BRENNEN

Industries / Food, farming and forestry / Food and beverage processing / Coffee roasting

Tasting and quality

Judging coffee by drinking it: roasting small samples, tasting them to a set method called cupping, describing their flavor in shared terms, and checking that each lot and each roasted batch tastes as it should.

6

Milestones achieved, from 1880s to 2024.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1880s

    Judged by the cup

    Before 1900, nearly all coffee bought and sold in the United States was judged by how the green or roasted bean looked, by William Ukers's 1922 account. In San Francisco, Clarence E. Bickford, then working for a broker named Hockhofler, found the cup qualities of high-grown Central American coffees; once roasters began tasting, Antigua and other high-grown coffees that had been marked down for their small beans sold at a premium. By the 1920s, Ukers wrote, every progressive buyer had a sample roaster and ran blind cup tests with weighed coffee, smelling the crust of grounds before breaking it and spitting out each sip.

    LessonBean size and color did not predict taste, so the price of a lot moved once buyers tasted it.

    gutenberg.pglaf.org

    Credit

    Clarence E. Bickford

  2. 2004

    One form for cuppers

    The SCAA published a cupping protocol and form in 2004, under its longtime director Ted Lingle. Daily Coffee News calls it the first widely adopted way to evaluate coffee in a quantifiable, structured way, used mostly by roasters and green coffee buyers, and companies still use private versions of it.

    dailycoffeenews.com

    Credit

    SCAA; Ted Lingle

  3. 2004

    A license to cup

    The Coffee Quality Institute (CQI) introduced its Q Coffee System in 2004, a cycle of training, examination and certification that licenses tasters as Q Graders. By 2024 CQI counted ten thousand Q Graders working along the coffee trade, and in 2025 it stopped running the program.

    coffeeinstitute.org

    Credit

    Coffee Quality Institute

  4. 2016

    Scores that did not match

    Mohammed Worku, Luc Duchateau and Pascal Boeckx compared cupping of 30 Ethiopian coffees by nine cuppers at three cupping centers of the Ethiopia Commodity Exchange (ECX). Total scores differed by about 3 points on a 100-point scale, close enough for grading, but on 70 other samples the exporter's (ECX) scores and the importer's (EFICO) scores did not correlate at all. In 2020 Davide Giacalone and colleagues found low agreement among expert cuppers on "clean cup", although each cupper repeated their own scores closely.

    LessonA cup score holds within one lab but cannot be taken as interchangeable between a seller's lab and a buyer's.

    api.crossref.orgapi.crossref.org

    Credit

    Worku, Duchateau and Boeckx; Giacalone, Steen, Alstrup and Münchow

  5. 2016

    A shared flavor vocabulary

    World Coffee Research and the Sensory Analysis Center at Kansas State University published a coffee sensory lexicon: 110 flavor, aroma and texture attributes, each with a reference for measuring its intensity, drawn from more than 100 coffees from 14 countries. A UC Davis team then had 72 experts sort the lexicon's terms to build a new Coffee Taster's Flavor Wheel, the first redesign of the SCAA's wheel in about 20 years. Lexicon 2.0 followed in October 2017 with references for 24 attributes that can be bought worldwide.

    worldcoffeeresearch.orgapi.crossref.orgapi.crossref.org

    Credit

    World Coffee Research; Edgar Chambers and colleagues (Kansas State University); Molly Spencer, Emma Sage, Martin Velez and Jean-Xavier Guinard (UC Davis)

  6. 2024

    Describing split from liking

    On 4 November 2024 the Specialty Coffee Association (SCA) adopted three standards, together called the Coffee Value Assessment (CVA), that supersede the 2004 protocol and form: SCA-102 for preparing samples, SCA-103 for describing a coffee's attributes and their intensities, and SCA-104 for rating the impression of quality. The SCA says the old form mixed descriptive tests with value judgments, which sensory science advises against, and that a 2021 survey found cuppers split on whether the old form was objective.

    sca.coffeesca.coffee

    Credit

    Specialty Coffee Association, with research by World Coffee Research and the Coffee Science Foundation (Jorge Berny and Mario Fernández-Alduenda)

Watch for

  1. Cup scores from a scan

    ProfilePrint, a Singapore company, showed its Mini Beluga analyzer in April 2026 and launched it in June with a cheaper service tier, ProfilePrint Lite. The 1.3 kg device takes a "molecular fingerprint" of a sample in minutes and sends it to ProfilePrint's AI, which, drawing on more than 30,000 specialty coffee samples assessed by Q Graders, predicts SCA cupping scores and flavor notes.

