How we source our numbers
Last updated 8 September 2026.
Most published grind charts do not say where their numbers came from. This one does, per setting, on every grinder page — and where the answer is "we worked it out", it says that too. This page is the standard behind those labels.
The two labels
From the maker means that number is printed in the grinder's own manual or grind chart, and that document is linked from the grinder's page — usually the PDF itself, hosted here so it cannot quietly disappear. Modelled means we derived it: interpolated along the maker's own published bands where there are some, or from the grinder's dial and burr geometry where there are none.
As of today 33 of 58 grinders have at least one maker-backed setting and every one of those can show you the document. The other 25 are modelled throughout, and their pages say so in a sentence rather than in a footnote.
The rules we adjudicate by
- A maker document wins outright on dial length and on any method it names.
- The maker's own document or spec page beats a retailer, which beats a review. A price-comparison site hosting a maker's PDF is not the maker.
- A maker and a projection are never averaged. Where they disagree we publish the maker's number and record the disagreement with its size. A median across a measurement and a guess is worse than either.
- Nothing is fitted. If a figure has to be derived from our own micron table times somebody else's clicks, it is not cited — it is a fit, and it gets labelled modelled however reasonable it looks.
- No source, no change. A grinder stays uncited rather than gaining a plausible number. That is the failure this whole audit exists to unpick.
- Microns per click a maker states as burr travel is a gap figure, not a particle size. Three separate grinders were wrong this way before we noticed.
- "Looked, found nothing" is a result and gets written down, because a search nobody recorded is a search somebody repeats.
What we have got wrong
Four grinders' settings were published under a maker's name while disagreeing with that maker's own document. Three of the four were wrong. The Niche Zero was the worst: our filter settings were roughly twice as fine as Niche's own published table, on four of the five methods Niche names. All four are corrected and the corrections are dated.
We also published, for months, a set of grind ranges labelled as cited when they were a third party's arithmetic. Twelve grinders carried them. They now say whose they are, on the page.
Tell us we are wrong
"Help us improve" sits on the top-right of every page and the first option is Wrong number — because that is the report worth having and the one that can be checked against a document. If you own the grinder and your dial disagrees with ours, that is the message we most want. Every one is read.
The atlas
Origin and region pages carry a sources footer that states what each source actually backs, rather than a bibliography implying it backs everything on the page. Production figures come from USDA's per-country PSD dataset; the five origins USDA does not carry take a FAOSTAT tonnage, converted and marked as such, with no species split invented for them.