Hello Dharmin!
Anomaly detection has a training period and a detection period, both can be configured to your needs.
It seems that the detection period in this case is around 24H, and within that timeframe there are values that are out of range. Any outlier within the detection period would cause the test to fail.
The training period and the detection period overlap, so you can see that the range keeps growing as the results have higher values for a few hours.