    Achieved whenan independent panel blind-cups a set of lots that the device has scanned and publishes how far the predicted scores fall from its own.

    dailycoffeenews.com

    Claimed by

    ProfilePrint, June 2026

  2. A scorecard that explains itself

    Jean-Xavier Guinard and colleagues at UC Davis published a test of their Coffee Cuality method, which pairs a 100-point quality rating with "just about right" ratings, a checklist of descriptors and comments, so that a low score comes with its reasons. In their study 56 experts rated 12 coffees with their usual method and with Coffee Cuality; the authors report the same overall ranking and more significant differences between coffees, although the order of the top three differs between the methods in their own table.

    Achieved whena lab or competition outside UC Davis uses the method and publishes how well its cuppers agree.

    pmc.ncbi.nlm.nih.gov

    Claimed by

    Guinard, Lim, Elliott and Cotter (UC Davis), February 2026

Open

  1. 01

    AI cup scores, blind-tested

    A blind test of tools that predict cup scores from instruments, scored against a panel of licensed cuppers. Take 100 lots, have at least five cuppers score each blind, run each tool on the same green or roasted samples, and publish each tool's average error beside the spread between the cuppers themselves.

    The brief
    Credit · unclaimed
  2. 02

    Ring-testing the CVA

    A multi-lab check of the SCA's new Coffee Value Assessment. Send the same 12 blind-coded coffees, roasted once by one roaster, to six labs in at least three countries; have each lab cup them twice with the CVA descriptive and affective forms; publish how well labs and cuppers agree on each attribute.

    The brief
    Credit · unclaimed

Open 01 · the brief

AI cup scores, blind-tested

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A blind test of tools that predict cup scores from instruments, scored against a panel of licensed cuppers. Take 100 lots, have at least five cuppers score each blind, run each tool on the same green or roasted samples, and publish each tool's average error beside the spread between the cuppers themselves.

Why it is still open

ProfilePrint says its predictions draw on more than 30,000 Q-graded samples, and Demetria sorts beans by predicted flavor, but searches in November 2026 found no independent blind test of either. Academic models have predicted cup scores from near-infrared spectra since at least 2016, when Tolessa, Rademaker, De Baets and Boeckx reported a prediction error (root mean square) of 1.04 on the total cup score for 20 of 86 green arabica samples, checked against the same study's cupping scores rather than a blind outside panel. The open dataset of near-infrared spectra and sensory scores from Universidad Surcolombiana (version 4, November 2025) comes with code to train models but states no accuracy. The gap between cuppers is itself about 3 points on a 100-point scale in the Ethiopian study, so a tool has to be judged against that spread.

dailycoffeenews.comdata.mendeley.comapi.crossref.orgebi.ac.uk

Built with what exists

A benchtop or handheld near-infrared spectrometer, the vendors' own devices where they agree to take part, partial least squares regression or gradient boosting in scikit-learn trained on the Surcolombiana data, and the SCA's free CVA forms for the panel.

  1. First step

    Write the test protocol and ask two importers to supply 30 lots with the cupping results they already hold.

  2. Done looks like

    For 100 lots, each tool's mean absolute error against the panel's average score, with its 95% confidence interval, published beside the cuppers' own average distance from that average, and the data released.

Open 02 · the brief

Ring-testing the CVA

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A multi-lab check of the SCA's new Coffee Value Assessment. Send the same 12 blind-coded coffees, roasted once by one roaster, to six labs in at least three countries; have each lab cup them twice with the CVA descriptive and affective forms; publish how well labs and cuppers agree on each attribute.

Why it is still open

The UC Davis team behind Coffee Cuality wrote in February 2026 that the CVA shares elements with their method and called for studies comparing them; their paper reports no reliability data for the CVA. The SCA's published work on its new system describes how cuppers used the old form, and searches in November 2026 found no inter-laboratory agreement figures for the CVA.

pmc.ncbi.nlm.nih.govsca.coffee

Built with what exists

The SCA's free CVA forms, one sample roaster and a color meter to keep the roast the same, a shared spreadsheet, and R packages such as irr and psych to compute intraclass correlation, a standard measure of how closely different raters agree.

  1. First step

    Recruit six labs through an importer or a coffee school and agree on a shipping and roasting date.

  2. Done looks like

    For every CVA attribute, an intraclass correlation between labs and within each cupper, with 95% confidence intervals, published with the raw forms, and set beside the 2016 Ethiopian figures for the old form.