Build an Analysis Dataset

For: Analytics Publishers

Use this chapter to create the structure that Readers and Writers will use. Your connection must include lab.analytics.write, and you need a Publisher role for the selected Analytics Project.

Workflow

1. Define the questions the dataset must support.
2. Create a table and define its columns.
3. Choose supported data types.
4. Add initial data from a CSV file when needed.
5. Review columns, imported rows, and validation results.
6. Create a reusable read-only view when appropriate.
7. Verify the dataset before sharing it.

Describe the dataset before creating it

Tell the AI agent:

• the question or workflow the dataset should support;
• the meaning of one row;
• the required columns and their expected values;
• which columns uniquely identify a record; and
• whether a reusable read-only view is needed.

For example:

In the Respiratory Cohort Analytics Project, propose a dataset for one laboratory measurement per sample. Include a sample identifier, measurement name, numeric result, unit, and measurement date. Show the proposed schema, table, columns, data types, required fields, and primary key before creating anything.

Review names, data types, nullability, column order, and primary keys before authorizing a structural change. After creation, ask the AI agent to list the table and its columns so you can verify the result.

Note

A Publisher can create structures. Loading rows requires Writer access, and querying data requires Reader or Writer access. One person may therefore need more than one Analytics Project role.

This manual will use product language such as dataset, table, column, and view. Database implementation details will remain in engineering documentation.