How We Verify Our Numbers

Every data study on this site runs on the same rules. Here they are, in full, so you can check our work.

When we publish a number like "black beans give you 10.4 grams of fiber per dollar," that number didn't come from vibes. It came from a specific USDA database entry divided by a specific shelf price, and both of those are documented. This page explains exactly where the numbers come from, how we handle the annoying edge cases, and what happens when we get one wrong.

Where the data comes from

The as-purchased rule

Every calculation uses the food as it sits in your cart, not as it sits on a lab bench. Dry lentils are priced and measured dry. Canned beans are measured drained, because nobody eats the can liquid on purpose. If a study compares cooked values, it says so explicitly and shows the conversion. Mixing "per 100g cooked" with "per 100g dry" is how you get rankings that look impressive and mean nothing, so we don't do it.

Edible-fraction adjustments

A pound of oranges is not a pound of orange. Peels, pits, rinds, and bones come out of the math before anything gets ranked. We apply USDA refuse percentages to adjust the price to the part you actually eat. This matters more than people expect: skipping this step quietly flatters foods with heavy waste, like avocados and winter squash.

The self-audit before publish

Before any study goes live, every value gets re-verified against two independent USDA pulls. Two separate queries, two separate sessions, compared line by line. If the pulls disagree, the row gets flagged and resolved manually before publication. It's tedious. That's the point. A ranking is only useful if the boring rows are as accurate as the surprising ones.

Every study ships its raw data

Each data study publishes its complete dataset as a public CSV: every food, every price, every source ID, every calculated value. No "data available on request." It's just there.

You're welcome to reuse the data. If you cite it, link back to the study so your readers can see the methodology too.

When we get it wrong

We correct errors publicly, in the study itself, with a dated note. We don't quietly swap numbers and hope nobody noticed.

Real example: on July 4, 2026, an adversarial audit of the fiber per dollar study caught six values that needed correcting. We fixed them, re-ranked the affected foods, and published a correction note in the article. The audit process that caught them is now part of the standard pre-publish checklist.

How often the numbers get refreshed

The studies this applies to

Questions about the data, or spotted something that looks off? Tell us. Getting corrected beats being wrong.