# Beverage chillers: research notes

Date: 2026-10-25. Scope: unpowered (non-refrigerated) things that chill a drink or keep it cold: ice and wine buckets and party tubs, freezer gel sleeves, in-bottle freezer sticks, vacuum-insulated bottle holders, whisky stones and steel cubes, and freezer-charged pour-through chillers for coffee and other drinks. Not plug-in wine fridges or thermoelectric bottle chillers (machines / appliances, `refrigerators`, "Refrigerators, Freezers & Wine Fridges"), not ice trays or big-ice moulds (sibling `molds`, "Molds, Ice Trays, Cutters & Stencils"), not can koozies and insulated tumblers (sibling `bottles`, "Bottles & Insulated Drinkware", under supplies / household), not picnic coolers (sibling `coolers`). Ice buckets overlap with `bottle-wine-serving-tools`; this register keeps them because their job is chilling, and leaves wine-preservation and pouring tools to that sibling.

Quote labels: **verbatim** = the sentence in quotation marks was in raw text I read myself (curl plus tag-stripping, or a brand's Shopify product JSON); **fetched excerpt** = the WebFetch tool summarised the page; **search excerpt** = the claim came only from a search-engine summary. No search-excerpt-only number is in category.json.

Vertical: the brief asked for `"vertical": "supplies"`, but the skill's `schema.json` only allows formulations, machines, fashion, furnishing or other, and the validator rejects `supplies`. The JSON uses `"other"`, as the concurrent supplies runs (mashers, pens, planners, labels) also did. The schema needs a `supplies` value before the brief's instruction can be followed.

## Seed

Seed output, recorded verbatim (first and only run used; a second accidental invocation, which drew car-tyres and was not used, was made only while saving output to the scratchpad):

```
=== Example entry (from an unrelated category) ===
category: olive-oil
what this shows: a facet that experts rank first, that almost no listing carries, and that is expensive to backfill. The honest answer is to list it, mark it researched, and limit backfill to the top of the catalog. Also shows a derived metric hanging off a researched facet.

<example>
{
  "id": "harvest_date",
  "label": "Harvest date",
  "definition": "The month and year the olives were picked, as opposed to the best-before date, which is set by the bottler and tells you nothing.",
  "why_experts_care": "Extra virgin oil is a fruit juice and loses the compounds that give it flavour and most of its claimed health effects within about 12 to 18 months of harvest. A bottle with a best-before two years out may already be a year old. People who buy well look for the harvest date and buy within a year of it.",
  "kind": "text",
  "roles": ["annotation", "sort_key"],
  "novice_answerable": false,
  "source": "researched",
  "listing_field_hint": "Rarely present. When it is, it is in a photo of the back label, not a field.",
  "backfill_cost": "high",
  "backfill_note": "Needs a label photo read or a producer page. Backfill the top 100 ranked oils and lazy-load the rest. Where the producer gives only a best-before, record it as unknown rather than guessing.",
  "visual": { "chartable": true, "encoding": "months since harvest on a horizontal scale, with the 12 and 18 month marks drawn" },
  "priority": { "narrows_set": 4, "cost_to_answer": 4 },
  "evidence": ["ucdavis_olive_center", "oliveoiltimes_freshness"]
}
</example>

Derived metric that follows from it:
{
  "id": "months_since_harvest",
  "label": "Months since harvest",
  "formula": "months_between(today, harvest_date)",
  "depends_on": ["harvest_date"],
  "unit": "months",
  "why": "The number a buyer actually needs. Under 12 is fresh. Over 18 and the oil is past the point most tasters would call it extra virgin."
}

=== Persona ===
A pastry chef, who thinks in ratios, temperatures and why a recipe fails

Ask yourself two things and write the answers under 'Seed' in notes.md:
  1. What is the beverage-chillers equivalent of the example above?
  2. What would this persona notice first about beverage-chillers that a typical buyer would not?
Keep what survives the data check. Drop the rest.
```

