How we calculate

Every nutrition, takeout and cost figure in Short Order is computed from public data with a written method and stated assumptions, and every one is an estimate.

Nutrition

Each ingredient line in a recipe is matched to a food record and a weight in grams. Per-serving nutrition is the sum of grams × the record's values per 100 g, divided by the servings. The app and this site use the same engine (version 1.0.0) and the same food list (461 foods, lexicon 2026.10).

• Food records: USDA FoodData Central Foundation Foods (release 2026-10-29) first, then SR Legacy (2018-04). A few packaged items, such as protein powder, use a brand-free median of several US product labels from FoodData Central Branded Foods.
• Cross-check: when the library is built, every line's text is parsed again on its own and checked against the written food and grams; a line that is more than 10% off is flagged as a build issue.
• Confidence: each dish gets a grade from how its lines matched. All 120 library dishes are graded High.
• Rounding: calories to the nearest 10, grams to the nearest 1. Always shown as “about” or “estimate”.
• The high-protein badge is a rule, not a vibe. Mains need at least 30 g of protein per serving and at least 7 g per 100 kcal; breakfasts 25 g and 6 g; sweets 15 g and 5 g; sides 10 g. 110 of 120 dishes meet it.

Takeout estimates

Restaurants don't publish one recipe for kung pao chicken, so the takeout side of every order is an estimate for a typical restaurant portion. We say which method each item uses.

• USDA survey foods (27 items): where the Food and Nutrient Database for Dietary Studies (FNDDS 2021–2023, released 2024-10-31) has the dish, we use its values per 100 g times a stated portion, such as 3 cups for a takeout entrée.
• Reference builds (29 items): where FNDDS has no such dish, we write a restaurant-style ingredient list, with every portion assumption stated, and compute it with the same engine as the home recipes.
• We never use a restaurant's own recipe or published numbers, and order names are generic. In the app you can enter a restaurant's published numbers yourself.
• Savings only appear when you enter what you paid. We never assume restaurant prices.

Prices

• Bureau of Labor Statistics Average Price Data, U.S. city average, for September 2026, where BLS publishes the item (for example boneless chicken breast, eggs, ground beef, milk, cheddar, rice and dried beans).
• USDA ERS Purchase to Plate National Average Prices (2017–18, price per 100 edible grams) for everything else, adjusted to September 2026 with the BLS consumer price index for the matching food group.
• Edible weight: bone-in cuts use a stated edible share, so the price is per gram you actually eat.
• Coverage rule: a dish shows its cost only when at least 90% of its ingredient weight is priced (112 of 120 dishes). Otherwise it shows “partial cost” and the share priced.
• Outside the US the app hides these estimates and uses only prices you enter. No sale prices, store claims or affiliate links.

Cooked weights

To tell you how many grams to scoop, the app estimates cooked weight with the USDA ARS Table of Cooking Yields for Meat and Poultry (Release 2, 2014) and the USDA Table of Nutrient Retention Factors (Release 6, 2007). Protein is conserved through cooking; water and fat change. Weighing the pot always beats the estimate, and the app lets you.

Food safety

• Doneness steps carry USDA FSIS safe minimum internal temperatures: 165°F for poultry, 160°F for ground meat, 145°F for whole cuts of beef and pork with a 3-minute rest, and 145°F for fish.
• Dishes with meat, poultry or seafood get at most 4 days in the fridge, following FSIS leftovers guidance of 3 to 4 days. The app never extends a date.

Protein targets

In the app you can pick a protein amount per plate (35, 45 or 55 g), or adults can work one out from body weight and activity: 1.2 g/kg for everyday, 1.6 for training and 2.0 for training hard, capped at 2.2, divided by meals a day. These follow the ISSN position stand on protein and exercise (Jäger et al., 2017) and Morton et al. (2018). It's a general guide, not medical advice; if you have a medical condition, ask your doctor.

AI

• Recipes are written with AI (Anthropic's Claude) and checked by automated QA gates and an AI review pass. We don't claim that a person cooked or tasted them. Nutrition and cost are always computed by the engine, never written by the AI.
• Food photos are AI-generated to show the finished dish.
• Nonna, the app's AI cook, quotes only the numbers the app computed and is labelled AI on every reply.

Sources

• USDA Agricultural Research Service, FoodData Central: Foundation Foods, SR Legacy, FNDDS 2021–2023, Branded Foods. Public domain (CC0). Downloads
• U.S. Bureau of Labor Statistics, Average Price Data and the Consumer Price Index (CPI-U food indexes).
• USDA Economic Research Service (2023), Purchase to Plate National Average Prices, 2017–18.
• USDA ARS, Table of Cooking Yields for Meat and Poultry, Release 2 (2014), and Table of Nutrient Retention Factors, Release 6 (2007).
• USDA Food Safety and Inspection Service, Safe Minimum Internal Temperature Chart and Leftovers and Food Safety.
• Jäger R. et al. (2017), International Society of Sports Nutrition position stand: protein and exercise, JISSN 14:20. doi:10.1186/s12970-017-0177-8
• Morton R.W. et al. (2018), A systematic review, meta-analysis and meta-regression of the effect of protein supplementation on resistance training-induced gains in muscle mass and strength in healthy adults, Br J Sports Med 52:376–384. doi:10.1136/bjsports-2017-097608
• Mona Sans, GitHub, SIL Open Font License 1.1. License text

Library built 2026-10-29. Prices as of September 2026.