Grindograph

An open knowledge graph · CC0

Every coffee, photographed.

A catalogue of roasted coffee where every record is grounded in a photograph of the bag it came from — searchable, linkable, and public domain. It is nowhere near complete. That is the interesting part.

coffees
roasters
times a machine made something up

Missing yours?

There is no scrape behind this and there is no licence deal. Every record here arrived because somebody photographed a bag — so the way it grows is one bag at a time, and it takes about a minute.

You do not have to type anything: upload the photo and the machine reads the bag back to you. But it is more fun if you read it first, because then you find out which of you got it right. Read a bag, beat the machine.

in

one photo of the label

you read it, then the machine does
out

a public-domain record, per bag

About a minute, nothing to sign up for, and every bag you add joins a public-domain graph anybody can query.

How a round works

  1. 1

    Snap the bag

    One photo of the side with the label is enough. Location data is stripped before anything is stored.

  2. 2

    Write what you see

    Five fields, none of them required. Type only what is printed — don't look the coffee up. A blank beats a good guess.

  3. 3

    See the scorecard

    Now the machine reads the same photo, and you find out which of you got what right.

The one rule

You never see the machine's answer before you write yours. The form does not pre-fill from an extractor and it never will. That is what makes the round a fair contest — and, not by coincidence, it is also the only thing that makes the resulting data worth having. Pre-fill the form and you are measuring agreement with the model rather than what the bag says, which is worse than no data, because it looks like data.

So it is enforced in code rather than trusted as a convention: there is deliberately no endpoint that hands out a reading on its own, and a test pins the list so adding one has to be somebody's deliberate decision.

What you're building

The scorecard is the fun part. This is the part that outlasts it — and the reason the round is designed the way it is.

A coffee, not a photo album

Two people photographing the same bag used to be two unrelated records. Now they meet at one coffee, under one roaster, at a permanent address anybody can link to — and every claim on it still points back at the photograph it came from.

A vocabulary counted, not argued

Bags say "Nat. Anaerobic", "anaerobic natural" and "Anaeróbico" and mean one thing. Collecting the ways they spell it is the work; being able to search for all of them at once is what it buys. The tally is public.

Public domain, permanently

Photos, readings, the graph, the generated shape, and what the models guessed — as versioned releases, and as data at every IRI in it. No attribution required, no strings. The grant is recorded per bag, so it cannot be revoked later.

The longer version — the three marks, the ordering guarantee, and how disagreeing with the form is itself a contribution — is on the why page.