Variables
Variables define the annotation schema for a project. Each variable represents one dimension that annotators (human or LLM) respond to.
Creating a Variable
Navigate to Variables in the project sidebar and click New variable. Choose a name and a type, then configure the type-specific settings.
Variable names must be unique within a project. Names are used as column identifiers in templates and exports.
Variable Types
Single Categorical
Annotators choose exactly one option from a predefined list.
Configuration:
- Options — a list of labels (e.g. positive, neutral, negative). Each option has an internal ID (generated automatically) and a display label.
Example: Sentiment classification with options positive / neutral / negative.
Multi-Categorical
Annotators can choose one or more options from a predefined list.
Configuration:
- Options — same structure as single categorical.
Example: Topic tagging where an article can belong to multiple categories.
Likert Scale
Annotators select a position on a numbered scale.
Configuration:
- Scale — number of points (e.g. 5 for a 1–5 scale)
- Labels — optional text label for each scale point
Example: Agreement level from Strongly disagree (1) to Strongly agree (5).
Text
Annotators type a free-form text response.
Configuration:
- Max length — optional character limit
Example: Asking annotators to rewrite a sentence or explain their reasoning.
Integer
Annotators enter a whole number.
Configuration:
- Min / Max — optional inclusive bounds
Example: Counting the number of named entities in a document.
Float
Annotators enter a decimal number.
Configuration:
- Min / Max — optional inclusive bounds
Example: A confidence score between 0 and 1.
Editing Variables
Click the edit icon next to any variable to rename it or change its configuration. Changing the options of a categorical variable does not retroactively affect already submitted annotations — existing task values reference the internal option ID.
Using Variables
Once defined, variables can be used in:
- Annotation forms — add an Input component block linked to the variable
- LLM pipeline prompts — each prompt targets one variable
- CSV/XLSX templates — appear as
annotation:{name}andllm:{name}columns - Exports — annotation columns are labelled
{variable name} ({annotator name})