LOWDOWN / OXALATE CORPUS

Where the data comes from

Lowdown is a glass box. Every oxalate and sodium number traces to a named source with a date. We keep conflicting sources side by side instead of averaging them, and we never fill a gap with a guess.

Corpus … · … records

Oxalate sources

Compiled into the published corpus. Harvard's 2023 lab-measured table is the trust anchor (priority 1); older and regional sources are kept alongside so you can see disagreement.

Source	What	Date
A Harvard T.H. Chan	Official oxalate table, 433 foods, lab-measured flags (Knight lab / UAB). hsph.harvard.edu	2023
Harvard 2008 list	~500 foods, via the oxalate.org compilation	2008
USDA 1984 / Dr. Duke	Vegetables + phytochemical DB (105 foods)	1984
FAO Bangladesh	Regional + soluble/insoluble (40 foods)	2014
NUTTAB	Australian food tables (17 foods)	2010
Zarembski	Historical assay (81 foods)	1962
Savage	Soluble/insoluble lab data (14 foods)	2001

Sodium source

Source	What	Date
B USDA SR Legacy	7,709 foods, sodium mg/100g. Token-matched to corpus foods with a per-row confidence and a salted/unsalted guard; ~71% coverage. fdc.nal.usda.gov	2018

Restaurant + packaged foods (crosswalk)

Ingested into the lakehouse for the menu decoder and barcode lookup. These carry live retrieval dates from our ingest registry — the table below is pulled from the system, not typed in.

Source	What	Last retrieved
Loading freshness…

How it stays current

Scout runs weekly — sweeps PubMed for new oxalate-assay papers and watches the Harvard table for changes, filing proposals for review.
Nothing auto-merges. Proposals sit in a review queue; a human approves each change, and it lands in a versioned, public changelog.
The corpus is versioned; the app checks for updates against a pinned version endpoint. The badge at the top is the live version.

How we handle it

Disagreement is data. Conflicting values are stored and shown, never averaged.
Unknown beats guessed. A missing match shows as "no match," never a nearest-neighbor guess.
Oxalate is an estimate — it varies by ripeness, origin, and prep. Sodium is label data per 100g.
Restaurant/photo/barcode oxalate is estimated from ingredients, with the per-ingredient chain shown.

Sources credited to their owners; USDA data is public domain, MenuStat via Harvard Dataverse (DOI-pinned). Estimates, not measurements. Not medical advice.