Data structures and meaning
Embedding, vector and model lineage
Lineage for derived artifacts: embeddings, features and models — what they were built from and when.
ddn@1New DDN source
Why use it?
A model you cannot trace to its training data is a liability, not an asset.
When it helps
Governing ML artifacts, features and their inputs.
How to read it
- Same rules as data lineage, plus version and owner on every artifact.
- A feature with no recorded source is an assumption with a GPU bill.
What you need
- The source datasets
- Feature/model artifacts
- Versions and owners
Things to check
- Do not let "the pipeline" be a single anonymous box.
- Do not version the model without versioning its inputs.
Scope of this diagram
Declared lineage; no evaluation of the models themselves (inference is out of scope by decision).
Work with the converted example
- Open the viewer and select
diagram. Inspect the rendered view and its source. - Open the designer. Keep the selected profile and inspect its legal palette and property controls.
- Change one displayed title or an appropriate data value. Apply the edit and check the resulting diagram.
- Run the verifier and review its diagnostics before moving complete content into review.
- Download your edited DDN source to retain the change.
The example uses the indentation-based ddn-next-1 dialect. The language guide explains reusable definitions, dotted data rows and explicit property blocks.