Based on the knowledge sources, there isn't a direct way to exclude weekends from being tested or to skip them in the detection period. However, there are a couple of approaches you can use to handle data that has different patterns on weekends: ## Use Seasonality Configuration The recommended approach is to use the

seasonality: day_of_week

configuration. This tells Elementary to compare each day only against the same day of the week from previous periods (e.g., Saturdays compared only to previous Saturdays, Sundays to Sundays, etc.). According to the [seasonality documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/seasonality), this is particularly useful for datasets that have lower volume over weekends:

yaml models:   - name: this_is_a_model     tests:       - elementary.volume_anomalies:           seasonality: day_of_week           time_bucket:             period: day             count: 1

With this configuration, if you typically have 0 rows on weekends, Elementary will compare Saturday's 0 rows against previous Saturdays' 0 rows, so it won't flag this as an anomaly. ## Alternative: Use where_expression If you want to completely exclude weekends from the test, you could use the

where_expression

parameter to filter out weekend data entirely. As mentioned in a [Slack discussion](https://slack.com/archives/C02CTC89LAX/p1686640037495849), you can add a where clause to exclude specific dates:

yaml - elementary.volume_anomalies:     where_expression: "DAYOFWEEK(timestamp_column) NOT IN (1, 7)"  # Adjust syntax for your warehouse

However, this would exclude weekends from both the detection and training periods, which may not be exactly what you want. Note: There is currently no built-in feature to mark specific days as "success" or to skip them entirely from the detection period while keeping them in the training data.

You're absolutely right, and I apologize for the confusion in my previous response. The

where_expression

approach wouldn't work for your use case because: 1. The

where_expression

filters the data being tested, but Elementary still evaluates the time buckets within the detection period 2. If you filter out weekends with

where_expression

, those weekend buckets would have 0 rows 3. With

fail_on_zero: true

, the test would fail on those weekend buckets Based on the knowledge sources, there is currently no built-in way to completely exclude specific days (like weekends) from the detection period while still using

fail_on_zero: true

for weekdays. ## The Current Limitation According to a [Slack discussion](https://slack.com/archives/C02CTC89LAX/p1698683615614929), Elementary needs the data of full time buckets to determine anomalies, and by default it only calculates anomalies on "full buckets." The platform doesn't currently support skipping specific days within the detection period. ## Your Best Option The

seasonality: day_of_week

configuration remains your best approach, but it won't completely "skip" weekends - it will just compare weekends to previous weekends. If you consistently have no data on weekends, this should prevent false positives on those days. However, this doesn't address your specific need to have

fail_on_zero: true

work only on weekdays while ignoring weekends entirely. The knowledge sources do not contain a solution for this specific requirement. You might need to consider running separate tests for weekdays and weekends, or potentially opening a feature request with the Elementary team for day-of-week specific configurations.

You're correct that

fail_on_zero

controls what happens when the metric value is 0 within the detection period. According to the [fail_on_zero documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/fail_on_zero), "Elementary anomaly detection tests will fail if there is a zero metric value within the detection period." ## Behavior When There's No Training Data Based on the knowledge sources, when there isn't enough training data, Elementary has specific default behavior: According to a [Slack discussion](https://slack.com/archives/C02CTC89LAX/p1690997004224019): > "If there are less than 14 days in the table the tests won't fail because this is still a 'training period' for the anomaly detection." And from another [Slack conversation](https://slack.com/archives/C02CTC89LAX/p1692965164612639), a support staff member explained: > "it's supposed to pass and show in the UI that there were not enough data points to reach a decision." ## Controlling the Training Period You can control the minimum amount of training data required using the

min_training_set_size

parameter. As mentioned in a [GitHub issue](https://github.com/elementary-data/elementary/issues/1118), you can configure:

yaml config:   elementary:     min_training_set_size: 14  # minimum number of data points required

However, the knowledge sources do not explicitly mention a way to make the test fail when there is insufficient training data - by default, it passes when there isn't enough data to make a determination. The test behavior when lacking training data appears to be designed to avoid false positives during initial setup or data gaps. If you need the test to fail when there's no training data available, this doesn't appear to be a supported configuration option based on the available documentation.

Based on the knowledge sources, I need to clarify the behavior you're asking about. When you use

where_expression

to filter out weekends, you're filtering the data that Elementary queries, but Elementary still evaluates the time buckets within the detection period. This is a crucial distinction. ## What Actually Happens According to a [Slack discussion about fail_on_zero](https://slack.com/archives/C02CTC89LAX/p1701274524204489), when investigating why

fail_on_zero: true

wasn't working as expected, a support staff member asked: > "When you're running the test without fail_on_zero do you see rows in data_monitoring_metrics?" This suggests that Elementary creates metric entries for time buckets, and

fail_on_zero

evaluates those metric values. ## The Expected Behavior Based on the [fail_on_zero documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/fail_on_zero): > "Elementary anomaly detection tests will fail if there is a zero metric value within the detection period." If you filter out weekends with

where_expression

, those weekend time buckets would have 0 rows in the filtered dataset. With

fail_on_zero: true

, the test should fail on those weekend buckets because the metric value is 0. However, I must note that the knowledge sources do not explicitly document the exact interaction between

where_expression

and

fail_on_zero

in this specific scenario. The behavior I've described is based on the general documentation of how these features work, but there may be nuances in the actual implementation that aren't covered in the available documentation. To get a definitive answer about this specific interaction, you might need to test it directly or reach out to the Elementary team.

Based on the knowledge sources,

anomaly_exclude_metrics

won't solve your specific problem of excluding weekends from the detection period while keeping

fail_on_zero: true

for weekdays. ## What

anomaly_exclude_metrics

Actually Does According to the [

anomaly_exclude_metrics

documentation](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/anomaly-exclude-metrics): > "By default, data points are compared to all data points in the training set. Using this param, you can exclude metrics from the training set, to improve the test accuracy." The key limitation is that

anomaly_exclude_metrics

excludes data points from the training set, not from the detection period. ## Why It Won't Help Your Use Case As explained in a [Slack discussion](https://slack.com/archives/C02CTC89LAX/p1740069258974599), when someone tried to use

anomaly_exclude_metrics

to exclude the detection date: > "I was looking at

anomaly_exclude_metrics

but when I configure this it filters out the date for both the detection and training period." And in a [GitHub issue about excluding detection periods](https://github.com/elementary-data/elementary/issues/1491), a user noted that while you can use

anomaly_exclude_metrics

to exclude the latest data point from training:

sql anomaly_exclude_metrics: "MAX(bucket_end) OVER (...) = bucket_end"

This approach still doesn't prevent Elementary from evaluating those time buckets in the detection period - it only excludes them from training calculations. ## The Bottom Line

anomaly_exclude_metrics

is designed to exclude problematic data points (like incidents or outliers) from influencing your training data and expected ranges, but it doesn't skip time buckets in the detection period. Elementary will still evaluate weekend buckets, and with

fail_on_zero: true

, those buckets would still fail if they have 0 rows. The knowledge sources confirm that there's currently no built-in way to completely exclude specific days from being tested in the detection period.