renditions — application/xml, text/html and application/pdf — each with its relative iirds:source path. Product variant, lifecycle phase and required qualification are stated once, on the topic. The portal displays the HTML rendition, offers the PDF for download and ignores the XML, which is kept for downstream reuse.

How it applies

Technical documentation

Renditions separate "what the content is about" from "how it is delivered." Editorial teams maintain metadata once and add formats as channels require. Because every path is relative to the package root and content lives in subfolders below it, never in the root or META-INF, a consumer can resolve files the same way on any system that unpacks the ZIP archive.

Migration

A legacy PDF manual can enter iiRDS as a Document with a single PDF rendition, long before it is rewritten as topics. Later, a RangeSelector on that same rendition can expose individual chapters. From DITA, the published HTML and the XML source can both be delivered as renditions of one unit.

MedTech

Where approved instructions for use exist as a fixed PDF, that file can serve as the rendition while metadata identifies device and revision. For archiving, iiRDS/A limits renditions to long-term formats such as PDF/A-3 or iiRDS XHTML5; iiRDS/H requires at least one PDF/A-2 or PDF/A-3 rendition per document. These are format rules for the package; they do not replace the regulatory review of the content.

AI and retrieval

Ingestion pipelines read META-INF/metadata.rdf, follow each rendition's source path and extract text. The danger is indexing the extracted text without the unit's metadata. Store the unit IRI with every chunk, so product, audience and lifecycle phase travel with the text into the index. When a unit has several renditions, index one text-bearing format and treat the others as alternatives.

Rendition vs. InformationUnit vs. Selector

Concept	Role	Carries metadata?
InformationUnit	Meaning: what the content is, for whom, for which product	Yes
Rendition	File: path in the package and format	No
Selector	Pointer to part of the rendition's file	No

Media follow the same logic. A video is not a special class: a self-contained clip is modeled as a Topic with a video/mpeg rendition, and a media file that is not self-contained must be modeled as a Fragment.

In RDF

A topic with an XML source rendition and published HTML and PDF renditions. Shown in Turtle for readability — inside a package, metadata.rdf is written in RDF/XML.

@prefix iirds: <http://iirds.tekom.de/iirds#> .

<https://example.com/iirds/change-sealing-jaw> a iirds:Topic ;
    iirds:has-rendition [
        a iirds:Rendition ;
        iirds:format "application/xml" ;
        iirds:source "content/source/change-sealing-jaw.xml"
    ] , [
        a iirds:Rendition ;
        iirds:format "text/html" ;
        iirds:source "content/html/change-sealing-jaw.html"
    ] , [
        a iirds:Rendition ;
        iirds:format "application/pdf" ;
        iirds:source "content/pdf/change-sealing-jaw.pdf"
    ] .

A rendition can also be given its own IRI instead of a blank node, which lets several units — for example one Document per language of a multilingual PDF — reference the same file.

In topic-based authoring and modular architectures

In topic-based authoring, the source topic and its published outputs are different files of the same module. iiRDS models exactly that: one unit, several renditions. A DITA map decides which topics are published; the iiRDS generator then records, for each published unit, which output files represent it. In modular delivery architectures, renditions are what the consumer actually renders or streams, so the choice between them becomes a channel decision — HTML for a tablet, PDF for print, video for training — made at runtime from metadata. Keep renditions free of metadata and keep units free of file assumptions, and the same package can feed new channels without re-authoring.