The learned 'normal' for a table or metric — the range against which anomalies are judged.
Reviewed by Francisco Ferreira ·
A baseline is what normal looks like for a given table, column or metric: the typical row count for a Tuesday, the usual null rate of a field, the expected range of a daily revenue number. Anomaly detection is only as good as its baseline — without one, every value is just a number with no context.
Good baselines account for rhythm: weekdays differ from weekends, month-end differs from mid-month. A naive fixed threshold like 'alert if rows are under 1000' fires constantly; a learned baseline knows that Sunday is supposed to be quiet and stays silent.
Tabkeel builds a baseline per table automatically from your history, then keeps it current as legitimate patterns evolve — so alerts mean something.