# Coffee: research notes

Scope: roasted coffee sold packaged as whole bean or ground. Capsules and instant appear only as format values.
Run: 2026-10-19. Seed persona: a cycling mechanic (compatibility, standards, the part that has to match). Seed example: bedsheets weave vs thread count.

How quotes are labelled. "verbatim" means WebFetch returned the sentence as a quote. "fetched excerpt" means WebFetch returned a model paraphrase of the page. "search summary" means only a search-result digest was available because the page itself was blocked (403/500/404) or failed. Numbers that rest on search summaries alone are not used as facts in category.json; they appear as judgements with confidence low or medium.

## Seed

1. The coffee equivalent of "weave vs thread count". The number everyone advertises is roast level (and, in the UK, a "strength" 1-5 printed on the bag, which is roast darkness, not caffeine). The claim everyone advertises is "100% Arabica", which almost every specialty bag satisfies, so it separates nothing above the supermarket tier. The facet that actually changes how the cup tastes, the way weave changes how a sheet sleeps, is the processing method (washed vs natural vs honey vs anaerobic) and, for whether the bag tastes like anything at all, the roast date. Candidate: process as a top split; roast date as the olive-oil harvest-date analogue.

2. What a cycling mechanic would notice first: the part that has to match. In coffee that is grind-for-method. Pre-ground coffee is ground for one brewer, and an espresso grind in a French press (or vice versa) ruins the cup regardless of what was in the bag. Second match: roast level to brewer, because light roasts are hard to extract on a home espresso machine and dark roasts over-extract on it. Third, the literal compatibility standard: capsule system (Nespresso Original vs Vertuo vs K-Cup) which is a hard filter in the pod sub-set. Fourth, the packaging standard: a one-way degassing valve is the sign the roaster expects the coffee to be fresh. Candidates: grind_for (hard filter on pre-ground), roast-to-method mismatch as a trade-off, capsule_system as a filter in the pod format only, valve as a low-cost freshness signal.

Survived the data check: grind_for (listing variant selectors carry it: Peet's "Whole Bean, Drip, French Press, Espresso"; Trade lists eleven grind options), roast-to-brewer as a trade-off (two sources on light-roast espresso), capsule system as a filter inside the capsule format (two sources on Vertuo lock-in), freshness policy and roast date as the top expert split (five sources). Process survived as a filter but not as a split: a novice cannot answer "washed or natural" cold, so flavor family (built from the SCA wheel's first tier) is the split and process sits under it.

Dropped: the degassing valve. No listing states it, it is not reliably visible in photos, and nearly every roaster that prints a roast date also uses one, so it adds nothing over the freshness policy. Also dropped "100% Arabica" as a facet in its own right; it is folded into species_blend where the useful signal is the opposite one (robusta share in espresso blends).

## Sources

### Standards body and trade

- sca_flavor_wheel_steps. SCA, "How to Use the Coffee Taster's Flavor Wheel in Eight Steps". https://sca.coffee/sca-news/how-to-use-the-flavor-wheel-in-eight-steps-ab2hy . Page fetch failed; search summary only: the wheel "is based on the World Coffee Research Sensory Lexicon" and is "a collaborative effort by the Specialty Coffee Association of America and World Coffee Research". Supports: the wheel is the SCA's standard sensory picture. Confidence medium (claim is uncontroversial and repeated verbatim by secondary sources below).
- dailycoffeenews_wheel_2016. Daily Coffee News, 19 Jan 2016, "SCAA Unveils a Whole New Coffee Taster's Flavor Wheel". https://dailycoffeenews.com/2016/01/19/scaa-unveils-a-whole-new-coffee-tasters-flavor-wheel/ . Verbatim: "The Lexicon currently contains 110 attributes — anything from pineapple or honey to cardboard or mold". Fetched excerpt: the WCR Sensory Lexicon "is the foundation for" the wheel. Supports: 110 attributes; lexicon basis. Confidence high.
- coffeester_wheel. Coffeester, "Coffee Taster's Flavor Wheel: History, Structure & How to Use It". https://www.coffeester.com/knowledge/sensory/flavor-wheel . Verbatim: "Fruity, Floral, Sweet, Nutty/Cocoa, Spices, Roasted, Green/Vegetative, Sour/Fermented, and Other." and "The Flavor Wheel is organized as a series of concentric rings, and the correct way to use it is to read from the center outward—moving from broad, general impressions toward increasingly specific, precise descriptors." Supports: nine first-tier categories; centre-outward reading. Confidence high.
- gevi_wheel. Gevi, "Coffee Flavor Wheel Explained". https://gevi.com/blogs/coffee-knowledge/coffee-flavor-wheel-guide . Verbatim: "Fruity, Floral, Sweet, Nutty/Cocoa, Spices, Roasted, Green/Vegetative, Sour/Fermented, and Other (which includes categories like Papery/Musty)." Supports: nine first-tier categories (second source). Confidence high.
- cloudcity_agtron. Cloud City Coffee, "Agtron Roast Value". https://www.cloudcitycoffee.com/blogs/notes-from-the-roaster/agtron-roast-value . Verbatim: "Agtron's scale ranges between 25 and 95, with the darker coffees corresponding to the lower numbers." and "Light: Agtron 70-85 / Medium: Agtron 60-69 / Dark: Agtron 50-59". Supports: scale direction and range; one roaster's bands. Confidence high for the scale, medium for the bands (disagrees with Coffee Review, below).
- coffeereview_roast_definitions. Coffee Review, "Understanding roast definitions for coffee". https://www.coffeereview.com/roast-definitions/ . Verbatim: "a reading of 55/68 would describe a coffee with an external, whole-bean M-Basic reading of 55 and a ground reading of 68" and "The number we use in assigning names for roast levels reflects an average of the two values indicated for each coffee, whole-bean and ground." Fetched excerpt table: Light > 80; Medium-Light 70-80; Medium 60-70; Medium-Dark 50-60; Dark 40-50; Very Dark 25-40; Extreme-Dark < 25. Supports: the slash format; a reviewer's bands. Confidence high.
- agtron_disc_search. Search summary (Crowd Roaster, Cafeinados Hub, Green Coffee Collective glossary): "In the 1970s, the Specialty Coffee Association of Americans (SCAA) developed together with Agtron Inc. a classification system made up of 8 colored discs ... with numbers varying between 25 and 95 (#25 / #35 / #45 / #55 / #65 / #75 / #85 / #95), with 95 being very light and 25 being very dark." Supports: origin of the 25-95 disc scale. Confidence medium (search summary). Not cited in the JSON on its own.
- wcr_geisha. World Coffee Research variety catalog, "Geisha (Panama)". https://varieties.worldcoffeeresearch.org/varieties/geisha-panama . Fetched excerpt: quality potential at high altitude "Exceptional"; optimal altitude "High" (>1600 m for 5°N-5°S, >1300 m for 5-15°, >1000 m for >15°); yield potential "Low"; lineage "Ethiopian landrace". Supports: variety names are a catalogued standard; altitude is latitude-relative. Confidence high.
- dailycoffeenews_cva. Daily Coffee News, 5 Nov 2024, "SCA Officially Adopts New Cupping Forms and Standards". https://dailycoffeenews.com/2024/11/05/sca-officially-adopts-new-cupping-forms-and-standards/ . Fetched excerpt: adopted SCA Standard 102-2024 "Sample Preparation and Tasting Mechanics", 103-2024 "Descriptive Assessment", 104-2024 "Affective Assessment", superseding "the SCA's coffee cupping form and protocol that were first published by the Specialty Coffee Association of America (SCAA) in 2004". Supports: the 2004 cupping form was replaced in November 2024. Confidence high.
- mtpak_cva. MTPak Coffee, Sept 2023, "Coffee Value Assessment: Understanding the new SCA protocol". https://mtpak.coffee/2023/09/coffee-value-assessment-sca-new-protocol/ . Verbatim fragments for the four assessments: Physical "The beans' colour, moisture content, defects, and size." Descriptive "The coffee's fragrance, aroma, mouthfeel, flavour, acidity, sweetness, and aftertaste." Affective "The impression of quality, and hedonic preference." Extrinsic "informational or symbolic attributes that are considered to add value to the coffee." Supports: CVA has four assessments. Confidence high.
- wikipedia_specialty. Wikipedia, "Specialty coffee". https://en.wikipedia.org/wiki/Specialty_coffee . Verbatim: "The widely accepted definition of specialty coffee is coffee scoring 80 points or above on the 100-point scale used on the Specialty Coffee Association Cupping form." and "to be considered specialty a coffee must have 0 to 5 defects every 350 g (12 ounces) of milled beans." and "Coffee scoring from 90 to 100 is graded Outstanding, coffee that scores 85–89.99 is graded Excellent, while coffee scoring 80–84.99 is graded Very Good." Supports: the 80-point line and bands. Confidence high.
- beanbox_specialty. Bean Box, "What Is Specialty Coffee? The 80-Point Score Explained". https://beanbox.com/blog/what-is-specialty-coffee . Verbatim: "Specialty means 80+ on the SCA's 100-point cupping scale, no more than 5 primary defects per 350g sample, judged by certified Q Graders." and "Estimates suggest only about 5 to 10 percent of the world's coffee clears the bar." Supports: 80-point line (second source); share of world coffee. Confidence high for the line, medium for the share (Wikipedia-derived sources say 3-5%).

