anomaly_sensitivity

is a configuration that tells Elementary how many standard deviations (z‑scores) away from “normal” a value must be to be treated as an anomaly. [[anomaly_sensitivity](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/anomaly-sensitivity)] Under the hood: 1. Elementary builds a training set of historical metric values (row counts, sums, null counts, etc.). 2. It calculates: - the average of that training set - the standard deviation of that training set [[Anomaly method](https://docs.elementary-data.com/data-tests/data-anomaly-detection)] 3. For each new value, it computes a z‑score: \[ z = \frac{\text{current value} - \text{training average}}{\text{training stddev}} \] 4. It then compares the absolute z‑score to `anomaly_sensitivity`: - if

abs(z) > anomaly_sensitivity

→ flag as anomaly - otherwise → not an anomaly [[Anomaly method](https://docs.elementary-data.com/data-tests/data-anomaly-detection); [anomaly_sensitivity](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/anomaly-sensitivity)] So with:

yaml anomaly_sensitivity: 2

you are telling Elementary: “mark a point as anomalous if it is more than 2 standard deviations away from the historical average.” That is more sensitive than the default of 3: the expected range is narrower, so more values will be flagged as anomalies. [[anomaly_sensitivity](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/anomaly-sensitivity)]