Query the dataset.

Run SQL against the covered scorecard dataset, right here in your browser. One table, agencies, holds every published feed record's latest snapshot: id, name, date, grade, score, days_until_expiry, producer-version fields, comparison_eligible, and the correctness, freshness, completeness, and realtime category scores. Nothing is sent to a server: the engine (DuckDB) and the data load into the page and the query runs on your machine.

Prefer a file? The same data is agencies.parquet, catalog.csv, and the JSON described in the data dictionary.

Comparison warning: the table includes old-methodology, duplicate-identity, and otherwise non-comparable rows so the public record stays complete. Do not compare or average scores unless you filter comparison_eligible = true and inspect the current contract in agencies.json. The examples below are support worklists and provenance checks, not rankings.

The query engine (about 6 MB) downloads the first time you press Run.

Remember the sampling frame: this is the covered set of feeds, not the universe of agencies, and absence means not covered, never failing. Data CC BY 4.0.