Data quality · the corpus, measured

How good is the catalog?

A per-topology data-quality report across the datasets we have probed - a measured slice of the larger catalog, not the whole of it. Every probed dataset is graded on ten quality criteria - completeness, validity, consistency, accuracy, and the rest - each split into Coverage (did the check run?) and Conformance (did it pass?). We grade on two planes - declared (the publisher's metadata) and measured (we read the bytes) - so a slice we have not measured can never render green. Across the corpus that gap is real: measured 93% vs declared 71% - reading the bytes catches what the metadata misses.