Mikyo

No worries, thanks for the feedback. Keep it coming

Def. file us github issues and we can try to iron out some of the pains you're feeling

But happy to explore both areas for sure. Annotations on datasets themselves make some level of sense too, which is why we have things like splits.

I think the typical workflow we imagine is that annotations happen on traces as part of an annotation workflow. Then you want to pick and choose which ones you want to transfer to a dataset (e.g. pick the interesting ones to move to a regression suite). Otherwise I think there's no real reason to move them to a dataset as the traces probably contain a lot more rich information

I see, yeah the dataset only captures the annotations when the dataset is created. They are not linked forever since the dataset content is versioned independently from the span. I get your pain here, but also a bit of a design decision since we typically would transfer interesting examples to a dataset post annotation. I guess what's the workflow behind annotating the spans after it's been added to a dataset?

Gotcha, on point 1 I think this makes a lot of sense but we'd have to probably build "transfer session" as there's some nuanced issues around when sessions end that it doesn't make sense to transfer the session when 1 trace is transferred. On 2, that seems like a bug depending on your flow. When you transfer 1 span you don't see the annotations in the dataset metadata?

I guess more concretely what is the project you are transferring the data to called - is it interesting sessions or cold storage?

Hey Darshan T. this is correct, the transfer feature only transfers the trace as a session can span multiple traces. The original design of this feature was intended for retention of annotations. I think to make this sufficiently work consistently we'd have to build transfer session as sessions are runtime recordings and there is messiness in transferring it. Can you help me better understand why you are trying to transfer the session over to another project?

Hey Shubham A. I help run the OSS work in Phoenix Support so I'm not sure how AX works here. Phoenix now supports pulling annotations. https://arize-phoenix.readthedocs.io/projects/client/en/latest/api/spans.html#client.resources.spans.AsyncSpans.get_span_annotations

It’s hybrid. The core of the functionality will be in Arize but the glue will be a masta module