### Regulator

- fda_date_labels. US FDA, "How to Cut Food Waste and Maintain Food Safety". https://www.fda.gov/food/consumers/how-cut-food-waste-and-maintain-food-safety . Verbatim: "Except for infant formula, manufacturers are not required by Federal law or regulation to place quality-based date labels on packaged food." Supports: no US rule requires any date on coffee, so a roast date is voluntary. Confidence high. (The USDA FSIS page and PDF returned 403 on three attempts and are no longer cited.)
- eurlex_1169. Regulation (EU) No 1169/2011. https://eur-lex.europa.eu/eli/reg/2011/1169/oj/eng . Search summary: mandatory particulars include "the date of minimum durability"; "the date shall be preceded by the words 'Best before …'". Supports: EU coffee must carry a best-before; nothing requires a roast date. Confidence low: EUR-Lex (HTML, consolidated HTML and OJ PDF) and the legislation.gov.uk mirror all returned empty to WebFetch and to curl in this run, so the regulation text was never read; the claim rests on the search summary and eClarion. Recorded as a low-confidence judgement in category.json.
- eclarion_dates. eClarion, "Best Before and Use By Dates: Rules for Food Labels". https://www.eclarion.com/en/support/resources/best-before-use-by-date-rules/ . Verbatim: "if the product is microbiologically perishable and could pose a health risk within a short period, a use by date is required. In all other cases, a best before date is sufficient." Fetched excerpt: no mention of a production or roasting date requirement outside frozen meat and fish. Supports: coffee carries best-before, not use-by; roast date is not a legal field. Confidence medium.
- wikipedia_decaf. Wikipedia, "Decaffeination". https://en.wikipedia.org/wiki/Decaffeination . Verbatim: "A caffeine content reduction of at least 97% is required under United States FDA standards." and "There is less than 0.1% caffeine in decaffeinated coffee and less than 0.3% in decaffeinated instant coffee in Canada." and "The 473 ml (16 ounce) cups of coffee samples contained caffeine in the range of 8.6 mg to 13.9 mg." Methods listed: dichloromethane or ethyl acetate solvents, Swiss Water, triglyceride, supercritical CO2. Supports: 97% US line; residual caffeine; methods. Confidence high.
- frequentcoffee_decaf. Frequent Coffee, "Decaf Deep Dive: U.S. vs EU Standards". https://frequentcoffee.com/blogs/journal/decaf-deep-dive-us-vs-eu-decaf-standards . Verbatim: "In the U.S., for coffee to be labeled as 'decaffeinated,' it must have at least 97 percent of the original caffeine content removed from the green coffee beans." and "The EU defines decaf coffee as having no more than 0.1 percent caffeine by weight in roasted coffee, or 0.3 percent for green coffee beans." and "The EU measures the amount of caffeine remaining by weight, not as a percentage of what was removed." Supports: US 97%, EU 0.1% roasted, and that the two are not comparable. Confidence high. Note: a search summary (Driftaway, Tasting Table) adds that the 97% figure "is not codified in official FDA regulations" and is an industry expectation. Disagreement recorded; see judgements.

