The custom query documentation is being refined to help users write more accurate data extractions. Developers building queries against run data will find clearer guidance on exactly what fields are available and how to access them, avoiding common pitfalls with missing data or incorrect extraction methods.
Specifically, the guide now explicitly notes that custom run metadata—the key-value pairs set within tasks—is not accessible from the Query page. To query JSON-based columns like outputs, errors, and metric attributes, users are now instructed to use direct rather than string extraction functions.
The accompanying reflect this change, demonstrating how to filter and select deeply nested properties natively. A typical query using the updated pattern looks like this:
1SELECT output.externalId AS external_id2FROM runs3WHERE task_identifier = 'my-task'4 AND output.externalId = 'something'5ORDER BY triggered_at DESC