# ~Baseline Percentiles~

The detection logic supports two deviation mechanisms that can run in parallel - threshold deviation and baseline deviation. Threshold deviation uses static performance levels that measurements are compared against. Baseline deviation, on the other hand, uses dynamic performance levels that are automatically calculated by the platform and allow for the continuous comparison of current performance with historical norms.

Baselines consist of statistical information including average and maximum measurement values. Unlike thresholds, baselines can change over time and are constantly adjusted by the platform. Baseline-based detection uses statistical deviations from a baseline average in order to determine when unusual performance conditions are happening and trigger an alert.

The baseline percentile selects the percentile of the baseline that will be used to define deviation. For example, if the baseline percentile is set to 90, it means that a measurement is declared as deviating when it is at or above the 90th percentile of the baseline average. And if the baseline percentile is set to 95, it means that a measurement is declared as deviating when it is at or above the 95th percentile of the baseline average.

The detection settings provide two baseline percentile values - one for a minor deviation, and one for a major deviation. The baseline percentile for the major deviation should always be greater than the baseline percentile for the minor deviation.

Aternity also uses sophisticated baseline splitting logic to manage multiple baselines based on attributes shared between the reporting End Points.