Consider Web Annotation

[BigBlueHat]

Really excited to see this project getting started! I've worked in similar spaces in the past (pre-LLMs). Specifically, it's great to see PROV in JSON-LD applied to this space.

Relatedly, you may also want to checkout the Web Annotation Data Model specification:
https://www.w3.org/TR/annotation-model/

It could serve as a meaningful data model for transporting and publishing the annotations (both human and bot generated) and allowing them to "interact" via more annotations (conversations, approvals, etc.).

Just an idea!
🎩

[Thru-Echoes]

Hi @BigBlueHat and thanks so much for sharing your ideas! Glad to see that the PROV in JSON-LD part of TRACE has been recognized.

I have look a bit into W3 related concepts, including just a tad of Web Annotation Data Model, but the conceptual + technical models you helped make there are great.

I do think that perhaps extending TRACE to have an optional Data Model as a serialization target. That could make it so that the 'creator' typed 'software' of the agent model would be included with TRACE's existing human / ai / system actor logging. Does that make sense?

Do you know of anyone using Web Annotation (or something similar?) as a provenance or decision logging target? Or, is PROV-O still the top choice there? I think you would know the answer to this far better than me.

[BigBlueHat]

The key thing Web Annotation would provide would be granular identification/selection. PROV-O would be a good choice for describing the actual provenance of whatever is "selected" (i.e. the target of the annotation). So they'd work best together, I'd think.