Pensum
How to track micronutrients
Where a tracker's vitamin and mineral numbers come from, and which filled-in values are worth trusting.
Most food trackers lead with calories, protein, carbs and fat because those are the numbers the data reliably has. Every nutrition label and nearly every database row carries those four numbers. Far fewer carry magnesium, folate or zinc. A branded yoghurt will list its energy and macros and say nothing at all about its minerals, so a day that looks complete in a tracker usually has gaps in its vitamins and minerals that you cannot see.
Tracking micronutrients well comes down to logging foods that carry measured values, and knowing which numbers were filled in by something else.
What a national food database covers per food
The most complete micronutrient data comes from national food composition databases, where foods are analysed in a lab and the results are recorded per 100 g. Pensum's catalogue is built on USDA FoodData Central, with the German BLS and the Swiss Food Composition Database adding more generic foods. Together that is 16,358 foods, and 11,504 of them have lab values for all twelve vitamins and minerals I used in my benchmark.
Pensum tracks 18 micronutrients plus fiber, sugar and saturated fat against daily targets. Each search result shows its source and how many of the 18 it covers, and generic whole foods rank above branded products, so "oats" or "chicken breast" returns the plain food first. So for complete micronutrient days, log the generic entry by weight whenever the food on your plate is a plain one. A weighed portion of oats from a national database brings its measured vitamins and minerals with it.
Where the gaps come from: branded products and mixed dishes
Branded products are the first gap. Pensum finds them through Open Food Facts, which holds millions of products kept current by people who scan them. Those entries carry what the package label prints, which is energy and macros and seldom much else. A diary built mostly from packaged food shows full macro coverage and thin micronutrient bars.
Mixed dishes are the second. A burrito or a ham pie has a micronutrient profile that depends on the ratio of its ingredients, and a dish name does not record that ratio. When you build a dish yourself as a recipe in Pensum, its micronutrients are added up from the ingredients you weighed, so a home-cooked dish is as complete as the foods that went into it.
How Pensum fills the gaps
The first step is a match. A plain branded product, such as a carton of milk, is usually the same food as a generic row in a national database, and that row has measured micronutrients. Pensum looks for that generic food and fills the product's missing micronutrients from it. The source line in the entry sheet shows which food was used. Tap it to choose a different generic food yourself, or to use an AI estimate only. When no generic food is a clear match, Pensum uses none, because a wrong reference would put confident wrong numbers in the diary.
The second step is an AI estimate for whatever is still blank. A known value, including a known zero, always wins, and the estimate is stored as a separate layer on top of the food, so removing it restores the original row exactly. Estimated values never leave the app in an export or a database contribution. Estimating missing micronutrients is free, and with Pensum Plus, which pays for what costs money to run, they fill in on their own as you log.
Which micronutrient estimates to trust
To measure the estimator, I took 250 foods with lab-measured values, balanced across USDA, BLS and the Swiss database, deleted their micronutrients and asked the model to put them back, using the exact prompt the app sends. The micronutrient benchmark has every table. Overall, the median error was 25% for single ingredients and 33% for mixed dishes, against 76% and 69% for a baseline that ignores the food and predicts the catalogue median. So the model is reading the food, and its accuracy depends heavily on the nutrient:
- Minerals are solid. Magnesium and potassium landed at 15% median error and zinc at 20%, and they were almost never wildly off.
- Vitamins are rough. Vitamin C was off by 73% at the median, vitamin D by 60% and vitamin A by 59%.
- Vitamin errors run low. The median bias was -51% for vitamin C, -44% for vitamin D and -39% for vitamin A. For raw rose hip the lab measured 1045 mg of vitamin C per 100 g, and the model said 426 mg.
- Mixed dishes fail the other way. The model anchors on the ingredient in the name: canned chicken liver pate measures 217 µg of vitamin A per 100 g, and the model answered 5500 because it estimated liver.
Mineral content is largely a property of the raw material and moves within a narrow band. Vitamin content swings with variety, freshness, storage, cooking and fortification, and a food's name does not carry that information. A larger frontier model did worse on every cell of the benchmark, the same result I found when testing photo portion estimates.
What the daily micronutrient bar shows
On every micronutrient bar in Pensum, the share that came from an AI estimate is painted orange, the colour used for estimated data everywhere in the app, and the measured share stays green. Each value is also flagged verified, partial or estimated. You can see at a glance how much of a bar is measured, and given the results above, the orange tends to show up on the vitamins.
In daily terms the typical estimate is small. Scored against the daily reference value, a single food's estimate was off by under one point of the bar, and a dish's by under two. The tail is wider: one estimate in ten was off by roughly ten points or more. A row that reads 40% covered could be 30% or 55%. Read a mostly green bar as measured, and a mostly orange vitamin bar as a rough guide that leans low.
I do not hide the vitamins or put them behind a warning. A labelled estimate of your vitamin A tells you more than a blank, as long as you can tell it apart from a measured value.
Tracking vitamins and minerals for free
Cronometer's free plan shows ads and keeps reports and charts to a 7-day window, and longer history and its nutrition scores sit behind Gold at $59.99 a year in the US. In Pensum the 18 micronutrients, the verified, partial and estimated flags, and trends over a week, a month or three months are free, with no ads and no account. The Pensum vs Cronometer comparison has the full list.
Weighing what you can and logging generic foods keeps your calories accurate too, and that weighed diary feeds Pensum's measured expenditure, explained on the expenditure page. For the meals you cannot weigh, how accurate AI calorie counting is covers how far a photo estimate can be trusted.