# Explore datasets

> Use the Dataset Explorer to verify variables, fix labels, and confirm group structure before you tabulate. Three views — Variables, Question Groups, Raw Data — share the same source of truth.

*Source:* https://recense.ai/docs/explore-datasets

## Variables view

A table of every variable in the dataset showing:

- Variable name (original from the file)
- Label (human-readable description)
- Type badge — colour-coded: C (Categorical), B (Binary), S (Scale/Numeric), T (Text), W (Weight)
- Group membership

> **ℹ Tip**  
> Use this view to scan the full variable list, search for specific fields, and spot type issues.

## Question Groups view

Variables grouped by their parent question. Useful for:

- Understanding array structures (e.g. a grid question with multiple items)
- Verifying that related variables are grouped correctly
- Identifying groups that need manual adjustment

## Raw Data view

A paginated record-level grid for spot-checking actual values.

- Filter columns to focus on specific variables
- Click a variable in the Variables tab to scroll to it in Raw Data
- Use this to verify suspicious categories, check missing-value patterns, or confirm that a weight variable looks reasonable

## Editing metadata

Variable names, labels, and value labels are editable directly in the explorer. Changes propagate to:

- Table headers and row stubs
- AI agent context (the agent sees updated labels)
- Exported project files

> **ℹ Clean metadata drives better AI**  
> Good AI output starts with clean metadata. If labels are unclear or grouping is wrong, fix it here first. The agent reads the same metadata you see in the explorer.

## Group-synced editing

For categorical array questions (grids), editing a value label in one member variable updates it across all members of the group. This prevents inconsistency in array-structured data.

## Next steps

- **[Stack question groups](/docs/stacking)** — Reshape grid questions when you want to analyse items as a single variable.
- **[Build tables and analysis](/docs/tables-and-analysis)** — Build cross-tabulations from your verified variables.
- **[Code open-ended responses](/docs/text-coding)** — Turn open-ended text answers into a categorical variable.
- **[Write dataset instructions](/docs/dataset-instructions)** — Give the agent persistent context about the survey.