1. The beverage-chiller equivalent of the harvest date is the measured time to serving temperature. It is the number that matters, every brand claims one, and no listing carries a measured value with start and end temperatures. Survived as `stated_chill_time_min` (researched, high backfill cost), kept as a claim with a judgement attached, because the data check found Vacu Vin's own site saying 15 minutes and its Amazon listings saying 5. A separate facet for a measured time was dropped: only two timed tests (Drinkhacker, Alcohol Professor) exist in what I could read, both for spirits in a glass.
2. The pastry chef notices the ratio, not the gadget: how much cold the chiller carries against how much the drink needs, and that melting is where the cold lives. A gram of ice absorbs about 334 J melting and only about 2 J per degree warming; frozen soapstone absorbs 0.79 to 1.0 J per gram per degree and never melts. Survived as the two derived metrics (`ice_equivalent_g`, `ice_needed_g`), as `cold_source` (split 3), as the freezer-stick judgement (the stick carries at most about half the cold a warm white needs) and as the colder-freezer judgement. Dropped: a salt-to-ice ratio facet (a technique, not a product attribute; kept in a trade-off and judgement instead); terracotta evaporative coolers (no fetched source in this run).

## Sources and what each supports

### Physics: latent heat, specific heat, density, freezer temperature

**Cooking Issues, "Cocktail Science in General: Part 1 of 2", Dave Arnold, 2 September 2010** https://cookingissues.com/2010/09/02/cocktail-science-in-general-part-1-of-2/ (curl, raw text)
- verbatim: "It takes 0.5 calories to heat a gram of ice from -1°C to 0°C (this value is called the specific heat of ice,) but almost 80 calories to melt that same gram (this value is called the heat of fusion of water)."
- verbatim: "There is no chilling without dilution. There is no dilution without chilling."
- verbatim (comments, by Arnold): "Rocks don't dilute but they also don't have that much chilling power. The specific heat capacity of granite (the first rock i found) was only 0.19 cal/gram, whereas water is 1 cal/gram."
- verbatim: "Even if your ice is below 0°C, it won't chill a drink that much better than ice at 0°C"; "super-frozen ice may actually chill drinks slower than ice at 0°C (even though the drinks reach a slightly lower final temperature)."
- verbatim: chilling without dilution is possible with "the highly-touted 'whiskey stones'".
- Supports: mechanism, the dilution trade-off, the colder-freezer judgement, granite's specific heat (0.19 cal/g·K = 0.79 J/g·K). Confidence high.

**Wikipedia, Enthalpy of fusion** https://en.wikipedia.org/wiki/Enthalpy_of_fusion (curl)
- verbatim: "1 kg of ice (at 0 °C under a wide range of pressures), 333.55 kJ of energy is absorbed with no temperature change."
**Engineering ToolBox, Latent heat of melting** https://www.engineeringtoolbox.com/latent-heat-melting-solids-d_96.html (curl)
- verbatim table row: "Water, Ice 334" (kJ/kg).
- Supports: 334 J/g. Three sources agree (Cooking Issues' 80 cal/g = 335 J/g). Confidence high.

**Engineering ToolBox, Specific heat of solids** https://www.engineeringtoolbox.com/specific-heat-solids-d_154.html (curl)
- verbatim table rows: "Granite 0.19 0.79" (Btu/lb°F, kJ/kg K); "Steel 0.12 0.49"; "Ice -4 oF 0.47 1.97"; "Ice 31 oF (0 oC) 0.49 2.09".
**Wikipedia, Table of specific heat capacities** https://en.wikipedia.org/wiki/Table_of_specific_heat_capacities (curl)
- verbatim table rows: "Steel solid 0.466 3.756" (J/g·K, J/cm³·K); "Granite solid 0.790"; "Water at 25 °C liquid 4.181".
- Supports: steel 0.47 to 0.49 J/g·K; granite 0.79; water 4.18; ice about 2.0 to 2.1 (ETB 1.97 to 2.09; Cooking Issues 0.5 cal/g = 2.09). Confidence high.

