How We Calculate Everything

Where the data comes from

Each compound record cites its own sources — the NIH Office of Dietary Supplements, Examine.com and indexed PubMed identifiers, linked directly on every page. Records carry an evidence level from A to C and a last-reviewed date.

Nothing is scraped from competitor sites, and no figure is invented to fill a gap. Where a value does not exist for a compound, the field is left empty rather than estimated.

Compatibility score

A pair starts at a neutral 70 — the baseline for any combination that has never been studied. From there the score moves only when the underlying records say it should:

• documented synergy: +12 at evidence level A, +8 at B, +5 at C
• documented antagonism: −30 for major clinical significance, −18 moderate, −9 minor
• both directions are checked — compound A may list B without B listing A
• the result is clamped to 5–99 and recalculated on every request, never stored

Elemental percentage

This is how much of a compound is the mineral itself, as opposed to the carrier the label counts as weight. It is calculated from the molecular formula and molecular weight: the atomic weight of the mineral, multiplied by the number of its atoms in the formula, divided by the molecular weight of the compound.

Worked example: magnesium glycinate is C4H8MgN2O4 at 172.42 g/mol, and magnesium weighs 24.305 — so 24.305 / 172.42 = 14.1% elemental magnesium. Magnesium oxide, by contrast, is 60.3% elemental magnesium, which is why it looks superior on a label.

Absorption score, and why it is not a percentage

This is the number we are least comfortable with, so here is exactly what it is. Each form carries a score from 0 to 100 that ranks it against other forms of the same compound. It is not the share of the dose that reaches your bloodstream.

We relabelled it on 2026-08-02 after auditing our own data. The scores had been presented as percentages, and they could not be: chelated forms sat at 85–95%, while absolute fractional absorption of calcium is about a quarter of the dose in any form, iron 5–35%, zinc 20–40%, magnesium 20–50%. The ranking the scores encode is directionally right — oxide really is far worse than citrate — but the numbers were never measured absorption.

Where a human pharmacokinetic study gives an absolute figure, that figure is shown separately as "Measured", with a link to the paper. Nine compounds carry one today: CoQ10 at 8% (PubMed 30558322), resveratrol under 1% (15333514), quercetin in low single digits (10363620), curcumin (30633561), magnesium oxide at 4% and citrate (34111673, 2407766), calcium citrate and carbonate (2273192). The rest carry a score only, and say so.

A distinction worth keeping: absorbed is not the same as available. Resveratrol is absorbed at roughly 75% and still has systemic bioavailability under 1%, because the gut and liver conjugate it on the way through. Sites that quote a single absorption number for it are quoting the wrong one.

Cost per unit delivered

Shelf price compares badly across forms, because a cheap capsule can be mostly carrier and poorly taken up. This metric divides monthly cost by the amount delivered: daily dose multiplied by the elemental percentage, then by absorption.

It reverses the ranking you get from price: magnesium oxide is the cheapest capsule and one of the most expensive ways to deliver magnesium, because only about 4% of it is absorbed.

Where absorption is measured, the figure is a real cost per absorbed milligram and is labelled "absorbed". Where only a score exists it is labelled "delivered" and is a comparative figure — valid for ranking forms against each other, not to be read as dollars per absorbed milligram.

The unit adapts to the compound: per 100 mg for magnesium, per 1 mg for selenium. Quoting "per 100 mg" for something dosed at 150 mcg is arithmetically correct and practically meaningless.

The metric is skipped where it would mislead — for compounds dosed in IU or CFU rather than milligrams.

Cost, safety rating and popularity

Three numbers that are estimates, stated plainly so you can weigh them accordingly.

