Weighed Portion Reference
Weighed Portion Reference. A test meal whose calories are known because every ingredient was weighed and converted with composition data before the meal was logged.
Building a meal you know the answer to
To test a calorie tracker, you need meals whose true content you already know. A weighed portion reference is the practical way to make one: weigh every ingredient on a calibrated scale, look up each ingredient’s composition, multiply, and add. The total becomes the ground truth the app’s estimate is scored against.
It borrows from the weighed food record used in nutrition research, long regarded as one of the more accurate ways of recording what people eat, and adapts it to the job of testing an instrument.
The steps, done properly
- Check the scale. A kitchen scale with 0.1 g resolution is enough, but resolution is not accuracy. Check it against a known reference weight, such as a certified 100 g mass, and note any drift.
- Tare between ingredients. Weigh each component on its own, zeroing the scale for the container each time.
- Record the state. Raw and cooked weights differ, sometimes enormously, because cooking adds or removes water. Use composition values for the state you weighed: raw values for raw weights, cooked values for cooked ones.
- Weigh what is hard to see. Oil, butter, sauces and dressings are small by weight and large in calories. They need the same care as the main ingredients.
- Look up composition values in a reference such as USDA FoodData Central, preferring analytical datasets over label data where possible.
- Record everything so another person could rebuild the meal: each ingredient, its weight, its state and the reference entry used.
Worked example: a portion from a pot of soup
Mixed dishes are cooked as a batch and eaten as a portion, so the reference uses the recipe method. A red lentil soup:
| Ingredient (raw) | Weight | kcal per 100 g | kcal |
|---|---|---|---|
| Red lentils, dry | 200 g | 358 | 716 |
| Onion | 150 g | 40 | 60 |
| Carrot | 200 g | 41 | 82 |
| Olive oil | 15 g | 884 | 133 |
| Water | as needed | 0 | 0 |
| Whole pot | 991 |
After cooking, the pot of soup weighs 1,600 g. A 400 g bowl is a quarter of the pot, so its reference value is 991 × 400 ÷ 1,600, about 248 kcal. The final weight matters: if the soup had simmered down to 1,300 g, the same 400 g bowl would hold about 305 kcal.
Notice the olive oil: 15 g, under a fiftieth of the pot’s raw weight, but more than an eighth of its calories. This is why references weigh fats separately and why apps that cannot see them struggle with mixed dishes.
Limits worth stating
A weighed reference is an excellent yardstick but not a perfect one. Composition values are averages, and a particular onion or batch of lentils will differ from them by a few percent. Small weights are less precise than large ones. And the method measures what was served, not what was eaten, so plate waste must be weighed and subtracted if it matters. A good study reports these limits instead of hiding them.
Why it is the standard for app testing
More direct references exist, such as chemically analysing a duplicate portion, or doubly labelled water for total intake over days, but they are too costly to repeat across dozens of meals each time an app is re-tested. Weighed references are accurate enough to reveal differences of a few percent between apps and practical enough to rebuild every quarter, which is why Methodology v3.2’s 50-meal battery is built from them. The trade-off is effort: preparing 50 weighed reference meals is several hours of lab work per re-test cycle. The MAPE entry explains how the app’s misses against these references are summarised, and the validation study entry how the whole exercise fits together.