Why so many coffees say “not stated”

Two different silences

There are exactly two reasons a fact is missing here, and the distinction matters enough that the site has separate wording for each.

That second case is not hypothetical, and price is where it is largest. 4,183 coffees here carry no price per kilo, and 2,560 of those are our gap rather than the roaster’s: 2,308 publish a price that our own £10 to £150 band or our repeated-price rule rejects, 95 defeated our reader on a page that does carry one, and 157 we have never read at all. Another 0 are priced in euros, which we do not convert, because a converted price is not a price anyone publishes. Not one of them is a roaster declining to publish a price.

Until recently every one of those pages said “no price published”, which told readers that some of the best-known roasters in Britain hide their prices. They do not. We had simply never read them, and each page now says which of the reasons above applies to it.

What is missing, and how often

Out of 12,793 coffees in the catalogue, with each gap split by whose it is. Most of these are a roaster’s choice about what to print and are not a measurement of anything. The right-hand column is ours.

6,880 of the gaps in that table are ours rather than the roasters’, and almost all of them sit in one row: 6,880 coffees whose roast level arrives from our pipeline as the literal word “unknown”. That is a figure we could not read, not a roaster who declined to print one, and the two had been added together here under a heading that blamed the roaster for both.

One more gap of ours is not in the table, because it is not an absence in the data at all: 225 coffees carry tasting notes we hold and could not sort into any flavour family. The roaster’s words are on the coffee’s page; it is our vocabulary that fell short, so those coffees are missing from the flavour filters on the catalogue and from the flavour pages those filters lead to.

Why we will not fill them in

Every method of filling a gap produces something that reads as a fact and is not one. A coffee called “Colombian Supremo” could have its origin inferred from its name, and that would be right most of the time and wrong often enough to matter. A process could be guessed from tasting notes. Last season’s altitude could be carried forward. All three produce a page that looks more complete and is less true.

So the rule in the code is that an absence has no value at all, and there is no third state that renders as a “no”. A missing organic certification cannot be displayed as “not organic”, because we have no idea. This has a cost: the pages look emptier than a competitor’s, and some of those competitors are filling the same gaps with inference. We would rather be the site whose facts you can check.

It has bitten us the other way too. Two fields arrive from our data pipeline as false on records nobody has ever examined — whether a coffee travelled by sailing ship, and whether its packaging is eco-friendly. Read literally, this site would be telling you that hundreds of roasters do not ship by sail, on the strength of nobody having looked. We treat both as “not stated”, which means a roaster who genuinely does neither is indistinguishable from one nobody asked. That is the right way round: an unfair silence is better than a false accusation.

What this costs you, the reader

Mostly, it means you will find fewer coffees than you expect when you filter hard. A filter for washed Ethiopian coffees traced to a farm, under £50 a kilo, returns a short list — not because few exist, but because few roasters state all four things at once. Filtering on this site is filtering on what roasters chose to publish.

The useful consequence is that where a page here does state something, it came from the roaster’s own words and usually links to the passage it came from. Not always, and the shape of that is worth knowing. We hold 64,420 stored passages against coffee facts. 61% of them are a whole sentence we can put in quote marks; 15% are a fragment, cut to a fixed width by the collector rather than to a sentence, so they are shown between dots and without quote marks; and 24% turned out to be a menu, a breadcrumb or a product grid, which a coffee’s page drops rather than quoting at you. Each page also checks that a passage mentions the fact it is offered as evidence for, so some of the rest are held and not shown.

And 3,443 coffees — 27% of the catalogue — carry no stored passage at all. Every coffee page links out to the roaster’s own listing, which is where its facts were read from, but for these we hold none of their words to show you beside it. The method page sets out how a passage becomes a quotation and when it does not.

Questions