Skip to main content

​
What is a Freshness metric?

Freshness metrics measure how recently data was loaded or updated in an asset. They are the earliest warning system for pipeline delays and stale datasets — detecting problems before downstream consumers notice them. Type: Automatic — Freshness metrics are enabled on all connected assets by default as part of the Essential observability profile. No configuration is required to activate them. Freshness metrics answer the question: Is the data current enough to be trusted? Prizm tracks the time elapsed since the last data update on each operational job run. The current staleness value is compared against a statistical baseline of normal refresh cadences — or against a user-configured time threshold — and alerts fire when data has gone too long without being refreshed.

​
Metrics

​
How Freshness monitoring works

Prizm reads the last-modified timestamp of the asset on each operational job run — using platform-native signals such as INFORMATION_SCHEMA timestamps, DESCRIBE HISTORY results, or watermark column values depending on the connected platform and asset type. The time elapsed between the current evaluation and the last update is the staleness value — expressed in milliseconds, seconds, minutes, or hours. This value is trended over time and evaluated against either a statistical baseline (Automated Threshold) or a hard time limit (Custom Threshold).

​
Example use cases

SLA enforcement for a reporting table: A finance dashboard reads from daily_revenue_summary, populated by a 6 AM batch job. Setting a freshness threshold of 2 hours means Prizm alerts by 8 AM if the job did not complete — before the finance team opens their dashboards. Catching a silent pipeline failure: An Airflow task fails without raising an error in the UI but stops loading data. The Last Updated timestamp stops advancing. Prizm raises a freshness alert on the next monitoring cycle — a failure that would otherwise go undetected until users notice stale numbers. Cross-platform freshness monitoring: A Snowflake materialized view is refreshed from an upstream Databricks pipeline. Prizm computes freshness from the view’s LAST_COMPLETED_REFRESH_TIME and alerts when the refresh window is missed — even when the downstream view itself shows no errors.
Set the freshness threshold to match the pipeline interval, not the maximum acceptable delay. A threshold equal to the pipeline cadence alerts on the first missed run. A threshold 4× the cadence lets three runs fail silently before anyone is notified.

Configure

Set time-based thresholds, view the metric detail page, and understand Freshness alerts.