**Greenstone Heat (soapstone masonry heaters)** https://www.greenstoneheat.com/why-is-soapstone-the-ultimate-material-for-any-masonry-heater/ (curl)
- verbatim: "Soapstone's specific heat capacity is approximately 0.98 to 1.0 kJ/kg·K, combined with a density of 2.8 to 3.0 g/cm³."
**Traditional Oven, Soapstone technical data** https://www.traditionaloven.com/articles/691/soapstone-data-info (curl)
- verbatim: "Specific heat capacity 785 J/Kg.°K"; "Mass density of soapstone 2.956 g/cm3".
- (Rumford's soapstone page repeats Traditional Oven's 785.2 figure "from Traditional Oven", so it is not independent and is not cited.)
- Supports: soapstone 0.79 to 1.0 J/g·K (sources disagree; the register gives the range); density about 2.8 to 3.0 g/cm³. Confidence medium on the specific heat.

**Engineering ToolBox, Metal alloy densities** https://www.engineeringtoolbox.com/metal-alloys-densities-d_50.html (curl)
- verbatim: "Stainless Steel 7480 - 8000" (kg/m³). With 0.49 J/g·K this gives 3.7 to 3.9 J/cm³·K, matching Wikipedia's 3.756. Used only in notes for the per-volume comparison.

**Wikipedia, Ice** https://en.wikipedia.org/wiki/Ice (curl): density "0.9167–0.9168 g/cm³". Notes-only (per-volume arithmetic).

**Wikipedia, Freezer** https://en.wikipedia.org/wiki/Freezer (curl)
- verbatim: "the freezer be regulated at −18 °C (0 °F)"; "Most household freezers maintain temperatures from −23 to −18 °C (−9 to 0 °F)".
- Second source: Drinkhacker measured its freezer-fresh materials at "1 degree Fahrenheit" (−17 °C); Samsara assumes a "-0°F freezer". Confidence high.

**Wikipedia, Frigorific mixture** https://en.wikipedia.org/wiki/Frigorific_mixture (curl)
- verbatim: "a mixture of sodium chloride, water, and ice form a frigorific mixture at −22 °C (−8 °F)". The table on the same page lists NaCl with snow/ice at "−18 °C / 0 °F". Two figures on one page; the register says "about −22 °C" in the reference line only and uses no salt temperature elsewhere.

### Timed tests (specialist reviewers who measure)

**Drinkhacker, "Experiment: Ice vs. Whiskey Stones vs. Tilt Chilling Sphere", 27 January 2013** https://www.drinkhacker.com/2013/01/27/experiment-ice-vs-whisky-stones-vs-tilt-chilling-sphere/ (curl)
- verbatim method: "Take 20ml of 80 proof vodka ... at room temperature (69.3 degrees Fahrenheit). Chill it with a similar volume of frozen material (all fresh from the freezer at 1 degree Fahrenheit)".
- verbatim results: ice "After 1 minute: 54.8 degrees ... After 20 minutes: 49.5 degrees (ice has melted completely)"; whiskey stones (soapstone) "After 1 minute: 56.5 degrees ... After 20 minutes: 63.7 degrees"; Tilt (stainless sphere) "After 1 minute: 59.2 degrees ... After 20 minutes: 46.1 degrees".
- verbatim: the stones "smelled — badly — of fish ... (The smell may be primarily due to the bag the stones are held in, but either way, the odor does transfer to the stones.)"
- Supports: stones chill briefly and warm back; steel does better; odour judgement. Confidence high for the test, low for the odour generalisation.