### Spec sheets (roasters and brands)

- onyx_geometry. Onyx Coffee Lab, "Geometry". https://onyxcoffeelab.com/products/geometry . Fetched excerpt with quoted labels: origin "Colombia, Ethiopia"; process "Washed"; variety "Mixed"; elevation "1950 MASL"; cup score "87"; roast level "Light" with "Ground Agtron: #130"; caffeine "215mg / 12oz"; grind "725µm" for filter; sizes 2oz, 10oz, 2lbs, 5lbs; transparency "Green Cost: $5.44", "Market Price: $4.04", "Pay Structure Rating: B". Supports: what a top specialty roaster publishes per coffee. Confidence high.
- onyx_gesha. Onyx Coffee Lab, "Mexico Finca Santa Cruz Gesha". https://onyxcoffeelab.com/products/mexico-finca-santa-cruz-gesha . Fetched excerpt: "Finca Santa Cruz"; Chiapas; "1650 MASL"; "Geisha"; "Washed"; cup score "86.75"; "Light" with "Ground Agtron: #130"; harvest "April"; grind "522µm"; transparency "Green Cost: $22.75", "Lot Size: 3800LBS". Supports: farm, variety, harvest month and cup score are publishable fields; Agtron on the gourmet scale runs above 95. Confidence high.
- onyx_collection. Onyx Coffee Lab coffee collection filters. https://onyxcoffeelab.com/collections/coffee . Fetched excerpt, quoted filter labels: "Roast Level" (Darker/Lighter), "Origin", "Caffeine" (Decaf, Full, Partial), "Profile" (Traditional/Modern), "Process" (Anaerobic Natural, Dry Washed, Innoculated, Natural, Red Honey, Washed), "Variety" (Batian, Bourbon, Catuai, Caturra, Chiroso, ..., Gesha, ..., SL28, SL34, Typica, ...). Supports: the facets a specialty roaster lets buyers filter on. Confidence high.
- counterculture_collection. Counter Culture Coffee collection. https://counterculturecoffee.com/collections/coffee . Fetched excerpt: card shows "Single-Origin | San Marcos, Guatemala", tasting notes "apple | simple syrup | cocoa", "light roast". Filters: Roast Level (Light, Medium-Light, Medium, Medium-Dark, Dark), Coffee Type (Blend, Single-Origin, Decaf), Flavor Profile (Fruity, Chocolatey, Earthy, Floral, Nutty), Origin (continent), Processing Method (Washed, Natural Sundried, Honey, Anaerobic), Certification. No elevation, variety or roast date on the cards. Supports: flavor-profile buckets are already a retail filter; process is a filter. Confidence high.
- timwendelboe_tatmara. Tim Wendelboe, "Tatmara". https://timwendelboe.no/products/tatmara . Fetched excerpt, quoted labels: Producer "Negussie Tadesse"; Region "Dakiti, Kaffa"; Country "Ethiopia"; Variety "74110, 74112"; Process "Washed"; Harvest "December 2025"; tasting notes "Citrus, floral & stone fruits"; 250 g. No altitude, roast style or roast date shown. Supports: harvest month is a published field at the top end; variety codes appear raw. Confidence high.
- drinktrade_78blend. Trade Coffee, "78 Blend" (roasted by Roasted Record). https://www.drinktrade.com/products/the-78-blend . Fetched excerpt: "Medium-Dark roast"; notes "chocolate, caramel and roasty"; process "Natural/Dry Processed, Washed"; origin "Colombia, Brazil"; variety "Mundo Novo, Red Catuai, Yellow Catuai, Colombia"; elevation "900-1650m"; grind options "Whole Bean, Aeropress, Automatic Drip, Chemex, Cold Brew, Espresso, French Press, Moka Pot, Percolator, Pour Over, Refillable Pods"; sizes "Standard (10.93 oz, ~20 cups)", "Big Bag (32 oz, ~60 cups)". Supports: grind-for is a variant selector; cups-per-bag at about 15.5 g per cup is how a retailer states it; marketplace listings carry elevation and variety. Confidence high.
- drinktrade_freshness. Trade Coffee, "Coffee freshness explained". https://www.drinktrade.com/blogs/education/coffee-freshness-explained-what-roast-date-actually-means . Verbatim: "Most specialty coffee tastes best between five and 21 days after the roast date." and "A best-by date is an arbitrary window — often six months to a year — that does not account for the rapid loss of volatile oils." and "Coffee found on grocery store shelves often lacks a roast date, showing only a best-by date instead. Because of the traditional distribution model, these beans may sit in warehouses or on shelves for months after roasting." A second fetch after the blind check returned the page's own timeline verbatim: "Days one through four: The coffee is in its heaviest degassing phase."; "Days five to 14: This is the peak window for most brewing methods."; "Days 15 to 30: The coffee is still fresh, though the more delicate floral and fruity aromatics may begin to fade."; "Beyond 30 days: The beans begin to oxidize more rapidly." So the page carries both 5-21 (text) and 5-14 (timeline). Search summary adds Trade ships "within 48 hours" of roasting. Supports: peak window as a range; best-by horizon; grocery coffee lacks a roast date. Confidence high.
- peets_major_dickasons. Peet's Coffee, "Major Dickason's Blend". https://www.peets.com/products/major-dickasons-blend . Fetched excerpt: roast "75 out of 100" light-to-dark; brightness "56 out of 100"; body "75 out of 100"; grind options "Whole Bean, Drip, French Press, Espresso"; origin "Americas, Indo-Pacific"; process "Washed and Wet-Hulled"; "roasted Monday-Friday", "Sealed immediately after roast"; no roast date, variety, elevation or caffeine. Supports: mass-premium brands publish 0-100 house scales, grind variants and a vague origin, and describe freshness as a policy rather than a date. Confidence high.
- flavia_lavazza_super_crema. Flavia (Lavazza) product page, "Lavazza Super Crema". https://www.flavia.com/products/lavazza-super-crema . Fetched excerpt: "7/10 intensity"; "medium roast"; "blend of Arabica and Robusta coffees"; "notes of roasted hazelnut"; 9 g per pack. Search summary (thecoffeelocator, beanbrewdigest) adds "approximately 60% Arabica / 40% Robusta". Supports: intensity scales (1-10) and Arabica/Robusta share are listing fields on Italian-style blends. Confidence high for intensity, medium for the 60/40 share.
- taylors_listings. Taylors of Harrogate Rich Italian across UK grocers (mcgrocer, Sainsbury's, Tesco). https://mcgrocer.com/products/taylors-rich-italian-ground-coffee . All pages 403; search summary: "medium roast ... strength rating of 4", "100% Arabica", "Rainforest Alliance Certified", "Ideal for Cafetiere or Filter". Supports: UK supermarket listings carry a strength number and a brew suggestion, not a roast date. Confidence medium (search summary).
- amazon_peets_listing. Amazon, Peet's Major Dickason's, ASIN B000LKZB6O. https://www.amazon.com/dp/B000LKZB6O . Two fetches returned HTTP 500. Not usable. Recorded so the gap is visible: Amazon field presence (Roast Level, Item Form, Caffeine Content, Flavor) is known from general experience but was not verified in this run.

