It seems likely that filtering traces by a specific feedback label (e.g., `feedback.label == "thumbs-up"`) may not work because the correct way to filter by annotation label is to use the evaluation name as an indexer on the special keyword `evals`, like so: `evals['feedback'].label == 'thumbs-up'`. The `feedback` column itself is an aggregation and not directly filterable in this way in the query DSL or UI filter bar.
For example, to filter for spans or traces with a "thumbs-up" feedback annotation, use:
```
query = SpanQuery().where("evals['feedback'].label == 'thumbs-up'")
```
This approach is supported in both the Python API and the Phoenix UI filter bar, as long as the evaluation/annotation is named "feedback" and the label is "thumbs-up" as per your configuration. Would you like more detail or examples for your specific setup?
Sources:
- <https://arize.com/docs/phoenix/tracing/how-to-tracing/extract-data-from-spans#how-to-explode-attributes|How to filter by evaluation results>
