This part of the description is unclear w.r.t the sequence diagram. How is this "deduplication" implemented?

nitpick: we usually use dashes as filename separator for doc files, please favor that over the underscore here

We should clarify which kind of metadata we are talking about here. In the past with @moranegg we have agreed on the following terminology:

• intrinsic metadata: those shipped as part of the source code artifacts that we ingest into the archive
• extrinsic metadata: those available out-of-band w.r.t. the above scenario (e.g., available on the forge / distribution platform, but not distributed as source code artifacts)

You should consider starting this document documenting this distinction.

Failing that (e.g., because we want to document the distinction properly elsewhere), you should at least stick to the terminology of intrinsic metadata in the documentation, because it's the workflow about them that you are documenting, not the other one.

In your sequence diagram, it looks strange that you try to retrieve existing metadata from the "Graph Storage", but you upload newly created metadata (in the alt box) only to the "Indexer Storage"

Indeed, thanks.

pretty sure what the thanks means but hey ;)

Explicitely content_metadata_get is a call from the indexer_storage api.

So:

IDX_REV_META->>IDX_STORAGE