### Enthusiast community and primers

- roastedrecord_agtron. Roasted Record, "What 'Light Roast' Actually Means". https://www.roastedrecord.com/blogs/news/agtron-roast-levels-explained . Verbatim: "Very roughly: mid-70s and up reads light, the 60s read medium, the 50s medium-dark, and the 40s and below are properly dark territory." and "'Light,' 'medium,' and 'dark' are house dialects, not standards — one roaster's medium is another's French roast." Supports: roast words are not comparable across roasters; a third set of Agtron bands. Confidence high.
- monastery_rest. Monastery Coffee, "How Long After Roasting Is Coffee Best?". https://monastery.coffee/blogs/news/how-long-after-roasting-is-coffee-best . Fetched excerpt (the blind check found these do not match the page word for word today; treat as paraphrase): "For espresso, a degassing period of 5-7 days is recommended to avoid excess gas"; "Pour-over or drip coffee benefits from waiting 4-10 days, depending on the roast level"; "Light roasts need a longer degassing period of 7-10 days to achieve peak flavour, while dark roasts are generally ready within 3-5 days"; "Most coffee achieves peak flavour within 3-14 days after roasting. However, it starts losing quality after 2-4 weeks". Supports: rest windows; light roasts rest longer. Confidence high.
- simonbearns_rest. Simon & Bearns, "Roast Date vs. Best Before Date". https://simonandbearns.coffee/en/blogs/kaffeeblog/roast-date-vs-best-before-date-what-really-matters-for-coffee . Verbatim: espresso "optimal after 7–21 days" and "stable for approx. 6–8 weeks"; filter "optimal after 3–14 days", good "for approx. 4–6 weeks"; "Darker roasts usually degas faster, lighter roasts slower". Supports: rest windows (second source); the 6-8 week horizon. Confidence high. Disagrees with Monastery on the espresso start (7 vs 5 days) and end (21 vs 7 days); both agree light roasts need longer.
- homecafe_light_espresso. Home Cafe by Charlie, "How to Dial In Light Espresso". https://homecafebycharlie.com/how-to-dial-in-light-espresso/ . Verbatim: "the higher the elevation, the denser the coffee bean, and so more heat and finer grind you'll use to extract all that goodness properly."; "a usual range for espresso is around 90-96°C, or 194-204°F"; "For espresso it's usually best to use beans that are between 7 days and around 40 days off the roast date". Supports: light and high-grown beans are denser and harder to extract; a 7-40 day espresso window. Confidence high.
- seattlecoffeegear_light_espresso. Seattle Coffee Gear, "Working with Light Roast Espresso". https://www.seattlecoffeegear.com/blogs/learning-center/working-with-light-roast-espresso . Verbatim: "Roasting breaks down the physical structure of coffee beans, making soluble compounds easier for water to extract. Lighter roasts have undergone less of this breakdown."; target "201 to 204°F for light roasts"; "Start at 1:2.5 (18g in, 45g out) and work toward 1:3". Supports: light roasts need hotter water, finer grind and longer ratios on espresso. Confidence high.
- wbur_hoffmann. WBUR Here & Now, 24 Oct 2025, James Hoffmann interview, "How to make great coffee at home". https://www.wbur.org/hereandnow/2025/10/24/how-to-make-coffee-at-home . Verbatim (Hoffmann): "Coffee is fresh food"; coffee is "perfectly good to drink for a very, very long time, it's not delicious for a very, very long time"; buy "within two, three, maybe four weeks of it being roasted"; "grocery stores can have older coffee on the shelves"; use whole beans "within another three or four weeks maximum"; "you'll see a real big drop in flavor within 24 to 48 hours after you grind coffee"; look for coffee "from a distinct place" with "traceability". Supports: buy within 2-4 weeks of roast; ground coffee fades in 1-2 days; traceability as a quality tell. Confidence high.
- headcount_shelf_life. Headcount Coffee, "The Real Shelf Life of Whole Beans vs. Ground Coffee". https://www.headcountcoffee.com/blogs/coffee-news/the-real-shelf-life-of-whole-beans-vs-ground-coffee . Verbatim: "Whole beans remain at peak flavor for about two to four weeks after roasting when stored properly"; ground coffee "Its peak is measured in days, not weeks. Within 24 hours, you'll already notice reduced aroma."; "Grinding increases surface area by more than ten thousand percent." Supports: whole bean vs ground freshness. Confidence high.
- coffeetower_staling. Coffee Tower, "Coffee: Oxidation and Staling". https://coffee.towerofrecords.com/coffee/oxidation-staling . Verbatim: "Ground coffee staling begins within hours of grinding through lipid oxidation; whole beans lose approximately 50% of primary volatile aroma compounds within 2 weeks of roasting without vacuum sealing."; "CO₂ trapped in roasted coffee acts as a carrier for volatile aroma compounds. As CO₂ degasses, it transports aromatics out of the bean matrix."; cites Marin et al. (2008), "Stability of roasted coffee during storage", Food Chemistry. Supports: staling mechanism; whole bean loses half its volatiles in two weeks. Confidence medium (secondary page citing a paper I did not open).
- tastingtable_roastdate. Tasting Table, "The Roast Date On Your Bag Of Coffee Isn't The Same As A Best By Date". https://www.tastingtable.com/1165590/the-roast-date-on-your-bag-of-coffee-isnt-the-same-as-a-best-by-date/ . Verbatim: "A coffee's roast date signifies the date on which the beans were roasted"; mass-produced coffees such as Folgers display only a best-by date, "giving you absolutely no indication of when your coffee was actually roasted". Supports: mass-market bags carry best-by only. Confidence high.
- roastar_valve. Roastar, "Degassing Valve Guide". https://www.roastar.com/blog/what-you-really-need-to-know-about-degassing-valves . Verbatim: "Once pressure inside the bag hits a specific threshold, the degassing valve allows CO₂ to escape while blocking oxygen from getting in." Supports: valve mechanism (used in the mechanism text only). Confidence high.
- plitek_valve. Plitek, "Coffee Valve". https://www.plitek.com/resources/technical-resources/coffee-valve/ . Verbatim: "allow the release of carbon dioxide gas naturally produced by the coffee beans during the roasting process, while preventing oxygen and moisture from entering the packaging." and "After roasting, freshly roasted coffee beans release a large amounts of carbon dioxide for several days." Supports: CO2 release after roasting. Confidence high.
- achilles_processing. Achilles Coffee Roasters, "Processing Deep Dive: Washed vs. Natural vs. Honey". https://achillescoffeeroasters.com/blogs/specialty-coffee-blog/processing-deep-dive-washed-vs-natural-vs-honey . Verbatim: washed "The outer skin and pulp are removed through a mechanical depulper" and "thoroughly washed with clean water to remove all remaining mucilage", giving "clean and crisp cups" with "bright, lively, and often citrusy or floral notes"; natural "The whole cherries—including the skin, pulp, and mucilage—are spread out in thin layers on drying beds or patios", giving "bold, fruity, and often intensely sweet flavor profiles"; honey "the mucilage—the sticky, sugary layer surrounding the bean—is intentionally left intact during drying", giving "sweet, syrupy body". Supports: process definitions and flavour direction. Confidence high. A second roaster set (search summary of Phoenix, Craft Coffee Canada, Brew Clan) says the same.
- seattlecoffeegear_anaerobic. Seattle Coffee Gear, "Anaerobic Coffee Processing". https://www.seattlecoffeegear.com/blogs/learning-center/anaerobic-coffee-processing . Verbatim: "Cherries or depulped seeds are sealed in airtight tanks before or during drying, removing oxygen from the fermentation environment." and "Anaerobic coffees tend to be characterized by intensity and unusual complexity. Common descriptors include tropical fruit (pineapple, passion fruit, mango), wine or cider-like fermentation notes, vivid florals, and sometimes spice or herbal qualities." Supports: anaerobic definition and taste. Confidence high. Search summary (Coffee Magazine, Bean Box) adds co-fermentation: "an anaerobic ferment with non-coffee ingredients tossed into the tank, such as fruit, spices, hops".
- lavazza_arabica_robusta. Lavazza USA, "Arabica & Robusta Coffee Beans". https://www.lavazzausa.com/en/recipes-and-coffee-hacks/arabica-robusta-meaning-vocabulary-curiosities . Verbatim: "Arabica's caffeine content ranges from 0.8% to 1.4% of bean dry weight." and "Robusta's caffeine content ranges from 1.7% to 4.0% of bean dry weight."; Arabica "roughly 60% of global output". Supports: robusta carries roughly double the caffeine. Confidence high for the direction, medium for the exact ranges (see next).
- coffeehero_arabica_robusta. Coffee Hero, "Arabica vs Robusta Coffee". https://coffeehero.com.au/blogs/news/arabica-vs-robusta-coffee-difference . Verbatim: "Arabica: 1.2-1.5% caffeine by weight"; "Robusta: 2.2-2.7% caffeine by weight"; "The bitterness comes from higher chlorogenic acid content (7-10% vs Arabica's 5.5-8%)." Supports: robusta roughly double caffeine; more bitter. Confidence high for direction. The two sources' ranges differ; both put robusta at about twice arabica.
