How MorselMetrics works out its nutrition figures.
Every nutrition figure in MorselMetrics is calculated by fixed code from published food composition data, except for foods you import yourself from Cronometer, which keep the values in your file. This page explains where that data comes from, how a recipe is calculated, what is assumed, what happens when data is missing, what has and hasn’t been checked, and how we correct mistakes.
Last reviewed: 31 October 2026.
Where the figures come from
MorselMetrics uses published Australian food data first. Each source below is used only for the jobs listed.
Australian Food Composition Database (AFCD), Release 3
Published by Food Standards Australia New Zealand (FSANZ). It is the main source of nutrient values, used by Food search, Check a recipe, the Recipe book, the meal planner, Foods by nutrient and Recipes by nutrient. Values are per 100 g of the edible part of a food. They are mapped into MorselMetrics’ list of nutrients and units, so what you see is a modified version of the original data.
AUSNUT 2023
Also published by FSANZ. It is used for three jobs:
- household measures, which turn cups, spoons and pieces of a particular food into grams (see Cups to grams);
- nutrient retention factors, for the cooking estimates described under Cooking and preparation;
- food records where the AFCD has no suitable one: a few ingredients in the Recipe book, palm sugar and natural Greek-style yoghurt in Food search, and records offered when you check a recipe, including any you pick with Find missing food.
A backup food database
MorselMetrics uses one backup food database, from outside Australia, in only two places. It is named in the ⓘ beside the Protein building blocks line and on Data & methods.
- Protein building blocks: where an Australian food doesn’t report a full set of essential amino acids, AFCD’s profile for the same food raw or dry, or for a close Australian food, is used first; failing that, the amino acid profile of a matching food from the backup database may stand in. Either is scaled to the food’s own nitrogen or protein and marked ≈. The matches were proposed and cross-checked by AI agents and approved by MorselMetrics’ operator. The figures themselves were computed by a script from the published files.
- Turning millilitres into grams for four liquids that AUSNUT has no measure for: maple syrup, hoisin sauce, tomato pasta sauce and spirits. These are labelled estimates, not nutrient data.
Nutrient Reference Values for Australia and New Zealand
Published by the NHMRC and the Australian Government Department of Health and Aged Care. These are the daily references behind the bars. Edition used: 2006 NRVs; sodium and fluoride updates incorporated in 2017. See the official Nutrient Reference Values.
Protein reference pattern
Protein building blocks compares the essential amino acids in a recipe’s or food’s protein with the WHO/FAO/UNU 2007 adult reference pattern, per gram of protein, with no adjustment for digestibility. For an adult, it also compares the amounts with a day’s needs: protein with the Nutrient Reference Values’ recommended dietary intake for protein, and each amino acid with what that much protein holds at the same pattern, because the Nutrient Reference Values don’t set an intake for single amino acids.
Product labels and recipes
One ingredient in the Recipe book, a vitamin B12-fortified yeast flake product, uses the values on its UK product label, because fortification differs from product to product. Its record names the exact product and the month the label was checked.
The Recipe book mixes MorselMetrics’ own recipes with adaptations of openly licensed recipes from the Wikibooks Cookbook and the NSW Government’s Healthy Eating Active Living program, and a few community recipes. AI tools helped write and revise MorselMetrics’ own recipes and its adaptations. Each recipe page names its source and licence, and lists the food record used for each ingredient, under About these numbers and sources.
MorselMetrics is independent. None of the organisations whose data it uses endorses it or its use of their data. FSANZ’s attribution, licences and limitation statement are on Data & methods.
How a recipe is calculated
- Match each ingredient to one food record. In Check a recipe, MorselMetrics suggests food records for each line. A clear match is filled in for you. Where a name could mean different foods, such as raw or cooked, dry or canned, a brand or a fat level, you’re asked to choose. Every match stays on screen for you to check or change. In the Recipe book, the food record for each ingredient was chosen when the recipe was prepared, and a search never swaps it.
