Yes, by default, Elementary includes failed data points (i.e., anomalies) in the training set for future calculations. This means that if an anomaly is detected (such as a sudden drop in row count), that data point will be used in the calculation of thresholds for subsequent anomaly detection runs. This behavior has been discussed in the community, and while it is intentional, there is ongoing debate about whether this is the best default approach. Some users have expressed concerns that including anomalous data in the training set can make future tests less sensitive to similar issues, as the anomaly becomes part of the "normal" range for the model. There are workarounds, such as using the anomaly_exclude_metrics parameter to filter out specific data points from the training set, but by default, failed data points are included in the training calculations [Exclude detection period from training period?](https://github.com/elementary-data/elementary/issues/1491) [Slack discussion on failed data points](https://slack.com/archives/C02CTC89LAX/p1695225048785209) [Slack discussion on anomaly inclusion](https://slack.com/archives/C02CTC89LAX/p1690997004224019).