- valor_caffeine_roast. Valor Coffee, "Does Light Roast Coffee Really Have More Caffeine Than Dark Roast?". https://valor.coffee/blog/does-light-roast-coffee-really-have-more-caffeine-than-dark-roast . Verbatim: "Light roast coffee beans, being denser, can have slightly more caffeine if you measure by weight (like 20 grams of beans)." and "Dark roast beans are larger due to expansion from roasting, so a scoop of dark roast might contain fewer beans than a scoop of light roast, potentially leading to less caffeine." Supports: roast level barely changes caffeine; measurement method explains the myth. Confidence high.
- podium_caffeine_roast. Podium Coffee Club, "Light Roast vs Dark Roast Caffeine". https://podiumcoffeeclub.com/blogs/blog/light-roast-vs-dark-roast-caffeine . Verbatim: "A raw green coffee bean and the same bean roasted to a dark French roast contain roughly the same absolute amount of caffeine — the losses are small, in the range of 5–10% through the roasting process."; "A tablespoon of ground light-roast coffee weighs roughly 5.5–6 grams. A tablespoon of ground dark-roast coffee weighs roughly 5–5.5 grams". Supports: 5-10% loss at most; density explains scoop differences. Confidence high.
- cannonball_strength. Cannonball Coffee, "'Strength 5 Coffee' doesn't mean anything". https://cannonballcoffee.co.uk/blogs/cannonball-coffee-blog/coffee-strength-numbers . Verbatim: "Contrary to popular belief this scale actually relates to the darkness of the roast, rather than a strong flavour or high caffeine content." and "'Strength 5' on the label just means the beans are roasted for a very long time." Supports: UK strength number is roast darkness. Confidence high.
- coffeejudge_strength. Coffee Judge, "Coffee Strength Guide". https://www.coffeejudge.co.uk/coffee-strength-guide . Verbatim: "The strength depends on the roasting time. NOT, as you might think, on the caffeine content." and "So it's possible to have a strength 3 with more caffeine than a strength 5." Supports: strength is roast time (second source); not standardised across brands (the page uses its own 1-6). Confidence high.
- merakitech_blend. Meraki, "Single-Origin vs Blend for Espresso". https://www.merakitech.com/blogs/coffee-knowledge/single-origin-vs-blend-for-espresso-which-shots-better . Verbatim: "A blend is crafted to taste the same in February and August, from crop to crop"; "Single origins are a seasonal snapshot of one place" and "change slightly from crop to crop"; blends are "balanced" and "hold up to steaming"; single origins suit people who "drink espresso without milk". Supports: blend = consistency and milk; single origin = seasonal character and black. Confidence high.
- commonroom_blend. Common Room Roasters, "Blends vs Single Origin Coffee". https://www.commonroomroasters.com/blogs/specialty-coffee/blended-coffee-vs-single-origin-coffee . Verbatim: "Blends combine beans from multiple origins. The goal is to build balance, consistency, and depth."; "once a harvest is gone, it's gone until next year, and flavors shift with each crop."; choose blends if you "drink espresso or milk-based drinks daily". Supports: same split, second source. Confidence high.
- mtpak_altitude. MTPak Coffee, "The impact of altitude on growing and roasting coffee". https://mtpak.coffee/2021/10/impact-altitude-growing-roasting-coffee/ . Verbatim (quoting Sweet Maria's): "Beans grown at a higher altitude tend to express greater density, which generally translates to better, specialty quality coffee." and "Higher altitude coffees develop higher acidity and smoother mouthfeel in the cup, whilst lower altitude coffees tend to develop heavier body and lower acidity." Fetched excerpt: most arabica benefits from 1,000-1,800 masl; robusta below 600 masl. Supports: altitude → density → acidity and extraction difficulty. Confidence high.
- pdg_green_fresh. Perfect Daily Grind, "How long does green coffee stay fresh for?". https://perfectdailygrind.com/2022/04/how-long-does-green-coffee-stay-fresh/ . Verbatim: "Most specialty coffee roasters consider green coffee to be fresh for six to 12 months."; Ahmed Mahyoub (Mocha Mill): "The generally acceptable timeframe for green coffee to be considered fresh is one year from the time it's processed and shipped."; "Once the year mark has passed and the next harvest lands at the roastery, the coffee is considered to be 'past crop' – a label generally associated with undesirable cereal or aged flavour notes." Supports: green coffee has its own freshness clock; past crop flavours. Confidence high.
- greencoffeecollective_pastcrop. Green Coffee Collective glossary, "Past Crop". https://greencoffeecollective.com/pages/glossary/past-crop . Verbatim: "There's no fixed definition of when a coffee transitions from current to past crop, but it's generally understood to occur after 18–24 months from the harvest date." and "In the cup it develops flat, papery, woody, or hay-like notes". Supports: past-crop timing (second source, disagrees on months: 12 vs 18-24). Confidence high.
- honestcoffeeguide_grind. Honest Coffee Guide, "Coffee Grind Size Chart". https://honestcoffeeguide.com/coffee-grind-size-chart/ . Verbatim: espresso "180-380 microns"; pour over "410-930 microns"; filter machine "300-900 microns"; French press "690-1300 microns"; cold brew "800 microns and more"; "The data from this tool is based on hands-on research and feedback from coffee enthusiasts like you!" Supports: grind ranges by brewer; crowd-sourced. Confidence medium.
- notabarista_grind. I'm Not a Barista, "Coffee grind size". https://notabarista.org/grind-size-chart/ . Verbatim: espresso "180–380 µm"; V60/pour-over "400–950 µm"; French press "1000–1200 µm"; cold brew "900–1500 µm". Supports: grind ranges (second source; French press range disagrees with Honest Coffee Guide). Confidence medium.
- beansandbarista_grind. Beans & Barista, "Coffee Grind Size Chart: What Has Actually Been Measured". https://beansandbarista.com/blog/coffee-grind-size-chart/ . Fetched excerpt: the two popular charts disagree (French press 690-1,300 vs 1,000-1,200 µm); only peer-reviewed drip measurements (UC Davis, laser diffraction) are "770 ± 10 · 928 ± 3 · 1,115 ± 74 µm" for fine, medium, coarse. Supports: published micron ranges are mostly crowd-sourced; the point for buyers is the order (espresso finest, French press coarsest), not the exact numbers. Confidence high.
- brewedbarista_grind. Brewed Barista, "Coffee Grind Size Guide". https://www.brewedbarista.com/guides/coffee-grind-size-guide . Verbatim: "Pre-ground coffee is ground for drip machines. It's way too coarse for espresso and will produce a watery, sour shot that runs in 8 seconds." Supports: default pre-ground is drip grind; espresso mismatch. Confidence high.
- coffeecraftguide_nespresso. Coffee Craft Guide, "Nespresso Original vs Vertuo". https://www.coffeecraftguide.com/nespresso-original-vs-vertuo/ . Verbatim: "Vertuo reads a barcode on each pod"; Original: "Dozens of brands produce compatible pods including Starbucks by Nespresso, Lavazza, L'OR, Illy, Peet's, Café Royal, and many supermarket own-brands."; Vertuo: "No third-party compatible Vertuo pods have reached meaningful commercial scale" and "you can only buy pods from Nespresso." Supports: capsule system is a hard compatibility filter; Vertuo is single-vendor. Confidence high.
- whichcapsule_vertuo. WhichCapsule, "Nespresso Vertuo Pod Alternatives". https://whichcapsule.com/us/blog/best-alternatives-nespresso-vertuo-pods . Verbatim: "There may be limited authorized or market-specific options, but Vertuo does not have the broad third-party ecosystem that Nespresso Original has." and "With Nespresso Original, many third-party brands make compatible capsules." Supports: same, second source. Confidence high.
- sca_310_2021. SCA Standard 310-2021, "Home Coffee Brewers: Specifications and Test Methods" (PDF hosted by CoffeeGeek; downloaded and read with pdftotext after the blind check). https://coffeegeek.com/wp-content/uploads/2023/10/SCAGoldCupStandard.pdf . Verbatim: "The brew ratio used for testing in this standard is 55 g of coffee per 1.000 kg of fresh water unless stated otherwise."; "reach a temperature in the 90 to 96°C range by the moment when 33% of the water mass has been sprayed onto the coffee"; typical conditions "a 100% arabica coffee of roast level between #55 to #65 using SCA/Agtron Roast Color Tiles; coffee particle size distribution of ≥65% in the 589 µm-1168 µm range"; "Coffees at least of 1-1.5 weeks off-roast". No "±10%", no "1:20 to 1:16.25" and no Fahrenheit figure appear in the document; those came from the earlier search summary and have been dropped. Supports: 55 g/kg standard ratio (so 13.75 g per 250 ml cup); 90 to 96°C; drip grind band; SCA calls #55 to #65 medium. Confidence high.
- podium_espresso_ratio. Podium Coffee Club, "Espresso Brew Ratio Explained". https://podiumcoffeeclub.com/blogs/blog/espresso-brew-ratio . Not fetched by the author; the blind check read the page and reports it states that an 18 g dose yields a 36 g shot at 1:2. Supports: espresso dose of 18 g. Confidence medium (checker-verified).

