The service runs outlier-detection algorithms on time-series metrics and alerts you when outliers are found.
Supported:
-PostgreSQL
-Email & Slack notifications
-Detection Methods: Threshold, Deviation from the Mean, Interquartile Range
The zh.wikipedia demo uses a simple method: it averages the previous 30 days and checks whether the current value is within ±10%.
As for the zh data itself, starting from late March 2025 it changes irregularly compared to other languages and even to its own 2024 pattern. That usually points to data issues — maybe bots or crawlers — though there could also be another cause.
What happened to the zh version that is full of outliers?