How Datadory Validates Every Dataset in Its Catalog

Every number on this page comes from the same audit pass over Datadory's master catalog, computed programmatically on 2026-08-22 and stated here as of August 2026. The unit of accounting is always the full denominator: 1,744 datasets drawn from 1,124 unique sources across 163 cataloged industries. Nothing on this page is sampled or extrapolated — each row below is a direct count over all 1,744 records, so any figure can be reproduced from the catalog itself.

The rubric runs on a 10-point scale, and the catalog-wide average lands at 7.81 out of 10. Scores skew high: 1,096 of 1,744 datasets (62.8%) score 8 or higher, and 145 records (8.3%) sit at the ceiling of 10. A low score does not remove a dataset from the catalog — 30 records (1.7%) score 4 or below, mostly in thin industries, and they are listed with their score attached so buyers can price the risk themselves. Across 24 industry briefs, government-named top datasets average 8.81 versus 8.47 for the rest, and 16 of the 22 ceiling-scored top datasets named in those briefs are government or public-science sources such as openFDA, NCBI GenBank, PubChem and ClinicalTrials.gov.

Geographic scope is bucketed from each record's coverage text using keyword normalization, and the result is nearly split down the middle: 597 of 1,744 datasets (34.2%) center on United States coverage versus 586 (33.6%) with global or multi-country scope, followed by Europe/EU at 125 (7.2%), the United Kingdom at 74 (4.2%), India at 32, Canada at 27, China at 11 and Australia at 8. Provenance skews institutional: 781 datasets (44.8%) are public_dataset type, and the catalog's 1,744 records compress into just 1,124 unique sources. Reach concentrates further — 703 datasets (40.3%) come from 149 sources serving more than one industry, led by the U.S. Census Bureau at 37 datasets across 29 industries, Eurostat at 32 across 28, Hugging Face at 27 across 24 and Kaggle at 25 across 22.

All figures on this page were computed programmatically from the 1,744-record master catalog on 2026-08-22; the page reads as of August 2026. Scope is the entire catalog — 1,744 datasets, 163 cataloged industries, 1,124 unique sources — with no sampling and no exclusion of failed or low-scoring records. Percentages are taken against the full 1,744 denominator unless a narrower base is stated, and denominators are never rounded away. Geographic buckets apply keyword normalization to the free-text coverage.geographic field, so mixed-scope records land in other/mixed rather than being forced into a single country. Format counts draw on a raw by_format field containing 1,061 distinct labels — json (api), json-stat, json-ld and sdmx-json variants among them — which pipelines must normalize before comparing; the CSV (734) and JSON (675) counts above follow that normalization. Quality uses the integer-valued 10-point rubric, with observed scores running 4 through 10 and the mean reported to two decimals (7.81).