## Disagreements kept side by side

- Agtron bands. Cloud City: Light 70-85, Medium 60-69, Dark 50-59. Coffee Review: Light >80, Medium-Light 70-80, Medium 60-70, Medium-Dark 50-60, Dark 40-50. Roasted Record: mid-70s+ light, 60s medium, 50s medium-dark, 40s dark. Onyx reports ground Agtron #130 on the gourmet scale, which runs above the 25-95 M-Basic disc range. Resolution: give the scale and direction as fact; give bands as "roughly" and cite all three; treat Agtron as an annotation and sort key, never a filter with hard cut-offs.
- Rest window after roasting. Monastery: espresso 5-7 days, filter 4-10, light 7-10, dark 3-5, peak 3-14 days, decline after 2-4 weeks. Simon & Bearns: espresso optimal 7-21 days and stable 6-8 weeks, filter optimal 3-14 days and good 4-6 weeks. Trade: text 5-21 days, timeline 5-14 days peak, oxidising faster beyond 30. Hoffmann: buy within 2-4 weeks, use within another 3-4. Home Cafe: espresso 7-40 days. Resolution after the blind check: no single window is claimed; the document says the peak sits somewhere between about 3 and 21 days depending on source and brewer, draws it as a gradient, and records the disagreement in a judgement.
- Caffeine by species. Lavazza: arabica 0.8-1.4%, robusta 1.7-4.0%. Coffee Hero: 1.2-1.5% vs 2.2-2.7%. Both: robusta roughly double. Resolution: say "roughly double"; do not print a single percentage.
- Decaf threshold. Wikipedia and Frequent Coffee: US "at least 97% removed" under FDA; EU 0.1% by weight in roasted coffee. A search summary (Driftaway, Tasting Table) says the 97% figure is an industry expectation not codified in FDA regulation. Resolution: report 97% as "the US convention" and 0.1% as the EU legal line; flag in judgements.
- Green coffee past crop. PDG: fresh 6-12 months; past crop when the next harvest lands. Green Coffee Collective: 18-24 months. Resolution: "a year is the trade's rule of thumb, with the hard decline by 18 to 24 months".
- Grind microns. Honest Coffee Guide French press 690-1300; Not a Barista 1000-1200. Beans & Barista: both crowd-sourced; only UC Davis drip numbers are measured. Resolution: publish the order and the rough espresso (about 200-400 µm) to French press (about 700-1300 µm) span; the buyer-facing fact is the grind_for label, not microns.
- Share of coffee that is specialty. Bean Box 5-10%; Wikipedia-derived summaries 3-5%. Not used as a number (the blind check found it had leaked into cup_score prose; removed).

## Data check: where each facet's value comes from

| facet | on listing? | where | backfill |
|---|---|---|---|
| format (whole bean / ground / capsule / instant) | yes, nearly all | title; "Item Form"; variant selector | none |
| grind_for | yes on ground SKUs and capsules | variant selector ("Grind: Whole Bean, Drip, French Press, Espresso"); capsule system in title | low: default pre-ground = drip grind (brewedbarista) |
| flavor_family | derived | tasting notes + roast level + process → SCA wheel first tier; Counter Culture already filters on five buckets | low (LLM map from notes); medium where no notes |
| freshness_policy | derived from roaster site + listing | "roasted to order", "ships within 48 hours of roasting", roast date printed, or best-before only | low: one lookup per roaster, not per SKU |
| roast_date | rarely a field | bag photo or shipment; Trade/Onyx ship within days | high: per-unit; backfill as roaster policy + ship date; top N only |
| roast_level | yes | title / "Roast Level" field; Peet's 0-100 scale; Lavazza "intensity" | none; normalise house words |
| agtron | no, except Onyx and Coffee Review | "Ground Agtron: #130"; Coffee Review "55/68" | medium: Coffee Review for top N; else map roast words with wide bands |
| process | specialty yes as a field; mass-market rarely | "Process: Washed"; Peet's "Washed and Wet-Hulled" | source marked researched, backfill medium: field on specialty pages, roaster lookup for mass-market, unspecified when silent |
| origin_type | yes | "Single-Origin" / "Blend" / "Signature Blend" | none |
| origin | yes (vague on mass-market) | country/region, or "Americas, Indo-Pacific" | none |
| variety | specialty only | "Variety: Geisha", "74110, 74112" | high; top N only |
| elevation | specialty and marketplaces (Trade) only | "1950 MASL", "900-1650m" | source marked researched, backfill medium; roaster page for top 200 single origins |
| species_blend | yes on mass-market ("100% Arabica"), inferred on specialty | title, bullets; Lavazza blend text | low: specialty without a claim = arabica |
| caffeine_level | yes | "Decaf" in title; Onyx filter Full/Partial/Decaf | none |
| decaf_method | specialty decaf yes; mass-market decaf usually silent | "Swiss Water", "Sugarcane EA", "CO2" in title/description | source marked researched, backfill low |
| roast_profile_for | not a field anywhere | words in the product name; mass-market brew suggestion ("Ideal for Cafetiere or Filter") | source marked derived, backfill low |
| harvest | top roasters only | "Harvest: December 2025", "April" | high; top N only |
| cup_score | few roasters; Coffee Review | "87", "86.75"; Coffee Review 50-100 | medium: Coffee Review for top N |
| strength_number | UK/Italian mass-market only | "Strength 4", "7/10 intensity" | listing where it exists; absence means not applicable, so none |

## Against the existing add-on schema

The current add-on (`data/schema-parts/formulations/food-beverages/add-ons/coffee.jsonc`) holds one field, `roast_level` (enum light…dark). Standards above it give `ingredients`, `high_signal_markdown`, `sizing.net_quantity`, `sizing.pack_count`, `sizing.serving_size`, `servings_per_container`, `nutrition`, `allergens`.