**Alcohol Professor, "The Ultimate Whiskey Ice Test", 29 February 2016** https://www.alcoholprofessor.com/blog-posts/blog/2016/02/29/the-ultimate-whiskey-ice-test (curl)
- verbatim: "The warmest temperature, at each interval, was with three TeroForma soapstone whiskey stones. The second warmest, at each interval, was with five of the soapstones."
- verbatim: "The stainless steel cubes from Kollea performed significantly better than the soapstones ... the initial chilling effect of Kollea's steel cubes, measured at two minutes, was within range of many of the actual ice variations."
- verbatim: "over the course of 20 minutes, their temperature would consistently begin to tick back up, whereas with nearly every single ice trial, temperature continued to decrease."
- Supports: same direction as Drinkhacker, from an independent test with Teroforma stones and Kollea steel. Confidence high.
- Note the one difference: Drinkhacker's steel sphere kept falling for 20 minutes, Alcohol Professor's steel cubes drifted back up. The register says "in one test they too drifted back up".

### Wine chilling methods and times

**Samsara Wine, "How to Chill Wine Fast: 5 Quick Methods"** https://www.samsarawine.com/education/how-to-chill-wine-fast-5-quick-methods-for-wine-champagne/ (curl)
- verbatim: "(62-68 degrees for red wine and 49-55 degrees for white wine)".
- verbatim: "a bottle of room temperature (70°F) wine will take about 40 minutes to cool to 50°F in a -0°F freezer."
- verbatim: "the ice bath method will take about 11-13 minutes to drop your wine to the appropriate serving temperature."
- verbatim: "Some sources recommend adding 3-4 tablespoons of salt to your traditional ice bath, however, this amount of salt isn't sufficient ... You will need at least ½ pound of salt ... you're looking at a 6-8 minute chill time."
- verbatim: spinning in salted ice water "will bring it to cellar temp in 2-3 minutes" (the winemaker's claim; not used as a number).
**Corkcicle blog, "How to Chill Wine Fast"** https://corkcicle.com/blogs/lifestyle/how-to-chill-wine-fast (curl)
- verbatim: ice bath "Allow the bottle to remain submerged for 15-20 minutes."; salted "Within 10-15 minutes, your wine should be nicely chilled".
**Usual Wines, "How to Chill Wine Fast"** https://usualwines.com/blogs/knowledge-base/how-to-quickly-chill-wine (curl)
- verbatim: salt "allows it to get much colder than 32 degrees Fahrenheit without freezing"; "Some say adding a few tablespoons of salt is sufficient, but such a small amount won't significantly drop the temperature ... pour at least a few cups of salt"; "In about 15 minutes or less, your wine will be ready to go."; freezer "set a timer for 30 minutes to an hour".
**NBC News Better** https://www.nbcnews.com/better/amp/ncna787426 (curl): "stash the bottle in the freezer for the next 45 minutes" (framing). Notes-only.
- Supports: plain ice water 11 to 20 minutes (Samsara, Corkcicle); salted 6 to 15 minutes (Samsara, Corkcicle, Usual); a few spoons of salt do little (Samsara, Usual); freezer slower than an ice bath (Samsara 40 min, Usual 30 to 60, NBC 45). Confidence high on the order, medium on the minutes. None of these compares ice alone with ice water, so the register does not claim that.

**Wine Folly, "Wine Serving Temperature"** https://winefolly.com/tips/wine-serving-temperature/ (WebFetch; curl returned a 961-byte shell)
- fetched excerpt: sparkling and light whites "38–45°F / 3-7°C"; rosé and full-bodied whites "44–55°F / 7-12°C"; light and medium reds "55–60°F / 12-15°C"; bold reds "60–68°F / 15-20°C".
- Second source for the bands: Samsara (above). The register uses only "38 °F to 68 °F" for the scale's span and the 60 to 68 °F bold-red band in no other place.

### The Kitchn test (could not read)
- "We Tested 6 Different Methods for Quickly Chilling a Bottle of Wine — And the Winner Only Took 15 Minutes" https://www.thekitchn.com/best-way-quickly-chill-wine-23186591 returned a PerimeterX bot wall to curl, and an Internet Archive `id_` snapshot returned the same wall. Search excerpt only: "equal parts ice, cold water, and ¼ cup kosher salt per quart". Not used.

