About data quality monitors

The Monitors tab in the Job Details page allows you to analyze the health of your Data Quality Job.

Data quality monitors allow you to:

Data quality monitors are out-of-the-box or user-defined SQL queries that provide observational insights into the quality and reliability of your data. Each out-of-the-box monitor is associated with a default data quality dimension. You can also associate user-defined SQL queries with data quality dimensions. Data quality dimensions categorize data quality findings to help communicate the types of issues detected.

For a job to have monitors, it must include one or more rows of data. If a job runs with zero rows, no monitors are executed, and they do not appear on the Monitors tab. The row count out-of-the-box monitor is the first monitor evaluated during a job run. If no rows are found, the remaining monitors are bypassed because they require rows in the job to function. All out-of-the-box monitors need at least four job runs with one or more rows before they can provide suggestions on whether they are breaking or passing. Until this condition is met, the row count monitor remains in the learning state.

Overview of the Monitors tab user interface

The Monitors tab allows you to assess the health of your data quality operations at a glance. The following image and sections describe the key functional areas of the page.

screenshot of monitors tab

Run history

Run history () shows the data quality score and number of findings from a job run on a given day. You can view these details in a chart or list display.

Tip Hover your pointer over a point on the data quality score line chart to view the exact score and run date, or click the point to open the monitors of any run on its associated date.

Run metadata

The run metadata section () contains various metadata of a job run.

Run metadata shows:

Monitor details table

The monitor details table () shows the output of each monitor. It includes insightful details, such as monitor type, the data quality dimension associated with each monitor, and its status. For more information, go to the overview of monitor details table.

The following table provides an overview of the monitors shown on the Monitors tab.

Monitor type Description
Monitor name The name of the monitor.
Column name The name of the column.
Monitor type

The type of monitor that actively checks for changes in a Data Quality Job.

Out-of-the-box monitors

  • Data type: Monitors changes to the inferred data type of a column.
  • Empty range: Monitors changes to the number of empty values across all columns.
  • Execution time: Monitors changes to the time it takes for a Data Quality Job to run.
  • Max value: Monitors changes to the highest value in numeric columns.
  • Mean value: Monitors changes to the average value in numeric columns.
  • Min value: Monitors changes to the lowest value in numeric columns.
  • Null check: Monitors changes to the number of NULL values across all columns.
  • Row count: Monitors changes to the number of rows in a Data Quality Job.
  • Schema change: Monitors schema evolution changes, such as columns that are added, altered, or dropped from a Data Quality Job.
  • Uniqueness: Monitors changes to the number of distinct values in fields across all columns.

User-defined monitors

  • Custom rule: Monitors a Data Quality Job based on the criteria provided in a user-defined SQL query.
Dimensions

The data quality dimension associated with the monitor type.

While you can create custom dimensions, the following list includes the out-of-the-box data quality dimensions.

  • Accuracy: The degree to which data correctly reflects its intended values.
  • Completeness: The degree to which potential data contained in a dataset is reflected. Refers to the percentage of columns that have neither EMPTY nor NULL values.
  • Consistency: The degree to which data contains differing, contradicting, or conflicting entries.
  • Integrity: The validity of data across the relationships and ensures that all data in a database can be traced and connected to other data.
  • Validity: The degree to which data conforms to its defining constraints or conditions, which can include data type, range, or format.
  • Duplication: The degree to which data contains only one record of how an entity is identified. Refers to the cardinality of columns in your dataset.
State

When a monitor is active, the following states are possible.

  • Breaking: Indicates that there are data quality findings for the monitor.
  • Passing: Indicates that there aren't any data quality findings for the monitor.
  • Learning: Indicates that the monitor is an adaptive rule that is still evaluating your data to determine whether there are data quality findings. Adaptive rules begin in the Learning state during the first run if no prior run history exists.

When a monitor is not active, the Suppressed state is possible.

  • Suppressed: Indicates that a user has manually instructed Collibra to exclude the findings for the monitor from the Data Quality Job results. When you suppress a monitor, Collibra recalculates the scores and marks the monitor as suppressed for existing runs that occur on or after the suppression date. Suppressed runs do not count toward the overall score calculation.
Actions

Depending on the type and state of the monitor, you can:

  • Open the details drawer to manage the monitor details.
  • Pass the finding when it is breaking.
  • Suppress the monitor when it is passing, breaking, or learning.
  • Activate the monitor when it is suppressed.
  • Edit or delete rule monitors.

Important To train monitor results, you need a global role with the Product Rights > Data Quality global permission or the Data Quality Editor or Data Quality Manager resource role with the Data Quality Job > Train Monitors resource permission.

Behavior of monitors when a job runs without data

When a job runs with zero rows, the monitors react differently based on their current state (learning or active).

Learning phase threshold

The adaptive rules learning model requires a completed run with at least one row of data to begin the learning process. If a job runs with zero rows, the system does not count those runs toward the minimum threshold required to exit the Learning state. If a job has prior consecutive zero-row runs, the first run that contains data does not show an adaptive rule status. The status appears only after the system processes a subsequent run that contains data.

Row count monitor

The row count monitor shows zero rows detected.

Scoring and status

The resulting scoring and status of the monitors depend on whether the monitor is in the learning state or is already active:

State Monitor status Adaptive rule status Scoring logic
Learning

The row count monitor passes.

 Other adaptive rules show the status “exception”. The score is 100% (1 learning monitor / 1 active monitor).
Active

The row count check runs first. The monitor status is based on the expected range. If the expected range includes 0, the monitor passes. If 0 is outside the expected range, the monitor shows a breaking status.

 Other adaptive rules show the status “exception”. Monitors showing "exception" or "suppressed" are not included in the denominator for score calculation.

What's next