Classification of what this document surfaces, against AGENTS.md's four allowed kinds. Proposals only.

- format: hard filter. Not in the add-on. Buyers filter by literal value (whole bean vs ground vs capsule); every row can carry it from the title. Propose `format` enum.
- grind_for: hard filter (compatibility: the brewer or capsule system it must match). Not in the add-on. It can differ across variants of one listing (grind is a variant selector), so under rule 2 it belongs inside `sizing`, alongside `net_quantity`. Propose `sizing.grind_for` enum (whole_bean, espresso, moka, drip, pour_over, french_press, cold_brew, nespresso_original, nespresso_vertuo, k_cup, other, unspecified).
- roast_level: keep as is. It is the category's one dense enum; the Agtron number is the quality currency behind it but is too sparse to commit to augmenting on every row, so Agtron stays in markdown.
- flavor_family: derived from tasting notes at ranking time. Not a schema field; the tasting notes themselves are high_signal_markdown material (the standard's comment already names provenance and process). Propose no field; the ranker reads notes.
- freshness_policy / roast_date: roast date is per-unit and never a page field, so it fails "present on every row". Freshness policy is per-roaster, not per-product. Both are high_signal_markdown material ("roasted to order", "ships within 48 hours of roasting"). Propose no field.
- process: candidate vertical-anchor-adjacent fact. It is dense on specialty, absent on mass-market, and we would be willing to augment it (one lookup on the roaster page) because it is the single strongest taste predictor after roast level. Propose `process` enum (washed, natural, honey, anaerobic, co_fermented, wet_hulled, other, unspecified) as a hard filter; mark unspecified rather than guess.
- origin_type (single origin / blend): hard filter, dense everywhere, zero augmentation cost. Propose `origin_type` enum.
- origin, variety, elevation, harvest, cup_score: provenance and grade. Under rule 5 these are Jev markdown material (the formulations standard's comment already lists "estate, single-origin, PDO/GI" and "grades"). Propose no fields.
- species_blend: "100% Arabica" and robusta share. Markdown material; the ingredient list already carries species when declared.
- caffeine_level: hard filter (full / half / decaf), dense, zero cost. Propose `caffeine_level` enum. decaf_method stays in markdown.
- roast_profile_for (espresso / filter / omni): filter, but sparse outside specialty and correlated with grind_for. Markdown.
- strength_number: brand-scoped, not comparable across brands. Noise as a structured field; keep in markdown or drop.

Add-on fields the document considers noise: none, there is only roast_level and it stays. Standard fields that are weak for coffee: `nutrition` (near-empty for black coffee) and `allergens` (none declared on plain coffee); both are inherited and harmless.

## Blind check, 2026-10-19

Eleven findings from the coordinator's blind check, all addressed in category.json:

1. cups_per_bag now divides by 13.75 g (55 g per kg at a 250 ml cup) for filter and 18 g for espresso, and says which cup size a retailer's count assumes.
2. The SCA PDF was downloaded and read; the reference is now sca_310_2021 with only what the document says (55 g/kg, 90 to 96°C, tiles #55 to #65, 1 to 1.5 weeks off roast). The ±10%, 1:20 to 1:16.25 and Fahrenheit figures were dropped.
3. 55 g/kg is cited to the standard directly; 18 g to the Podium page the checker read; the 5 to 10 percent specialty share was removed from cup_score prose and from the Bean Box supports line.
4. The peak window is stated as a range across sources (about 3 to 21 days), Trade's own 5 to 14 timeline is recorded, and the judgement lists every source's numbers.
5. Grind ranges now follow the sources: espresso 180 to 380 µm, pour-over about 400 to 950, French press about 700 to 1,300, cold brew about 800 to 1,500.
6. The origin_type cold-start question now asks the facet directly; milk-drinking is demoted to a preference weight in the rationale and removed from skip_if_known.
7. The freshness question now carries the consequence ("knowing it may have sat for months").
8. Wikipedia's 0.1 percent sentence is attributed to Canada; the EU figure is cited to Frequent Coffee only; "legally means" became "the widely accepted trade definition".
9. elevation, decaf_method and process reclassified as researched; roast_profile_for as derived; strength_number kept as listing with the not-applicable justification in its backfill note.
10. The FDA page was fetched and quoted verbatim (US claim now high confidence). EUR-Lex and legislation.gov.uk returned empty to WebFetch and curl on every attempt, so the EU claim is a separate low-confidence judgement that says so.
11. Freshness window origin says it is a proposal; the Monastery entry above is relabelled a paraphrase; "at most" became "in the range of".

Not changed: the Trade reference keeps both the 5 to 21 (text) and 5 to 14 (timeline) figures because both are on the page.

## Branches

Added 2026-10-19. Each branch is a claim about how the answer narrows the path; the sentence it follows from is given. Values that change nothing (format = instant, flavor_family = chocolate and nuts) have no branch and fall through to the next split.

- format = ground, capsule: then ask grind_for. From the format rationale: "a buyer without one needs a grind that matches the brewer, and a capsule buyer needs the one capsule system that fits."
- format = whole bean: then ask roast_profile_for; grind_for stops mattering. From grind_for's own value "whole bean (any brewer)" and the data check row ("grind_for: yes on ground SKUs and capsules"); the brewer match moves to the roast, per Seed 2 ("Second match: roast level to brewer, because light roasts are hard to extract on a home espresso machine") and roast_profile_for ("Buying the roast made for your brewer removes most of that fiddling").
- flavor_family = fruity and floral: then ask process and roast_profile_for. From flavor_family ("'fruit and flowers' usually means a light roast and often a natural or anaerobic process") and the light-roast trade-off ("Push toward light, fruity, high-grown coffee and espresso gets hard").
- flavor_family = fermented and funky: then ask process. From the process trade-off ("Naturals and anaerobic lots give the biggest fruit and the most surprise, and also the widest swing between bags") and the mechanism family note that anaerobic is "not the bag to recommend to someone who asked for 'a nice coffee'": the process answer separates a natural from an anaerobic or co-fermented lot.
- flavor_family = roasty and smoky: origin and process stop mattering. From flavor_family ("'roasty and smoky' means the roast dominates and the origin barely shows") and the dark-roast trade-off ("What you give up is everything the origin and process put in the bean").
- freshness_policy = roasted to order and shipped within days, roast date printed: then show roast_date. From roast_date ("It is the number experts read first" and "The useful value is days since roast"); the derived metric days_since_roast depends on it.
- freshness_policy = best-before date only: then show strength_number; roast_date and harvest stop mattering. From the freshness_policy definition ("or only a best-before date"), the roast_date definition ("the best-before date, which is set by the packer and says nothing about age"), the harvest definition ("which the very top roasters print alongside the roast date") and strength_number ("Keep the number on the card for buyers who navigate by it"; data check: "UK/Italian mass-market only").
- origin_type = single farm or microlot: then show origin, variety and harvest. From origin ("A farm name, lot size and producer ... means a traceable lot"), variety ("Only the top tier of roasters print it") and harvest ("the very top roasters print [it] alongside the roast date").
- origin_type = single origin (country or region): then ask origin. From origin_type ("single origin for black coffee and for tasting where it came from") and origin ("it is how buyers remember what they liked").
- origin_type = blend: then ask species_blend; harvest stops mattering. From species_blend (robusta "is what makes a traditional Italian espresso blend taste the way it does") and origin_type ("A single origin is a snapshot of one place and harvest", while a blend's "components go in and out of season").

