Data quality assessment provides a comprehensive health check for training datasets. The assessment engine evaluates data across six dimensions: completeness, consistency, accuracy, timeliness, uniqueness, and relevance. Each dimension produces a normalized score along with detailed diagnostics.
Completeness analysis identifies missing values, truncated records, and partially populated fields. For text data, it detects incomplete sentences, missing paragraphs, and broken encoding that could corrupt training signals.
Consistency checks validate that labels match content, that formatting follows declared schemas, and that cross-references between records are valid. Inconsistencies are flagged with severity levels and suggested corrections.
The assessment report includes actionable recommendations prioritized by expected impact on model performance. Historical trend tracking shows how dataset quality evolves across versions, helping teams measure the effectiveness of their curation improvements.