Open Data Quality Requires Looking Inside the File

A dataset can be published, licensed, machine-readable, and highly rated by a portal, yet still fail real-world reuse. Portal and metadata scores indicate access and discovery, not the quality of the data itself.^[1][2]

Start with the reuse case and legal gate: verify a licence, avoid terms that prohibit commercial exploitation, and check whether the data is available in standard formats through APIs, web services, or downloads.^[3]

Then inspect the schema against a domain standard. Check accurate feature names, required fields, semantic meaning, and completeness. Test actual values for consistent types, and measure missingness from nulls and empty strings.^[4]

Do not stop at structure. Profile values for inconsistencies and anomalies, including unusual observations or possible sensor faults. The evidence supports content-level checks, but not a single universal anomaly score or dedicated anomaly dimension.^[5]