Overview

What is a Statistics metric?

Statistics metrics compute numeric summaries of column values — central tendency, spread, quantile boundaries, distribution shape, and aggregate totals. They provide the statistical fingerprint of a column and detect subtle data drift that isn’t visible in row counts or null rates. Type: Configurable — Statistics metrics are part of the Structural (Advanced) profile. They are enabled per column through the Configure Structural Metrics panel. Statistics metrics answer the question: What are the statistical properties of this column, and have they changed?

Metrics

Metric	Level	Description
Mean	Attribute	Arithmetic average of all non-null values
Median	Attribute	Middle value when all values are sorted — more robust than Mean for skewed distributions
Mode	Attribute	Most frequently occurring value
Sum	Attribute	Total of all numeric values — useful for financial columns where the aggregate should match an expected total
Q1	Attribute	25th percentile — the value below which 25% of observations fall
Q3	Attribute	75th percentile — the value below which 75% of observations fall
Range	Attribute	Difference between the maximum and minimum values
Standard Deviation	Attribute	Measure of how spread out values are around the mean
Variance	Attribute	Square of Standard Deviation — tracks distribution spread over time
Skewness	Attribute	Measures asymmetry of the distribution — positive skew means a long right tail; negative skew means a long left tail
Kurtosis	Attribute	Measures tail heaviness relative to a normal distribution — high kurtosis means more extreme values
The Margin of Error	Attribute	Statistical margin of error for the column’s mean estimate — indicates confidence in the Mean metric at the current sample size

How statistics monitoring works

Statistics are computed on each profiling run across all enabled numeric and date columns. Prizm baselines each statistic independently and applies anomaly detection to flag meaningful drift. Because these metrics describe the population rather than individual records, they are particularly effective at detecting upstream changes that affect many records simultaneously — loader bugs, ETL logic changes, or population shifts in source systems.

Example use cases

Financial reconciliation: The payment_amount column in a payments table should have a relatively stable Mean and Sum across daily loads. A sudden Mean spike on a Tuesday with no corresponding business event triggers an investigation that uncovers a misconfigured FX conversion multiplier applied to a subset of records. Detecting population shift: A machine learning feature column session_duration_seconds has a long-established right-skewed distribution (Skewness ≈ 2.3). After a mobile app update changes how sessions are defined, Skewness drops to 0.4. Prizm alerts on the shape change — flagging that the feature distribution has fundamentally changed before it causes model degradation.

Use Q1 and Q3 alongside Mean to distinguish between mean drift and distribution shape changes. A Mean that stays constant while Q1 and Q3 diverge indicates the distribution is widening — which Mean alone would not reveal.

Configure

Enable Statistics metrics per column, set thresholds, and configure alerts.

Get Started

Core Concepts

Platform

Architecture

Data Sources

Integrations

Security & Access Control

Deployment

Help

Settings

MCP

Release Notes

API Reference

What is a Statistics metric?

Metrics

How statistics monitoring works

Example use cases

Configure

​What is a Statistics metric?

​Metrics

​How statistics monitoring works

​Example use cases

Configure

What is a Statistics metric?

Metrics

How statistics monitoring works

Example use cases