Evaluation rules are the scoring logic that data quality scans run against your data. You build a rule once, pick the quality dimension it measures, define what it looks at, and set the logic that turns a value into a result and a score. Once saved, you attach the rule to any object in a Monitoring Suite by mapping its inputs to that object's fields, so one rule can be reused across many datasets.
Overview
Every evaluation rule lives in a central manager showing its name, its active or inactive status, and the dimension it belongs to. From here you can create a rule, edit one, clone a rule as a starting point for a similar check, or toggle it on or off without deleting it. A rule only affects data once it's attached to an object inside a Monitoring Suite — building a rule and activating it are separate from actually applying it.
Quality dimensions you can measure
When you create a rule, you first choose the dimension it evaluates. Each represents a different way of judging trustworthiness:
Validity checks whether values follow the correct format, content rules, and relationships between attributes.
Uniqueness flags duplicate records and confirms a value appears only once where required.
Completeness checks whether required fields hold values instead of being empty or null.
Accuracy compares your data against a trusted reference source to confirm correctness.
Timeliness assesses whether data was available when needed and still reflects the current state.
Choosing Uniqueness also lets you pick a rule category. A standard rule validates individual records one at a time; an aggregation rule evaluates data across a whole group of records instead, which suits checks like "no two records in this group should share a value."
Defining inputs
Before writing rule logic, you define what it can reference. Attributes are the data fields the rule checks, each typed as text, whole number, decimal number, true/false, or date, and later mapped to real fields when the rule is attached to an object. Parameters are configurable values, such as a threshold, supplied when you attach the rule rather than when you build it. Variables derive a new value from an attribute by applying a chain of transformations — such as uppercasing text, trimming whitespace, or rounding a number — in an order you can drag to rearrange. Aggregation rules use groups, named sets of attributes, in place of parameters. An input already referenced in your rule logic is protected: deleting it warns you to remove it from the logic first.
Rule sets and how scores are computed
Rule logic is an ordered list of rule sets. Each pairs a condition over your inputs with an outcome: a result appropriate to the dimension, plus a numeric score. Results depend on the dimension — a validity rule resolves to Valid or Invalid, a completeness rule to Complete or Not Complete, a timeliness rule to Timeliness Ok, Minor Delay, or Major Delay.
Rule sets evaluate top to bottom and stop at the first match, so their order sets priority. If nothing matches, a fallback result you configure is returned instead. Rule logic stays locked until you've chosen a dimension and created at least one input — one group, for aggregation rules.
Testing a rule
As you build a rule, a preview panel mirrors your inputs and logic. From there you can open the tester, add test case rows with sample values for each attribute and parameter, and run the test to see the result each case produces. Use this to confirm edge cases land in the rule set you expect before relying on the rule in production. Saving a rule activates it immediately, so use the manager toggle afterward if you want to deactivate it without deleting it.
Notes
Saving a rule activates it right away — deactivate it from the manager toggle if you're not ready to use it yet.
Cloning a rule is a fast way to reuse an existing dimension, inputs, and logic as a starting point.
An input referenced anywhere in your rule logic can't be deleted until you remove that reference.
Aggregation rules only appear under the Uniqueness dimension and use groups instead of parameters.