Comparison
Distribution checkvsVolume anomaly
Distribution check
the average order value did not change but the shape did, because a whole segment of small orders stopped arriving.
Comparing a column's distribution — percentiles, category frequencies, cardinality — against its recent history. It catches the class of problem that passes every structural test: right types, right row count, wrong data. It is the most likely check to produce false alarms, so it works best as a warning that a human reviews rather than as something that blocks a pipeline.
Full entry →Volume anomaly
the load succeeded and inserted eleven rows, where every previous day inserted about four hundred thousand.
A row count far outside the range that period normally produces. It is the highest-value check per line of code in existence, because almost every ingestion failure — a filter matching nothing, a partial extract, a source outage — shows up here first. It needs day-of-week and seasonality awareness to be usable, otherwise every Sunday is an incident and the alert gets muted.
Full entry →