### Brand spec sheets (fields brands publish)

**Vacu Vin, coolers collection** https://www.vacuvin.com/collections/coolers (product JSON, verbatim)
- Wine Cooler Sleeve Silver: "Before use, keep the cooling sleeve in the freezer for 6 hours for best results. When ready, slip it around your wine bottle, and let it cool for 15 minutes. The sleeve fits tightly around 0.7-1L bottles and drink bottles."
- Champagne Cooler Sleeve Silver: same instructions; "slip it around your champagne bottle". The brand sells wine and champagne sleeves as separate products, and a "Wine & Champagne Cooler Flexible Artico Sleeve" whose selling line is "Universal fit: Fits both wine and champagne bottles".
- Beer Cooler Sleeve: "fits snugly around 0.33 L beer bottles and half-size wine bottles"; "will stay cool for up to 1.5 hours".
- Wine Cooler Elegant Stainless Steel: "chills your wine in 15 minutes and keeps it cool for hours"; an "active cooling sleeve" inside a case.
- Wine Cooler Hive Stainless Steel: "Designed with double-wall stainless steel insulation, it keeps your pre-chilled bottle cool"; steps "Chill your wine or champagne bottle in the fridge. Insert the cold bottle into the cooler—no ice required."; "Fits bottles from 0.7 to 1.5 L, including most sparkling and champagne bottles."
- Supports: gel sleeves 6 h freeze and 15 min claim; insulated holders need a pre-chilled bottle; fit is a stated field; sparkling fit is a separate product line. Confidence high.

**Vinglacé** https://vinglace.com/products/wine-chiller and https://vinglace.com/products/grande-wine-champagne-chiller (product JSON, verbatim)
- "The Vinglacé wine & champagne chiller keeps your bottle cold without ice. It's made of stainless steel and vacuum insulated, so you don't have to worry about condensation or warm wine."
- Grandé: "It's our largest model, so you don't have to worry about your favorite bottle of bubbles fitting!"
- Supports: insulated holders keep cold; no condensation; sparkling fit needs a bigger model. Confidence high.

**Corkcicle Air** https://corkcicle.com/pages/corkcicle-air-presell (curl, verbatim)
- "Maintains chilled white wines for up to one hour and cools down room temperature reds to cellar temperature in 15 minutes."
- "Freeze your Corkcicle Air for at least 90 minutes in advance of using it."
- "If you're drinking a white it will keep it chilled."
- Dimensions: "Height: 12.0in"; "Weight: 0.11lbs".
**Amazon, Corkcicle Air (B00IULFU4C)** https://www.amazon.com/dp/B00IULFU4C (curl, verbatim)
- bullet: "brings room-temperature reds to cellar temp in 15 minutes and keeps whites perfectly chilled for up to an hour"; "Just freeze for 90 minutes"; Item Highlight row "Reds Chilled in 15, Whites Cold 1 Hour, Built-In Aerator, Flavor & Aroma Boost, Freeze for 90 Min, No Ice or Dilution"; "Item Weight = 0.1 kg"; "Material Type = Cork" (wrong for an acrylic stick; Amazon's material row is not trustworthy here).
- Supports: freezer sticks hold rather than chill whites; the stick weighs about 50 g in all. Confidence high.

**Corkcicle Ice Bucket** https://corkcicle.com/products/ice-bucket (product JSON, verbatim)
- "super insulated to keep beverages/ice cold for up to 48 hours"; "Slip in your favorite bottle of wine or pile the 100 oz ice bucket high with ice"; "triple-walled insulation".
- Supports: capacity is stated in ounces; insulated buckets exist. Single brand; the 48 h claim is not in the JSON.

