FAQ
Data and privacy#
Does my data leave Databricks? No. DataPulse runs inside your account. The AI model is a Databricks Model Serving endpoint in your workspace. Notifications go only to your own Databricks notification destinations.
What does the AI model see? Column names, types, summary statistics, query text, and error text. While Share sample values with the model is on (the default), it also sees up to 5 real, unmasked values from each column with 20 or fewer distinct values; an admin can turn this off in Setup → Models. It never sees whole rows or columns with many distinct values. Text from your data or errors is passed as data, not as instructions.
Whose permissions does DataPulse use? Yours, when it reads or writes your data. The Unity Catalog rules, row filters, and column masks that apply to you apply in the app. DataPulse's own history (actions, incidents, budgets) is written by the app itself so users cannot edit it.
Where is DataPulse's own data stored? In Delta tables in datapulse_app.state in your Unity Catalog. There is no separate database.
Numbers#
What does "Mock Data" mean? The panel is showing sample data because your real system table could not be read. It is not your bill. Ask your admin to check the grants.
Why is Verified Savings zero? An action counts as verified only after the observation window (DATAPULSE_VERIFY_DAYS, default 7 days) has passed after it executed and a person has pressed Verify on it in the Actions inbox. The window lets both the before and after billing periods close fully. AI estimate is a guess and never counts as verified savings.
Is the AI estimate a saving? No. AI estimate is a guess and never counts as verified savings.
Why are other people's emails shortened? To protect privacy. Only admins see full emails.
How old is the data? Databricks system tables can lag by hours. Panels are cached for one to five minutes. Use Refresh Overview or reload the page.
Actions#
Can I undo an action? Most of them. Executed actions show Rollback in the Actions inbox, and rollback restores the previous setting. Two cannot be undone: Retry update and Full refresh on a declarative pipeline (the confirm dialog says so). A Repair Run can be cancelled while it runs, but tasks it already re-ran stay re-run. See Pipelines.
Why is the Approve button grey? Your role is too low for that risk level, or the action is not in the right state. Hover the button to read why.
Who can approve a change? Operators (for the modules they may change), cleansing engineers (Data Cleansing only) and admins for LOW RISK. Admins only for HIGH RISK. See Roles.
Can DataPulse change things without asking? Only if an admin set a budget to auto_idle mode. Even then it acts only on idle resources and only for low-risk actions. See Budgets and guardrail modes.
Data cleaning#
Will cleaning change my source table? Never. It writes a new table (by default <catalog>.silver.<table>_cleaned, set in Setup → Data conventions).
What if the AI writes something dangerous? A safety gate blocks DROP, DELETE, UPDATE, ALTER, GRANT, and stacked statements. You get DESTRUCTIVE_SQL and nothing runs.
Why did my request fail after I clicked Run as background job? A check ran first to confirm you can read the source and write to the target. If you cannot, you get PRECHECK_DENIED and no job starts. Ask for USE SCHEMA and CREATE TABLE on the target schema.
Cost of DataPulse itself#
What does DataPulse cost to run? It uses a small serverless SQL warehouse, small jobs, and model calls when you press AI buttons. Job schedules are installed paused. See Cost control.
Why did the first page load take a minute? The SQL warehouse was stopped and started on your first query.