USDA's Local Food Portal is the canonical US farmers market dataset, but the raw feed is rough: around 40-char truncated names, stale records, a state= filter that substring-matches state NAMES (querying "WA" returns Delaware rows), and thousands of missing websites/hours.
I've been cleaning and enriching it: merged in state sources and filtered the result down to 8,863 real markets (~270 merged entries turned out not to be farmers markets at all — single-farm stands, mobile food pantries, retail grocers), backfilled websites from the live API + state sources (~50% coverage now),, added season/SNAP/organic fields, and published the whole thing CC-BY as CSV/JSON with a DOI (Zenodo, also mirrored on HF/Kaggle/GitHub).
The site on top of it has per-state statistics pages and a season finder. Happy to answer anything about the data quality issues — the state-filter substring bug was a fun one to find, but the weirdest was ~60 listings whose "official websites" are expired domains that gambling-SEO farms have since hijacked — the pages return HTTP 200, so naive liveness checks pass them.
1 comment
[ 2.6 ms ] story [ 18.8 ms ] threadI've been cleaning and enriching it: merged in state sources and filtered the result down to 8,863 real markets (~270 merged entries turned out not to be farmers markets at all — single-farm stands, mobile food pantries, retail grocers), backfilled websites from the live API + state sources (~50% coverage now),, added season/SNAP/organic fields, and published the whole thing CC-BY as CSV/JSON with a DOI (Zenodo, also mirrored on HF/Kaggle/GitHub).
The site on top of it has per-state statistics pages and a season finder. Happy to answer anything about the data quality issues — the state-filter substring bug was a fun one to find, but the weirdest was ~60 listings whose "official websites" are expired domains that gambling-SEO farms have since hijacked — the pages return HTTP 200, so naive liveness checks pass them.