• monthly cost is a typical retail figure for a mid-range product at the listed dose — it is not a price we track, and 397 of 427 values are round to the nearest $5, which tells you the precision
• the safety rating from 1 to 10 is an editorial summary of the documented risk records: severity of the listed adverse effects, how far the therapeutic dose sits below the toxic one, and whether the compound is restricted. It is not a regulatory classification
• popularity rank orders compounds into broad bands rather than exact positions; we use it to decide what to show first, never as a recommendation

FDA adverse-event reports

Pulled from CAERS, the FDA system that collects reports about food and supplements, through the openFDA API. Two numbers are shown per compound: how many reports mention it at all, and how many name a supplement containing it as the suspect product.

The gap between those two matters, and it is where most published counts go wrong. openFDA matches conditions at the level of the whole report, not the individual product — so a query for "magnesium AND suspect" also returns a report where magnesium was a concomitant product and the suspect was a hair gel. That is how you get the widely quoted figure of about 1,800 reports for magnesium. We fetch the reports and inspect each product entry: 379 of them actually name a magnesium supplement as the suspect.

We also split out reactions that are not pharmacology. For magnesium the single most common entry is choking, followed by foreign body and dysphagia — people struggling with the tablet, not with the mineral. Those are listed separately rather than mixed into the effect list.

What these numbers are not: rates. Reporting is voluntary and unverified, a report records what someone experienced while taking something rather than what it caused, serious outcomes are far more likely to be reported than mild ones, and counts follow how many people take a thing. Use them to see what gets reported, not how likely it is.

Known limitations

We would rather state these than have you discover them:

• the absorption score is a ranking, not a measured percentage, for every compound except the nine that carry a "Measured" figure
• absorption is dose-dependent — fractional uptake falls from roughly 65% at low intakes to 11% at high ones, while our figures are single values per form
• absorption also varies with food, gut health, age and genetics, none of which is modelled
• the compatibility score covers supplement-to-supplement records only; medication interactions are listed separately and never folded into the number
• "no known interaction" means no documented evidence, not evidence of safety — most pairs have never been studied
• evidence level B or C means the effect is plausible, not established
• adverse-event counts are reports, not rates, and the query caps at 1,000 mentions per compound — where it caps, the figure shown is a floor

What we deliberately do not do

• no star ratings or review counts — we have no reviewers, and inventing them is fraud
• no Product or Offer markup on information pages, and no aggregate ratings anywhere
• no pages generated for combinations nobody searches for
• no claims that a supplement treats, cures or prevents disease

Who maintains this

The database is compiled and maintained by Denis, the site owner. Every figure is either taken from a cited source or derived by a formula published on this page — nothing here rests on personal authority.

To be explicit about the limits: I hold no medical or nutritional qualification. I built this because the answers I wanted — which form actually absorbs, what a combination does, how much reaches the bloodstream — were scattered across product pages that had something to sell. So the site shows the arithmetic and the sources instead of asking you to trust me.

What that means in practice: use these numbers to compare options and to prepare questions for a clinician who can see your full picture. Do not use them as a treatment decision.

Open to review

If you are a physician, pharmacist, registered dietitian or researcher and willing to check a section, write in — you will be credited by name on every page you review, with the review date. That applies whether you confirm the data or tear it apart; a corrected figure is worth more than an unchallenged one.

Corrections

If a dosage, interaction or absorption figure looks wrong, tell us and point to the source — corrections to the database take priority over new features. Each correction is recorded in the compound record with its reason and date.

The largest correction so far was on 2026-08-01. Every PubMed identifier in the database was checked against the PubMed API: of 428 citations, 408 pointed at real papers on unrelated subjects and 9 at nothing at all — the identifiers were keyboard patterns such as 31234567. All 419 were removed and 9 verified ones kept. A "studies indexed" count was removed the same day for the same reason: nobody had counted them.

On 2026-08-02 the absorption figures were relabelled as scores rather than percentages, nine measured values were added with their papers, and the values for CoQ10, resveratrol, quercetin, curcumin and several plant extracts were lowered — they had been ranked among the well absorbed compounds when the literature places them at the bottom.

Write to [email protected]. If you are a clinician or researcher willing to review a section, we will credit you on the pages you review.