Datasets

Inspect imported and generated data before using it in analysis.

A dataset is a table of data in Dawn. It may come from a file upload, an external connection, or the output of a pipeline. Pipelines, models, dashboards, and reports can use it as an input.

Open and inspect a dataset

Find the dataset in Workspace, using search or Filter assets where helpful, and open it. Check the preview and available information about its columns and source.

Before using a dataset, establish:

• What one row represents.
• Which column or combination of columns identifies a row.
• Which period the data covers and when it was refreshed.
• Which values are missing or use special codes.
• Whether amounts include tax, discounts, or refunds.

See loading data to create a dataset and data engineering to prepare one for repeated use.

Explain the columns

Use the dataset’s data dictionary to record column meaning, units, valid values, and important exceptions. Review automatically generated descriptions before relying on them: column names and samples do not establish the business meaning on their own.

For example, describe amount as “order-line amount in AUD, excluding GST, before refunds” when that is the actual definition. A description such as “the amount” adds little useful information.

Use Context → Data for relationships between business concepts and datasets, and Catalogue for shared calculations. See context.

Analyse the data

Attach the dataset to a conversation through Add context or image. Start by asking Dawn to inspect its structure, then request the analysis you need.

Explain the structure of this dataset. Identify its likely row key, date
coverage, and columns that need business definitions. Check duplicate keys
and missing values before proposing an analysis.

Verify the proposed row key and definitions yourself or with the dataset owner.

Change or refresh data deliberately

A new import or pipeline run can change the data used downstream. Check the operation’s replace, append, or update behaviour before running it. Preserve the original source when developing cleaning rules and write prepared data to a separate output where practical.

If you use the API to ingest or update records, follow the current API reference for the endpoint’s authentication, schema, and limits. Do not assume a dataset is immutable simply because it was first created by a pipeline.

Retire a dataset

Check pipelines, dashboards, reports, models, and scheduled work that depend on it before deletion. Renaming or restoring under a new name can affect SQL references. Follow workspace cleanup and Bin recovery.