**Teroforma Whisky Stones ICON** https://teroforma.com/products/whisky-stones-icon (product JSON, verbatim)
- "Set of 3 / 7/8" cubes"; "polished, engraved soapstone beverage cubes".
- "Whisky Stones beverage cubes can be used to cool down your favorite spirits just enough to take the edge off without 'closing down' the flavors."
- "Store in freezer for at least 4 hours before use. Add 3 stones to a glass. Pour until liquid has reached the height of 1 stone (about 2 fl oz). Let stand 5 minutes."
- "Do not chew or ingest. Not intended for use with large volumes of liquid."
- "Natural soapstone is classified Generally Regarded As Safe (GRAS) by the US FDA"; "non-porous and will impart neither flavor nor odor"; "FDA-approved, food-safe polish".
- Supports: 4 h freeze; the maker's own "take the edge off"; the GRAS claim (the maker's claim, not checked against an FDA listing); the odour disagreement with Drinkhacker. Confidence high that the page says this.

**HyperChiller V2** https://hyperchiller.com/products/hyperchiller-instant-coffee-beverage-cooler-black-2 (product JSON, verbatim)
- "Our Patented design uses regular water to chill and because there are no chemicals or Gels, all parts of the HyperChiller are dishwasher safe!"
- "Simply fill with water and place in the freezer, remove and place in your Coffee maker and brew or pour from a carafe in to the HyperChiller and then into your cup."
- "chilling all of your favorite beverages whether it's a special blend of tea, juice, wine or other adult beverages without dilution".
- "The HyperChiller will chill your Coffee in less than 60 seconds!" (single brand claim; not in the JSON).
- Supports: the coffee family is a frozen water core you brew into; no freeze time is stated. Confidence high.

### Retail listings (what data is present)

**Amazon, Vacu Vin Rapid Ice Champagne Cooler (B0000DDVX0)** https://www.amazon.com/dp/B0000DDVX0 (curl, verbatim)
- bullets: "Chills champagne within 5 minutes and keeps it cold for hours - without the need for a refrigerator or ice bucket"; "Sleeve is ready for application to the outside of a champagne bottle after 6 hours or more in a freezer"; "Designed to fit any standard-size champagne bottle".
- detail rows: "Size = fits standard 0.7-1 L champagne bottles"; "Material Type = Polyester"; "Item Weight = 9.6 ounces"; category "Wine Chillers"; Amazon product type "BEVERAGE_INSULATOR".
**Amazon, Vacu Vin Rapid Ice (B00CNFTOWA)** https://www.amazon.com/dp/B00CNFTOWA (curl, verbatim)
- title "Vacu Vin Rapid Ice Can Cooler - Silver" but bullets "Chills wine within 5 minutes and keeps it cold for hours"; "Designed to fit any standardsize wine bottle".
**Pak-it Products (distributor)** https://www.pakitproducts.com/product/vacu-vin-rapid-ice-chiller/ (curl, verbatim)
- "Once frozen, the rapid ice wine chiller will chill a bottle in five minutes and will keep it cold for hours."
- Supports: the 5 versus 15 minute disagreement within one brand, and where the freeze time and fit appear on a listing. The Amazon search page was not scraped (the harness blocked it as bot-check circumvention), so only these two listings were read; fields marked `listing` were limited to the capacity figure that brands put in the title.

### Regulators

**21 CFR 174.5, via Cornell LII** https://www.law.cornell.edu/cfr/text/21/174.5 (curl, verbatim)
- "(2) Any substance used as a component of articles that contact food shall be of a purity suitable for its intended use."
- "(1) The quantity of any food additive substance that may be added to food as a result of use in articles that contact food shall not exceed ... that which results from use of the substance in an amount not more than reasonably required to accomplish the intended physical or technical effect in the food-contact article".
**European Commission, Food contact materials** https://food.ec.europa.eu/food-safety/chemical-safety/food-contact-materials_en (curl, verbatim)
- "Constituents of food contact materials that transfer from these materials into the food may affect the chemical safety of the food and affect human health, as well as the quality of the food, its taste and smell, and its appearance."
- "To ensure a high level of food safety, all food contact materials must comply with Regulation (EC) No 1935/2004".
- "These rules ensure that packaging does not release substances into food that could endanger health, change the food's composition, or affect its taste or smell." (said of packaging on the page)
- EUR-Lex (the regulation itself) returned an empty body to curl twice, and legislation.gov.uk returned 0 bytes; the regulation is cited through the Commission's page. FDA's own food-contact-substances page returned a 10-byte body.
- Supports: in-drink parts are food-contact articles in both markets. Confidence high.

