Data Quality

Be Told a Number Is Wrong Before a Stakeholder Is

Checks run against a published Data Mart at three levels: the mart itself (empty table, duplicate rows, primary-key uniqueness), the field (null rate, type mismatch, freshness, constant column, negative values) and the relationship (join values that exist on both sides). On demand, or on a schedule. Every finding carries its violation count, its example values and the SQL that produced it, so you can read why a check says what it says instead of taking a red badge on trust.

Freshness is separate, needs no configuration and costs no credits: every mart shows when its source tables were last written to, on the mart’s page, on the model canvas and in MCP responses. It reports writes, not coverage, so a backfill that rewrites a table today with 2021 figures reads as written today, and the product says so rather than letting you assume otherwise.

Features

Data Quality Features

Data Quality Checks

F01

Data Quality Checks

Finance asks the Data Owner why the dashboard's order count differs from theirs, and the cause can be duplicate orders or orders whose customer is missing from the joined Data Mart, which look exactly like correct data. Data Quality Checks test the Data Mart, its fields and its joins, on demand or on a schedule.

  • Duplicate rows and join values with no match in the target show up as failed checks, each with its violation count, up to three example values and the SQL that found it.
  • You choose the checks and each one's severity, Error, Warning or Notice, and Run History keeps the configuration every run used.
  • A preset of basic checks is on from the start, and a Data Quality Run trigger reruns them on your schedule.
Data Freshness Tracking

F02

Data Freshness Tracking

A founder asks whether the AI assistant's figure for yesterday's sales is current, and the Data Owner is the one who has to answer: a stale figure looks as convincing as a fresh one. OWOX shows when the source tables behind each Data Mart were last written to, with nothing to configure.

  • Hover the value for the exact timestamp and each source table's own time; SQL and views are traced to their base tables.
  • It reports when tables were written, not what period they cover, and shows Unknown or ≥ rather than guess when a source can't be read.
  • Each query an AI assistant runs through OWOX MCP returns the date with the rows; the same value sits on the Data Mart page, in the list and on every canvas node.

Recent updates