Interpreting Built-in Metrics¶
Note
To understand the concept of Data Quality, you can read a full description in the Concept Overview page here.
Overview¶
Each Data Quality Check is performed by a Data Quality Application and reports its result as a Metric.
Each Check reports a different type of result, so the same number means different things depending on which Check produced it.
Use this page to understand what the values shown on the Data Quality Dashboard are telling you about your data.
Note
Some Data Quality Applications will output regular reports, like Edge Data Availability, and others will only output reports when a problem is detected, like Outlier Detection.
Note
These are the five Checks that come built in with the Kelvin Platform, each performed by a Data Quality Application built by Kelvin.
Developers can also build their own custom Data Quality Applications, which run on the same infrastructure and report their own Metrics. Check with your developer for details of any custom Metrics on your Kelvin Platform.
Data Availability¶
Ensures the expected number of messages are received in a given time window.
The result is the number of messages actually received divided by the number of messages expected in that window, expressed as a percentage.
The expected number is set with the Expected # of Messages and Interval parameters in the Data Quality Configuration.
| Value | Meaning |
|---|---|
| Below 100% | Fewer messages arrived than expected. Data is missing, delayed, or the Connection is offline. |
| 100% | Exactly the expected number of messages arrived. |
| Above 100% | More messages arrived than expected. The Connection is producing data faster than the configured rate. |
Warning
A value above 100% is not a bonus. It is a mismatch between the number of messages you declared and what the Connection is actually sending.
For example, if the configuration expects 1 data point per minute and the Connection sends 4 data points per minute, the metric reports 400%, which is 300% more data than expected.
This matters because models and Applications are often built around a specific frequency or sampling rate, so receiving data far more often than expected can distort their calculations.
To correct it, either change the Connection polling frequency, or update the Expected # of Messages and Interval parameters so they match the real data rate.
Outlier Detection¶
Uses statistical methods to detect outliers over a moving window.
The result is a count of the statistical outliers detected.
Range Validation¶
Validates whether values fall within an expected range.
The result is a count of the out-of-range values detected.
Stale Detection¶
Detects when a Data Stream value has not changed over a configurable time window.
The result is 1 when a stale value is detected and 0 when normal.
Timestamp Anomalies¶
Detects anomalies or irregularities in the timestamp sequence.
The result is a count of the timestamp anomalies detected.