There’s no config that says “start from today” automatically when volume changes a lot, but there are two supported ways to effectively restart metrics from now: ### 1. Force a full metrics recomputation (recommended) Run your tests with:

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

This forces Elementary to recalculate metrics instead of relying on cached history, so the anomaly model is retrained on the current data pattern. [[Retraining tests](https://slack.com/archives/C02CTC89LAX/p1698154501770259); [force_metrics_backfill](https://slack.com/archives/C02CTC89LAX/p1705323587448999)] You can combine this with a shorter

training_period

(or

days_back

on older configs) so only recent data (e.g., last 7–14 days) is used going forward. [[training_period behavior](https://docs.elementary-data.com/data-tests/anomaly-detection-configuration/training-period)] ### 2. Manually clear old metrics for that table If you want to truly “start from today” for a specific table: 1. Delete that table’s rows from

elementary.data_monitoring_metrics

with a

DELETE ... WHERE

on the relevant identifiers. [[Reset metrics for a table](https://elementary-community.slack.com/archives/C02CTC89LAX/p1681289966274379)] 2. Run

dbt test

again (optionally with

force_metrics_backfill: true

). Elementary will then rebuild metrics only from the remaining (new) data. --- There is no documented per‑test YAML flag that automatically discards all history when today’s volume changes a lot; you need to use

force_metrics_backfill

and/or delete old metrics as above.