> For the complete documentation index, see [llms.txt](https://docs.collibra.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.collibra.com/data-quality-self-hosted/core-concepts/findings/data-quality-monitors/schema.md). # Schema The Schema monitor detects changes to columns and data types and down-scores any observations. {% hint style="info" %} Collibra DQ requires a minimum of 4 previous runs to establish a baseline calculation of a dataset's table properties before it can detect any schema changes. {% endhint %}

Column	Description
Observation	The type of schema anomaly detected.
Action	Lets you label and train a finding. The available dropdown menu options are Validate, Invalidate, and Resolve. Validate instructs Collibra DQ to either assign a finding to a specific user for review, which then appears in the View the Assignment Queue or acknowledge without an assignee that the finding is a valid observation. Invalidate instructs Collibra DQ to ignore a finding and allow the value to pass. There are two invalidation options: Save lets you mark a finding as invalidated. Save & Retrain lets you invalidate a finding and any previously saved invalidated findings, if any. When you have many findings to invalidate, it may be best to use the Save option to invalidate them at the same time, once all findings are reviewed. Resolve Instructs Collibra DQ to mark the finding as an observation and prevents it from appearing in future runs. Resolving a finding does not immediately affect data quality scores.
Status	The status of your data quality item, for example, Observation.
Profile	The user account that is assigned to this schema finding. When the Status is Assigned, a user profile displays in this column. When a schema finding is unassigned, the profile column is empty.

### Exporting schema records Click **Export** above the drill-in table to generate an Excel file with the details from the drill-in. ## Retrain Updates the DQ models used for automated data-quality checks. The Retrain process consumes current training data (and optionally labels/config), builds a new model or rule-set, validates it, and publishes a new version that will be used to score incoming data and compute DQ metrics. Use **Retrain** when data distributions change, new labels/ground truth are available, business rules change, or on a scheduled cadence to prevent model drift.