Semantic Metrics

Semantic is one of the six metric contexts in Prizm (see Metrics). A semantic metric is a quality rule that is tied to a business concept in the glossary, a Term, rather than to a single hand-picked column. You write the rule once against the term, and Prizm evaluates it on the attributes that carry that term, so the same business definition is checked the same way wherever it appears.

​

Where to find them

Select Metric in the top navigation, then select the Semantic tile in the row of context tiles. The tile shows how many semantic metrics exist. The list below it switches to semantic metrics only, with sub-tabs for the metric subcategories in use (for example Query and Conditional), each with a count. Semantic metrics are marked in the list as follows: As with other metrics, Select Columns in the More menu lets you show more columns, such as Term, Term ID, Domain, Product, Application, Tag, Owner, Weightage, Target Score, and Metric Priority.

​

Create a semantic metric

1. Go to Metric and select the Semantic tile.
2. Select the + icon above the list to open Add Metric.
3. Enter a Metric Name. This is required.
4. Choose a Subcategory for the rule type. The list offers Conditional, Query, Behavioral, Comparison, Lookup, and Standalone Query.
5. Optionally describe the Purpose of the metric.
6. Select Create, then configure the rule on the metric’s detail page.

​

The metric detail page

Selecting Edit on a row opens the metric’s detail page. The header shows the metric name, its overall score and status, the path to the asset it runs on (source, schema or pipeline, table, and attribute), and summary counts: Alerts, Issues, Views, Weightage, Total, Valid, and Invalid records, plus the owner. The page has five tabs:

​

Overview tab

The Overview tab contains these sections, from top to bottom. Metric Detail. Shows the Asset the metric runs on and a Downstream Impact count of the assets affected if this metric fails. Description, Purpose, and Rationale tabs let you record what the check does and why it exists. Prizm AI can suggest a description. Metric configuration. The heading names the metric’s rule type and the glossary term it is tied to. Under Select criteria you pick the columns the rule uses, and under Condition rules you build the rule. Rules can be combined with AND, OR, and NOT, and you can add single rules or groups of rules. A Conditional metric offers these rule types: Select Validate to check the rule before you save it. A validation badge shows whether the rule is valid. For a Query metric, you write a SQL query instead of building condition rules. See Conditional metrics and Query metrics for the full authoring reference. Scoring. Set the Weightage (as a percentage), the Pass Criteria, and whether the metric is Exported. Weightage controls how much the metric contributes to the score of the asset it runs on. Threshold. Choose when a result is unusual enough to raise an alert: For Auto thresholds, the page shows four bands, based on how far a result is from its expected value: Low (0σ to 1σ), Medium (1σ to 2σ), High (2σ to 3σ), and Critical (3σ or more). Each band displays the current expected range and a short explanation of what to do. Semantic context. Tag the metric so alerts reach the right team and it appears in search. You can set Domain, Application, Product, Tag, and Dimension (for example Consistency). See Semantics. Owners. Choose who is accountable for the metric and who fixes it when it fails. You can assign a Custom owner and a Business Owner. Audience. Choose the people or groups who receive alerts and summaries for this metric.

​

Row actions

Each row in the list has these actions on the right: Edit, Copy (clone the metric), Schedule, Run job now, and Delete. Use Schedule to run the metric on a schedule, or Run job now to run it immediately. See Schedules and Jobs.

​

Related pages