- Turn each amount into grams. Gram amounts are used as entered. Cups, spoons and pieces use the AUSNUT 2023 household measure for that food where there is one, with Australian 250 mL cups, 20 mL tablespoons and 5 mL teaspoons unless you choose US measures, and a liquid uses its own density. Other amounts, such as a can to be drained or “a pinch”, and conversions without a published measure get a labelled working estimate you can edit, or wait for you to enter a weight.
- Scale the food record. For each nutrient, an ingredient contributes its weight in grams ÷ 100 × the record’s value per 100 g.
- Add the ingredients up. The whole-recipe amount is the sum of the ingredients’ amounts. Ratios and classification codes are never added.
- Divide by servings. One serve is the whole recipe divided by the number of servings. In Check a recipe, changing the servings divides the same ingredients into more or fewer serves. On a Recipe book page, Make servings scales the ingredient amounts instead, so a serve stays the same. In the meal planner, each person’s figures follow the portion you give them.
- Finished weight, if you enter one. If you weigh the cooked dish, MorselMetrics also shows amounts per 100 g of it. Entering a finished weight doesn’t change the totals.
The Recipe book is calculated by the same code as Check a recipe. For a worked example, see How to calculate recipe nutrition.
Cooking and preparation
Each food record describes a food in one state: raw, cooked, canned, drained and so on. MorselMetrics uses a record in the state it describes. A record that is already cooked is never adjusted a second time.
For some raw records you can choose a cooking method. MorselMetrics then applies FSANZ’s published nutrient retention factors from AUSNUT 2023, linked to that exact food record through the AFCD Release 3 recipe file. Only nutrients with a published factor are adjusted. The rest stay as recorded and are marked as estimates.
If you don’t choose a method, or none is published for that food, the record is used as it is and labelled as an ingredient estimate: losses from cooking aren’t modelled. The weight a dish gains or loses in cooking isn’t applied to individual ingredients.
When data is missing
- Not every food record reports every nutrient. A value the source doesn’t report is shown as Not reported. It is never counted as zero.
- When some ingredients report a nutrient and others don’t, the total includes only what is reported. It is marked * (partial data) because it may be understated.
- An ingredient without a food match or an amount is left out of the totals, which are then marked as partial.
- Rankings leave gaps out rather than guessing. Foods by nutrient lists only foods that report the nutrient, and Recipes by nutrient includes a recipe only when every ingredient reports it.
- Protein building blocks appears on a recipe page or in Check a recipe only when there is at least 5 g of protein, and in Food search only for a food with at least 2 g of protein per 100 g. Either way, 95% or more of the protein must have an amino acid profile.
- Source values may be analysed, calculated, borrowed from a similar food, estimated or imputed, depending on the food and the nutrient.
Source values, estimates and daily references
A food’s values per 100 g are the published figures from its source, converted to MorselMetrics’ units. Everything shown for a recipe, a serve or a day is calculated from them. It is an estimate of what the ingredients contain, not a laboratory analysis of your dish. Food composition varies between batches, brands and seasons, and the amounts don’t account for how much of a nutrient the body absorbs.
≈ marks an estimate: every value is reported, but something is assumed, such as a stand-in food, an estimated amount or a cooking state you haven’t confirmed. A stand-in is used only when no record exists for an ingredient, and only from a short reviewed list of foods with similar composition and use. Sumac, for example, can use paprika. A stand-in is always disclosed and never becomes the ingredient’s name, shopping-list entry or allergen information.
The bars compare an amount with the Nutrient Reference Values for the age, sex and life stage set for each person, including pregnancy and breastfeeding. Recommended intakes (RDI and AI), upper levels (UL) and suggested dietary targets (SDT) are kept separate and never combined into one score. Upper levels are shown as a warning, not a target.
These are population references. They are not personalised advice, and a bar is a comparison, not a diagnosis. Your own needs can differ. If a health professional has given you targets, you can enter them as your own.
Where AI is used, and where it isn’t
No nutrient value comes from AI. The nutrition maths is done by fixed code from the sources above.
