Corrections

Most ingredient charts on the web publish a single number for a cup of anything and do not say where it came from. When you compare those numbers against USDA FoodData Central measurements, or against the independent physical standards that exist for particular substances (Codex Alimentarius for oils, FDA identity standards for dairy), a pattern appears: a small number of sources publish figures that are impossible on physical grounds and have propagated across every downstream site that copied from them.

The most common of these is the King Arthur Baking Ingredient Weight Chart’s figure of 227 g per cup for hydrated dairy products. That number is 8 × 28.35, the conversion for weight ounces, applied to a fluid-ounce volume. It appears on King Arthur’s chart for milk, yogurt, heavy cream, half-and-half, Greek yogurt, buttermilk, pumpkin purée, ricotta and evaporated milk. Every one of those products is more than 90% water plus dissolved solids that are denser than water; the mixture must be denser than water. A density lighter than water (0.960 g/mL) is physically impossible for them. It is a unit-conversion mistake, not a measurement.

The extension of this argument beyond dairy: King Arthur’s figure for olive oil is 50 g per quarter-cup, which extrapolates to 200 g per cup and implies a density of 0.8455 g/mL. That figure falls 7.1% below the Codex Alimentarius standard range for olive oil (0.910–0.916 g/mL at 20°C, CXS 33-1981). No olive oil can be that light and still be olive oil under the international standard. The underlying arithmetic pattern is the same as the dairy rejections: 7 × 28.35 = 198.45, applied to a fluid-ounce volume that should have been converted through the substance’s density instead. King Arthur’s published figure of 200 g is the pattern rounded to the nearest 25 g interval their chart uses.

This page lists every rejection currently in the dataset, with the accepted figure the ingredient page publishes instead, and links to a per-rejection subpage that walks through the physics or the standards for that specific case. Each rejection is falsifiable: if a source can produce a bench measurement supporting the rejected figure, the rejection reverses. So far, none have.

The list will grow as the dataset does. Every ingredient added with a rejected source generates a new correction subpage automatically.