Cobalt Recipe Check

Give it a recipe and it estimates how much cobalt is in one serving. The estimate uses median values from lab analyses of about 430 foods.

How the score works

Each ingredient line is matched to a food in the dataset and turned into grams. Then grams × the food’s median cobalt concentration gives its cobalt. The total for the recipe is divided by the number of servings and scored on the dataset’s own scale:

  • 0 under 100 ng per serving
  • 1 100 to 800 ng
  • 2 800 to 1,600 ng
  • 3+ one more point for every 800 ng after that

Medians are used rather than averages because some foods have a few extreme samples. For example, cooked ground meat averages 59 ng/g, but its median is 7 ng/g. The “typical range” uses the middle half of samples for every ingredient (25th to 75th percentile).

That scale was made for single foods, so a whole meal can score higher than any one of its ingredients. Use the score to compare recipes, or to spot the ingredient that contributes most. It isn’t an exact measurement.

Assumptions and limits
  • Ingredients are scored as bought (raw, dry or canned), because that’s how recipes list them. Lentils and dried beans count as dry unless the line says canned or cooked.
  • Cups and spoons are converted to grams with typical densities, and counted items (“1 onion”) with typical weights. Amounts marked estimated come from these conversions. You can type the real weight instead.
  • Some ingredients have no data of their own and borrow a close relative, e.g. kale uses spinach. Those are labelled. Foods with only one or two samples are marked few samples.
  • Lines with no amount, such as “salt to taste”, count as a pinch for seasonings and as nothing for other foods. Serving suggestions (“rice, to serve”) aren’t counted until you give them an amount.
  • This is not medical advice. If you follow a low-cobalt diet for a medical reason, check with your doctor or dietitian.
Where the data comes from

Cobalt values come from the Low Cobalt Diet spreadsheet (release 2.0.1, 2024). It combines about 4,000 lab samples from over 100 sources, including total diet studies from Canada, Australia, France and Sweden, the New Zealand food composition database and peer-reviewed studies. Its author cautions that the data may contain errors. This tool takes per-sample medians from that file and doesn’t change any values.

List every source