In the app, an AI model from OpenAI is used only in these features, and only when you use them:
- Reading a recipe. When MorselMetrics can’t read a recipe’s text on its own, it may ask the model for help. That applies wherever a recipe is read: a recipe you paste, or import from a link, in Check a recipe or when you add a new recipe to the meal planner, and a recipe written by Write me a recipe or by Make another version. A meal you describe in the planner is first split into ingredient lines by our own code, in your browser, keeping your words and amounts; those lines are then read the same way. The model can reword an ingredient so our search looks again, but it can only choose records our search returns. It can also suggest a working amount for vague quantities such as “a pinch” or “to taste”, how many servings the recipe makes when it doesn’t say, or how much of a marinade is eaten, all labelled as estimates you can change. Its choices go through the same checks as any other match.
- Writing a recipe. In Use what I have, Write me a recipe asks the model to write a recipe from the ingredients you type: ingredients with gram weights, servings, method and an estimated time. The model is sent our own names for your foods and whether each is raw, cooked, canned, dried, frozen or a leftover, never the words you typed. Before you see it, our own code checks it: no nutrition, health or diet claims; every ingredient weighed, with sensible amounts per serve; every food the title or method mentions on the list; your foods in the state you have them; raw meat, poultry, fish and eggs cooked, with poultry, mince and sausages cooked to 75°C in the centre and leftovers heated until steaming hot; and a time no shorter than the method. If you choose vegetarian or vegan, a recipe is shown only when our own food data confirms it for every ingredient. Each ingredient is then matched to a food record and calculated like any other recipe. A written recipe is marked as ours, so it can’t be shared as your own.
- Planning your own recipes. To use one of your saved recipes in meal planning, the model may suggest its meal slots, dish type, servings and preparation time, marking each as stated in your recipe or suggested.
- Writing another version of a recipe. On a MorselMetrics recipe, Make another version asks the model to rewrite it the way you ask: vegetarian, without an ingredient you name, with more protein or with less saturated fat. It writes ingredients with gram weights, the same number of servings, and a method. Its text is checked for nutrition, health and diet claims. Each ingredient is then matched to a food record and calculated like any other recipe. Whether a version is vegetarian, or really is without the ingredient, is decided by our own ingredient reviews and never by the model; where those reviews can’t tell, the page says so. More protein or less saturated fat is claimed only when the calculated values per serve show it. A version is saved as a new recipe; the original is not changed.
- Reading a typed goal. In Autofill you can type what you want in a sentence. Our own code reads it in your browser and fills in the form, so most sentences never leave your device. If it can’t read part of a sentence, the model is sent the words you typed and nothing else, and it can answer only by choosing from the same fixed settings the form has: days, meals, diet, listed ingredients to leave out, cooking time and nutrients to put first. It can’t add advice, or a number you didn’t type. A sentence that mentions health is not sent at all. Nothing is saved until you preview the meals and add them.
- Pics you post. Before a pic you post on a recipe goes up, OpenAI’s moderation model checks it for harmful content, and the model checks that it shows only food and kitchen things (a hand at the edge is fine). A pic that passes both goes up; any other waits for MorselMetrics to look at it. When someone reports a pic, the moderation model checks it again. The models get the picture only, with its location and camera details removed, never your name or account.
- Recipes you share. When you share one of your recipes to Community recipes, the model reads its name, description, ingredient names and method (never its nutrition, your name or your account) and checks it for unsafe cooking, food for babies, alcohol-heavy dishes, health claims, copied or branded text and non-food ingredients. A recipe it finds clear goes up; anything it flags, or any time it can’t answer, waits for MorselMetrics to look at it. The nutrition shown with a shared recipe is calculated by the same fixed code as every other recipe.
- Importing from Cronometer, if you choose it. AI column help suggests which columns of your file hold which details, from a small sample. Use AI to match these foods suggests a food record and an estimated weight for each food. You review both before anything is saved.
