Yes, it is possible to force a "restart" or retraining of a test in Elementary if you want it to disregard previous periods and use only the latest data. You can do this by running your dbt command with the variable

force_metrics_backfill

set to

true

. For example:

dbt test --vars '{"force_metrics_backfill": true}'

This will ensure that the metrics are recalculated and the test is effectively "restarted" for the next run, rather than relying on the previous training period's data. This is particularly useful if the underlying data has changed significantly (e.g., the table was dropped and recreated), and you want the test to reflect the new state rather than historical failures or patterns. By default, Elementary optimizes for performance and does not always retrain on the full period unless this flag is set, so using

force_metrics_backfill

is the recommended approach in such scenarios [Slack thread on retraining tests](https://slack.com/archives/C02CTC89LAX/p1698154501770259).

I'm sorry, but the knowledge sources do not mention any built-in functionality in Elementary Data to automatically increase the detection_period or to re-check a failed period until it passes. The detection_period parameter is static and must be configured manually in your test configuration. There is no documented feature that dynamically adjusts the detection window based on test outcomes or automatically re-tests failed periods in subsequent runs. If you need to re-check or retrain tests after failures, you can use the force_metrics_backfill option to recalculate metrics, but this does not dynamically adjust the detection period or automate the process you described. Any such workflow would need to be managed externally, for example, by adjusting your test configurations or orchestration logic outside of Elementary Data itself [Slack thread on retraining tests](https://slack.com/archives/C02CTC89LAX/p1698154501770259). If you have a specific use case, you may want to suggest it as a feature request to the Elementary team.