Create Task
To create a task, click Add Task on the tasks page. The Create Task wizard opens.

Task details
In Fill Task Details, enter:
- Task Name
- Task Description
- Dataset to use for the task

Click Next to continue.
Select Recipe
Select the recipe for this annotation job. Click a recipe to inspect it, then Select.
The recipe type must match the dataset type.

Click Next.
Select Columns
For text (CSV) datasets, choose which columns appear in the annotation UI. The export can still include the full CSV plus annotations.
Column mapping depends on the recipe tools (for example NER, QnA, Text Matching, RLHF).

Image, medical imaging, document, and audio tasks skip this step.
Select Data Points
Choose which items to include. Use Sample for a sampling method, or select/deselect rows manually.
| Sampling method | Details |
|---|---|
| Select All | Include every data point in the dataset |
| Select Range | Consecutive range — useful for batching |
| Select Random | Random subset — useful for pilots |


You can export partially annotated data before the full task is done.
QC workflow
Choose a quality workflow: Simple, Standard, or Strict.

Simple
Fast path with no built-in review stage. You can still add QC later from the task overview.

Standard
Balanced speed and quality. Two QC stages with configurable review percentage and minimum precision. Failed precision can send items to review and rework.

Strict
Every labeled item is reviewed. Rejected items go to rework.

Annotators and reviewers
Select annotators and reviewers from project members (checkboxes). Only users already on the project appear here — add them on the project Users tab first.

Finish
Confirm the summary (task name, dataset, recipe, annotators, reviewers, QC workflow), then click Finish.
