How we research coffee.

When we have reviewed a coffee, we say so: our tasting always appears under the heading OUR REVIEWS. We tasted that coffee ourselves and scored it on our five-star rating system. Every page discloses how we got the coffee — purchased by us, sent to us for review, or sponsored — before a word of the review. We never publish a score for a coffee we haven't tasted: a coffee we haven't reviewed says NOT YET REVIEWED BY US, plainly, and shows only outside reviews from online sources until we have.

We research the public web for reviews of each exact coffee by professional publications, recognized reviewers, competition results and retailers. Every outside review is stored with its reviewer, publication, date, original score, original scoring scale, and a link to the original. We never republish copyrighted review text — each summary is written by us, in our own words, from the source. And we never invent a review, a reviewer or a score. On every coffee page these appear only under the heading OUTSIDE REVIEWS FROM ONLINE SOURCES — never mixed into our own. When a source gives no numeric score, we show NO NUMERIC SCORE, never an invented one.

A review of Finca X's Pink Bourbon, 2025 harvest, roasted by Roaster A is a different coffee from Finca X's Pink Bourbon, 2024, roasted by Roaster B. Every potential match is assigned one of five confidence levels:

• EXACT MATCH — same product, producer, farm, variety, process, roaster, and harvest where both state one. The only level that counts toward consensus.
• LIKELY MATCH — nearly everything matches, but one identity field (usually harvest or roaster) is unstated or ambiguous.
• RELATED COFFEE — same farm or producer, but a different lot, harvest, variety, process or roaster. Shown as context, never as a review of this coffee.
• FARM / PRODUCER CONTEXT — discusses the producer or farm generally, not a specific coffee.
• UNRESOLVED — we can't determine whether it's this coffee, so it doesn't count as one.

Every discovered review carries an identity check naming the fields that matched or differed. Low-confidence matches go to a human review queue before anything is published.

A 4/5 user review and an 80/100 professional cupping result are not interchangeable just because both numbers can be rescaled. They may reflect different evaluation methods, samples and expectations. We preserve linked outside reviewers' actual scores in their original scales, identify the source and release where documented, and avoid a synthetic five-star aggregate in our outside-source panels. A mathematical conversion is not a scientifically valid claim of equal quality or equivalent methods.

The Specialty Coffee Association's Coffee Value Assessment separates descriptive sensory traits, affective impressions and information about a coffee rather than reducing all value to one universal number. Read the SCA Coffee Value Assessment framework ↗. This explains evaluation categories; it does not verify our individual coffees.

We treat multiple reviews as supporting context only when their sources are attributable and the coffee's identity — including the release or harvest where relevant — can be matched. Two comments from one publisher are not two independent publishers. A single historical review is valuable evidence but not a consensus. When comparability, sample size or identity is uncertain, we show individual sources and limitations instead of manufacturing a precise combined five-star rating.

Public sentiment summarizes what attributable writers reported, not a laboratory measurement or a score awarded by our reviewers. When there are at least two distinct, linked publishers, we may explain a qualitative editorial interpretation and its limitations. Source counts, individual ratings, lot identity and whether coverage is historical must remain visible. We do not transform words like 'excellent' into our tasting stars, and a producer's marketing page does not qualify as independent praise.

Our research pipeline scans public blogs, video reviews, forums and discussions for each coffee. Only qualifying public sources are counted; sentiment is reported only when the evidence supports it, and each discussion is attributed and linked. We never fabricate internet data, sentiment or sources — when we haven't researched a coffee yet, the page says so.

• TIER 1 — Primary and authoritative: research institutions, variety registries, competition records, producer documentation.
• TIER 2 — Professional coffee media and recognized reviewers.
• TIER 3 — Commercial sources: roasters, retailers, importers.
• TIER 4 — Community: forums, Reddit, social discussion.

We use AI as a research and drafting tool, under human editorial control. AI-written summaries are generated only from real sources we've found and are human-reviewed before display. Structured database updates are automated; Journal articles are always drafted, fact-checked and published by a human. Every illustration on this site is original AI-generated artwork in our sacred-geometry visual language — it is never presented as a photograph, and we never generate fake portraits of real people.

If a producer, farmer, roaster, researcher or reader spots incorrect information, we want to hear it. Corrections enter a review queue, are checked against sources, and are fixed openly — we never silently rewrite significant factual errors. Pages note when they were last updated.

Independence is the whole point: our review scores are ours alone, outside reviews are always attributed, producer and roaster marketing is always labeled as such, and nothing publishes from the automated pipeline without human fact-check. When we highlight a coffee, we always disclose whether that highlight is sponsored — there are no paid placements in disguise. Coffee existed long before the logo on the bag — our job is to keep telling the story of who actually grew it.