Anomalies

An anomaly records one or more quality check failures produced during a Scan operation. It points to a potential issue in a record, data structure, or summarized set of violations and provides failed-check evidence that teams can review with their business context. An anomaly does not by itself prove that the data is incorrect or that conditions outside the configured checks and Scan scope were evaluated.

Both AI Managed and Authored checks can produce anomalies. When every failed check linked to an anomaly is AI Managed, the anomaly displays an AI badge (purple pill with a four-point star icon). Hovering it shows "Identified by AI managed checks."

Anomaly Types

Qualytics classifies anomalies into two types: Record Anomalies and Shape Anomalies. Record anomalies identify individual records that failed one or more record-level checks. Shape anomalies identify structural failures, such as a missing field or schema change. They can also summarize additional record failures when a Scan reaches its configured per-check record anomaly limit. Each anomaly reflects the active checks and scope evaluated by its Scan. Data stewards review the evidence, impact, and known exceptions before assigning a status or resolution.

Anomaly Detection Process

The anomaly detection process starts with connecting a datastore and syncing its containers and fields. Profile operations calculate observed patterns and can create or refresh most AI Managed checks. Observability measurements maintain Volumetric and Freshness checks. Data stewards add Authored checks for business expectations that require human context. A Scan evaluates the active checks in its selected scope and records failures as anomalies for investigation and resolution.