What are entities
Entities define the data structure for an item: what experts will SEE (input entities) and what they will CREATE (output entities).
Input entities: Extract data from your dataset—this is what experts see and work with.
Output entities: Collect annotations from experts—this is what experts create and what you’ll receive back.
Entities also provide format validation to ensure data consistency.
Passthrough Data
In addition to entities, your dataset may contain passthrough data—fields that pass through the system unchanged (metadata, IDs, timestamps). These are not entities (experts don’t see or modify them), but they are included in your results to maintain data traceability.
Examples: item IDs, timestamps, original metadata, conversation IDs.
These fields are automatically included in results without expert interaction.
What to check
✅ All input entities match fields in your dataset (what experts need to see).
✅ All output entities capture what you need from experts (what you’ll receive).
✅ Entity types are appropriate (e.g., ratings need numbers, not text).
✅ Passthrough fields (if any) are correctly identified and will be included in results.
✅ The structure supports your annotation goal.
Data path
You can Edit settings via UI
Each entity shows its mapping to your dataset fields (using JSONPath).
Input entities map to fields in your dataset (where to extract data from).
Output entities map to fields where annotations will be saved (where to store results).
The agent automatically sets up these mappings based on your dataset structure, you don’t need to do it manually.
If something is wrong
Tell the agent what needs to be changed.
The agent will update the entity structure.
Review again in the Entities panel before proceeding.