Glossary
Data Quality
The degree to which training or evaluation data is accurate, consistent, relevant, and representative of the desired behavior.
Data quality is one of the most important — and most underestimated — factors in model behavior. A model trained on inconsistent, ambiguous, or biased examples will internalize those inconsistencies, and no amount of prompt engineering can fully compensate. Quality issues can range from outright errors (a label applied incorrectly) to subtler problems like annotation guidelines that different raters interpret differently. For behavior architects, maintaining data quality means investing in clear guidelines, calibration sessions with annotators, and regular audits of what’s actually in your training data. The phrase “garbage in, garbage out” is old but still accurate.