Measurement / An explorable

Can a food log find a deficiency?

You weighed everything for a month. The macros are dialled in. So the natural next question is whether the micronutrients are covered too, and it turns out that question splits in half: one part your food log answers cleanly, and one part it cannot answer at all, no matter how carefully you weighed.

Drag the sources of error and watch what happens to the verdict~9 min

How you get here

Macros are counting. Micros are chemistry.

Tracking macros is close to bookkeeping. Protein, carbs and fat are measured in grams, they are printed on the packet, and the numbers add up the way numbers do. After a few weeks of weighing food you develop a real feel for it, and the feeling generalises: I know what I ate, therefore I know what I took in.

Then you start wondering about the other column. Am I getting enough magnesium? Zinc? Iodine? The food log is right there, every gram of it. Surely you can just look them up.

You can. Matching a month of weighed entries against a composition table like USDA's SR Legacy is tedious but entirely doable, and it produces a tidy number for every nutrient. The trouble is what that number is. A macro figure is a count. A micronutrient figure is a chain of estimates, and each link multiplies its own error into the result.

The instrument

Watch the verdict break

Below is one nutrient's daily intake, expressed as a percentage of the requirement. Start with the raw table figure and no error accounted for, then switch on the things a food log genuinely cannot pin down. The band is the honest interval. The red line is the requirement.

Intake vs requirement -

requirement

Answerable

Switch every source on for each of the three profiles in turn. The absent nutrient stays absent and the abundant one stays abundant: their bands are wide, but the whole band sits on one side of the line, so the verdict never moves. The one sitting just over the requirement loses its verdict entirely. Its band straddles the line, and no amount of extra weighing narrows it, because the uncertainty is not in your scale. It is in the table, the gut and the pan.

Where the error actually comes from

Six honest reasons the number moves

None of these are exotic. They are the ordinary conditions under which any composition table is built, and every one of them is invisible from inside a food-logging app.

The table is an average of a variable world

Selenium in a plant tracks the selenium in the soil it grew in, and that varies by more than an order of magnitude between regions. Iodine in milk swings several-fold with the season and the dairy's teat-dip practice.

10x to 100x between samples

Eating it is not absorbing it

Non-heme iron, the kind in plants, is absorbed anywhere between about 2% and 20% depending on what is on the plate beside it. Vitamin C lifts it; tea, coffee and phytate suppress it. Your log records the mouth, not the bloodstream.

up to 10x spread on one figure

Phytate quietly taxes the minerals

Oats, flax, chia and pumpkin seeds are excellent food and also rich in phytate, which binds zinc and iron in the gut. A daily wholegrain breakfast can make a comfortable-looking zinc intake land near the requirement absorbed.

a bias, not a wobble

Cooking is subtraction

Composition tables usually list raw values. Boiling takes a large share of the folate and a quarter to half the vitamin C, and time on a shelf takes more. The table never subtracts it, so your estimate is systematically generous.

folate 50-80%, vitamin C 25-50%

The table has a nationality

American pasta, flour and rice are enriched by law. European versions mostly are not. Match your food against a US table while eating in Italy and the folate, thiamin and iron columns are all quietly inflated.

a one-way error of 5-15%

Some columns are simply empty

The most useful discovery of the whole exercise: USDA SR Legacy contains zero iodine values across all 7,793 foods. Sodium is listed but every entry is unsalted, so it ignores the salt cellar. Two nutrients that matter, and the method has nothing to say about either.

not uncertain: absent

The reframe

It is not a measuring instrument. It is a triage instrument.

The food log cannot tell you whether you are sufficient. It can tell you which questions are worth paying a laboratory to answer.

Once you stop asking the estimate to adjudicate and start asking it to prioritise, it becomes genuinely useful, because triage only needs the coarse verdicts that survive the error bars. Something structurally missing shows up loudly. Something abundant is safely ignorable. Everything in the ambiguous middle gets escalated to a blood test, and blood tests cost money, so the log has just done real work by shortening the list.

There is a second filter, and it catches people out. Some nutrients have no useful blood test either. Guess which of these can be settled by a needle before you turn each card over.

What happened next

Marking the estimate against blood

The honest way to grade a method is to write its predictions down, buy the tests, and see. Here is how one month of weighed logging scored when the results came back, without the personal numbers, because the pattern is the interesting part and the values are not.

Confirmed The nutrient the log called abundant, at high confidence Discrete animal-source foods, well characterised in the table, no meaningful cooking loss. The log said comfortably covered. Blood came back solidly mid-range. This is the easy case, and the method handled it.

Confirmed The nutrient the log hedged as low confidence Adequate on paper, but the table lists raw values and boiling destroys a lot of it. The log said covered while flagging its own uncertainty. Blood came back normal, in the lower third. The hedge was the accurate part of the prediction.

Confirmed The store the log admitted it could not see The log could estimate intake but said outright that intake tells you nothing about the reservoir. The blood test for the reservoir came back healthy. Being right here meant naming the gap, not filling it.

Overturned The one the log said needed no attention Intake was high but well within what healthy kidneys clear, so the prediction was that blood would read normal. It read slightly over the range instead. A useful reminder that "the body regulates this" is a mechanism, not a measurement.

Untestable The flagged deficiency the blood could not judge Diet supplied a small fraction of the reference, which looked alarming on paper. Blood came back fine, because the dominant source was never food at all. The log's refusal to call this one was the correct answer, not a failure.

Still open The four findings that started all of this The genuine flags, the ones the estimate was confident about, all landed on nutrients with no useful blood test. They were never going to be resolved by a lab. They were behaviour changes from the start, and they still are.

What to take from it

The method is worth running. Once.

Matching a month of weighed food against a composition table takes an afternoon and it will not tell you whether you are deficient in anything. What it will do is convert a vague background worry into a short, specific list of questions, sorted by whether a laboratory can answer them.

That is a real result. Most people carrying the question "am I getting enough micronutrients?" are carrying something unresolvable, and the useful move is not to answer it but to break it into the part you can settle for the price of a blood draw, the part you can settle by changing what is in the cupboard, and the part nobody can settle and you can therefore stop thinking about.

And run one calibration check before trusting any of it. Sum the matched entries back up into calories and compare against what your tracker recorded independently. If those agree closely, your food identification and portion parsing are sound, which is worth knowing. It says nothing whatever about whether the micronutrient columns are right, which is the whole point of this page.