None of these answers has a field for a nutrient value, and the model can’t decide allergen or diet suitability. How recipe text sent to the AI provider is handled is explained in the Privacy Policy.
Behind the scenes, AI tools also helped write and revise MorselMetrics’ own recipes and adaptations in the Recipe book, and proposed and cross-checked the backup amino acid matches described above.
What has been checked, and what hasn’t
These statements are about the recipes MorselMetrics wrote or adapted, and about MorselMetrics’ own calculations. An original source may have had its own checks.
Checked
- Each revision of a Recipe book recipe is published with a recorded checklist: the food record for each ingredient, allergen and diet labels, the servings the method supports, timings and equipment, the method text and the right to use the recipe. The checks are done with AI tools and automated tests, not by a dietitian, nutritionist or chef.
- An automated check reads every method in the Recipe book and fails if it gives a temperature below Australian food safety guidance for mince, sausages, poultry or eggs.
- On 30 October 2026, every recipe summary that mentioned a nutrient was checked with AI tools against the per-serve conditions for nutrition content claims in Schedule 4 of the Food Standards Code. Two were reworded.
- The calculation code has automated tests.
Not done
- MorselMetrics hasn’t kitchen-tested the recipes. We haven’t cooked them to confirm timings, yields or results.
- No dietitian or nutritionist has reviewed MorselMetrics’ recipes and adaptations, its calculations or this site.
- No outside organisation has audited the calculations.
- MorselMetrics is not a medical or diagnostic service. Its figures aren’t suitable on their own for managing a medical condition or a food allergy. Recipe allergen notes aren’t an allergy-safety guarantee, so check product labels.
How we correct mistakes
When a recipe in the Recipe book changes, the new version is usually published as the next numbered revision at the same web address, so links keep working. A copy you saved to My recipes keeps the version you saved; it doesn’t update.
If a change matters to anyone who used the earlier version, it carries a correction notice, classed as safety, suitability, source or nutrition. If you planned an earlier version, the meal planner puts a dot on that day and lists the notice in its recipe updates panel, which you open to read. A shopping list can’t be saved or printed with a corrected meal in it; to clear that, replace the meal with the current recipe or leave it out of the list. A list you have already printed or saved as a PDF can’t update itself. After a safety, suitability or source correction, the earlier version, including a copy saved in My recipes, can’t be added to the planner again.
Recent corrections
- 29 October 2026, safety. Three methods gave doneness temperatures below Australian food safety guidance. The lamb couscous now says at least 63°C in the centre (medium-rare), or 71°C for medium, and the Greek lamb meatballs and curried sausages say 75°C. Read the What’s new entry for safer cooking temperatures.
- 29 October 2026, safety. Six more methods with mince or sausage meat now say 75°C, and the cheesy scrambled eggs 72°C. Read the What’s new entry for safer cooking temperatures.
- 29 October 2026, suitability. The mushroom, pea and parmesan risotto, the creamy pumpkin soup and the minestrone now say vegetable stock, so they stay vegetarian as labelled. Their nutrition is unchanged: they already used an all-flavours reduced-salt stock record, which stands in for vegetable stock. Read the What’s new entry for the cook’s pass.
- 30 October 2026, wording. After the Schedule 4 check above, two recipe summaries were reworded. The cheddar and pear plate no longer mentions riboflavin, and the prawn and egg fried rice no longer compares its salt with takeaway. Nothing else in either recipe changed, so neither carries a correction notice.
What’s new lists the changes you’re most likely to notice, but not every correction. The recent recipe corrections are listed above.
Report a problem
If a figure, food match or recipe looks wrong, tell us through the contact form: choose Food or nutrient data issue, and include the page and the food or recipe. If you have an account and are signed in, you can also use Feedback on the right-hand edge of any page.
Who is responsible
MorselMetrics is an independent project, made in Australia. The operator is named in our Terms of use. To ask about this page, a figure or a correction, see Report a problem.
Nutrition information is for general educational use and is not personal medical advice. See the Nutrition disclaimer and Terms. Source attribution and licences are on Data & methods.