## Trimmed detail (2026-10-19)

Sentences cut from category.json to meet the validator's caps, by field. Nothing here was changed, only moved; numbers and sources are as they were.

- summary: "Scope: roasted coffee sold as whole bean or ground, with capsules and instant appearing only as a format."; the qualifier that the 5 to 21 day peak is "depending on brewer and roast"; and that the expert checks "not the roast level on the front of the bag but whether there is a roast date on the back".
- mechanism.plain: "Roasting also puffs the bean and drops its density, which is why dark beans are bigger and lighter per scoop but carry almost the same caffeine by weight." and "A one-way valve on the bag lets the gas out and keeps air out." (roastar_valve and plitek_valve stay cited on the mechanism; the vocabulary defines the valve).
- splits[format].rationale: "It also sets the freshness ceiling, because ground coffee fades within a day or two of grinding however good it was."
- splits[freshness_policy].rationale: a best-before date "can sit a year out and says nothing about when the beans were roasted".
- splits[origin_type].rationale: "Asked in the facet's own words so the answer maps straight onto single origin or blend." and "If memory already says the buyer drinks mostly with milk, use that only as a weight toward blends and medium to dark roasts, not as the answer."
- grind_for.why_experts_care: the micron ranges by brewer, "espresso about 180 to 380 microns, pour-over and drip in the middle at about 400 to 950, French press about 700 to 1,300 and cold brew about 800 to 1,500" (judgement 10 records these as crowd-sourced; the Grind size scale grammar keeps the order).
- flavor_family.why_experts_care: the wheel's nine centre-ring groups, "fruity, floral, sweet, nutty/cocoa, spices, roasted, green/vegetative, sour/fermented, other".
- freshness_policy.why_experts_care: the per-source windows "(Trade: days 5 to 14 for most methods; Monastery: 3 to 14 days, losing quality after 2 to 4 weeks; Simon & Bearns: 7 to 21 days for espresso)"; "James Hoffmann's rule is to buy within 2 to 4 weeks of the roast and use it within another 3 to 4."; "the EU rules we could only read second-hand require a best-before, not a roast date" (judgement 6, low confidence); "Supermarket coffee usually shows only the best-before and may have sat in a warehouse for months."
- roast_date.why_experts_care: "Whole beans lose about half their aroma compounds within two weeks of roasting unless sealed" (coffeetower_staling); espresso wants rest "because the trapped gas disturbs extraction".
- roast_level.why_experts_care: "the dark bean is simply bigger and lighter, so a scoop of it holds a little less coffee."
- agtron.why_experts_care: Onyx's gourmet scale "runs past 95 for very light roasts"; "The SCA's own home-brewer test standard defines its medium test coffee as roast colour tile #55 to #65." (sca_310_2021); the ten-point band shift was stated relative to Coffee Review's bands ("from those").
- origin_type.why_experts_care: "Neither is higher quality in itself."
- origin.why_experts_care: the traceable-lot example "(Tim Wendelboe: 'Producer: Negussie Tadesse, Region: Dakiti, Kaffa')".
- variety.why_experts_care: "Gesha (also spelt Geisha)"; Onyx's green prices, "$22.75 a pound" for Gesha against "$5.44" for blend components; Ethiopian research codes "('74110, 74112')".
- elevation.why_experts_care: the worked example "so a 1,900 m light roast needs hotter water and a finer grind on an espresso machine than a 1,200 m medium".
- species_blend.why_experts_care: "The useful signal runs the other way."; robusta carries "more of the bitter chlorogenic acids" (coffeehero_arabica_robusta: 7-10% against arabica's 5.5-8%); the crema gloss "(the tan foam on an espresso)", which the vocabulary defines.
- caffeine_level.why_experts_care: "Note that roast level does not change caffeine in any way a drinker can feel; the only lever is the species and the decaffeination." (rephrased to "Roast level does not change caffeine noticeably; the levers are species and decaffeination.").
- roast_profile_for.why_experts_care: espresso extracts "in under a minute"; espresso roasts are developed "a little" further.
- harvest.why_experts_care: the green clock "runs before roasting".
- cup_score.why_experts_care: specialty also requires "no more than 5 primary defects per 350 g sample" (wikipedia_specialty, beanbox_specialty); the CVA "separates describing a coffee from scoring it" (kept in judgement 8's note).
- strength_number.why_experts_care: "not how much caffeine is in it" (now "not its caffeine").
- judgements[1].claim: "espresso sits at the later end and light roasts later than dark" (now "espresso and light roasts sit later").
- judgements[1].note: Monastery also gives "quality loss after 2 to 4 weeks and espresso rest of 5 to 7 days"; Simon & Bearns also give "stable to 6 to 8 weeks". Both are in Disagreements above.
- judgements[2].claim: a light roast gives slightly more per scoop "because the beans are denser"; "in a way a drinker would notice" (now "noticeably").
- judgements[5].claim: "a roast date on a US bag" (now "a roast date"; the sentence still opens with "US federal law").
- judgements[7].claim: the EU 0.1 percent limit is "by weight in roasted coffee" (0.3 percent for green beans, per frequentcoffee_decaf).
- judgements[8].claim: "in the market".
- judgements[12].claim: field presence rests on "brand pages (Peet's, Lavazza via Flavia), the Trade marketplace, and search summaries of UK grocer listings".
- references, supports lines: eurlex_1169 "The date of minimum durability ('best before')"; onyx_geometry "caffeine per 12 oz", "roast level with Ground Agtron", "recommended grind in microns"; counterculture_collection "Retail filters"; drinktrade_freshness "most specialty coffee tastes best", "the peak for most methods"; peets_major_dickasons "A mass-premium brand"; taylors_listings "pages blocked; read via search summary"; sca_310_2021 and podium_espresso_ratio wording only; fda_date_labels is now a paraphrase of the verbatim quote recorded under Regulator above.

## Corrections log

### 2026-10-19

- Branches: 10 added across the four splits (format 2, flavor_family 3, freshness_policy 2, origin_type 3). See Branches above.
- Questions: 12 question fields rewritten to at most 12 words with no preamble: the 4 split questions and their 4 matching cold_start_questions (format, flavor_family, freshness_policy, origin_type), plus grind_for, species_blend, decaf_method and caffeine_level. Each split question and its facet's cold_start_question are now the same text.
- Caps: 39 fields trimmed to the validator's caps (summary, mechanism, 4 split rationales, 16 why_experts_care, 7 judgement claims, 1 judgement note, 9 reference supports lines). Cut sentences are under Trimmed detail above. No number, threshold, name or claim was changed; no facets, judgements, references, evidence ids or priority scores were added, removed or changed.
- Validator clean; category.md and viewer.html regenerated.

### Correction, 2026-10-19 (blind check of branches)
- format = whole bean: `roast_profile_for` removed from `then_ask`. The checker graded the leg WEAK: it rested on the seed reasoning in these notes, not on a register sentence. The branch keeps `grind_for` as irrelevant, which "Experts buy whole bean and grind at home" supports. The microlot leg (variety, harvest) stays: a traceable lot is what carries those two.