Diarrhoea and Acid Reflux Share Our 'High' Confidence Grade — Their Rates Are More Than 3× Apart

If a side-effect rate on our site carries a "high" confidence grade, does that tell you the effect is common? Or rare? Neither — and two rows in our own table make the point better than an explanation could: diarrhoea and acid reflux carry the identical `high` grade, and their pooled rates are 17.2% and 5.0% — more than three times apart.

The grade is a bucket on distinct source count, nothing else. `confidenceFromSources` takes exactly one input — how many distinct studies state a rate for that effect — and returns one of five labels: `very_low` (≤1 source), `low` (≤3), `moderate` (≤9), `high` (≤24), `very_high` (≥25). It has no way to read the rate itself; the function's own signature only accepts a count. Whatever the pooled percentage turns out to be plays no part in which label an effect gets.

Read as a table — every one of our current eight `high`-grade effects, sorted by rate, each figure computed live by the same function that runs the site:

One badge, eight effects, rates from 5.0% to 17.2% — a 3.4x spread. Source count doesn't order them either: fatigue's 12 sources (the fewest of the eight) sit behind a *higher* rate than acid reflux's 14, and the three effects tied at 24 sources each — diarrhoea, constipation and reduced appetite, the most in this tier — span 17.2% to 9.3%, a 7.9-point spread on an identical source count.

It doesn't stop at the tier boundary. Vomiting carries `very_high` — the grade one full tier above every row in that table, backed by 27 distinct sources — and its pooled rate is 10.3%. Three `high`-grade effects in the table above it — diarrhoea (17.2%), constipation (11.8%) and headache (10.7%) — all sit higher, and the five below it run from 9.3% down to 5.0%. A reader who used the grade to rank "which of these matters more" would get diarrhoea and vomiting backwards, comparing a `very_high`-graded 10.3% against a `high`-graded 17.2% and concluding the wrong one is bigger.

What the grade is actually for

None of this is a defect — it's what the grade was built to measure, stated plainly in a disclosure note our API publishes with every response that carries a pooled rate: `sourceDiversity` "measures SOURCE DIVERSITY, not precision" and is "computed without reference to the confidence interval." What this piece adds is that the same is true of the rate itself — the grade doesn't read that either, by construction. It answers one question honestly: how many independent studies have weighed in? It was never built to answer, and structurally cannot answer, "how common is this?" or "how big a deal is this?" — those are what `ratePct` and the 95% interval are for, and they have to be read directly, not inferred from the badge sitting next to them.

The practical rule, then, for reading any confidence or evidence grade attached to a medical rate — ours or anyone's: it tells you how much evidence stands behind the number, never the size of the number. A `high`-graded 5% and a `high`-graded 17% are both exactly as well-evidenced as their shared grade says; neither is bigger or smaller because of it.

Reproduce it

Every figure above was computed on 20 September 2026 by the same `pooledClinicalEstimate`/`confidenceFromSources` functions that run live on magistra.health (methodology), cross-checked against the site-wide total published in this week's composition piece (1,142 stated rates, 31 distinct studies — unchanged since). Read any effect's current grade, rate and source count at `GET /api/data?q=effect&id=` — the corpus grows daily, so a grade or a rate quoted here can move; the endpoint is the source of truth, not this article.

The full dataset — 15 effects, both tracks, every source with its URL, CC BY 4.0 — is at magistra.health/en/data-api.