### Could not reach
- WebSearch budget for the session ran out (200 of 200 calls, shared across the concurrent supplies runs) after five searches of mine. Everything after that was fetched from known URLs.
- The Kitchn chilling test, Food & Wine, The Spruce Eats and Saveur wine-chiller reviews (bot walls or empty shells). No specialist review of wine sleeves or sticks with measured temperatures was read, which is why `stated_chill_time_min` stays a claim.
- USDA FSIS freezing page (Access Denied); FDA food-contact page (empty); Huski and Spinchill stores (not Shopify, or no response). Motorised spinning chillers are therefore not a value of `cold_source`; the Samsara spinning claim stays in these notes.

## Data points used for derived metrics

Constants (each from two sources above): latent heat 334 J/g; ice 1.97 to 2.09 J/g·K; water 4.18 J/g·K; steel 0.47 to 0.49 J/g·K; granite 0.79 J/g·K; soapstone 0.79 to 1.0 J/g·K (disputed); soapstone density 2.8 to 3.0 g/cm³; freezer −18 °C.

- `ice_needed_g`, 750 ml bottle from 22 °C to 10 °C: 750 × 4.18 × 12 = 37.6 kJ ÷ 334 = 113 g. Wine is treated as water (Cooking Issues notes ethanol's heat capacity is about half water's, so wine needs slightly less); the glass bottle is ignored, so 113 g is a floor.
- 60 ml pour, same drop: 60 × 4.18 × 12 ÷ 334 = 9 g, which is 8 percent of the bottle.
- Mug of coffee, 240 ml from 90 °C to 10 °C: 240 × 4.18 × 80 ÷ 334 = 240 g, about twice the bottle. The mug size and coffee temperature are illustrative, not sourced, so the register says only "more than a bottle does".
- Three 7/8 in Teroforma soapstone cubes: 2.22 cm per side → 10.97 cm³ each → 32.9 cm³; × 2.8 to 3.0 g/cm³ = 92 to 99 g. Warming from −18 °C to 10 °C (28 K) at 0.79 to 1.0 J/g·K: 2.0 to 2.8 kJ ÷ 334 = 6 to 8 g of ice. Teroforma's pour (2 fl oz, about 59 ml) needs 59 × 4.18 × 12 ÷ 334 = 8.9 g for a 12 °C drop, so the stones cannot quite manage that even if they reached 10 °C, which matches "take the edge off".
- Per gram, from −18 °C to a 10 °C drink: ice 2.05 × 18 + 334 + 4.18 × 10 = 413 J; soapstone 22 to 28 J; steel 13 to 14 J. Ice does 15 to 19 times the work of soapstone by mass.
- Per cm³ over the same 28 K: steel 3.76 × 28 = 105 J; soapstone (0.79 to 1.0 × 2.8 to 3.0) × 28 = 62 to 84 J; ice 413 × 0.917 = 379 J. So by volume steel holds about 1.3 to 1.7 times soapstone's cold, which is why the steel cubes in both tests did better although steel's per-gram figure is lower.
- Corkcicle Air: 0.11 lb (brand page) or 0.1 kg (Amazon) = about 50 g in all. Even if all 50 g were ice: 50 × (2.05 × 18 + 334) = 18.5 kJ, plus 50 × 4.18 × 10 = 2.1 kJ warming the melt, = 20.6 kJ, against 37.6 kJ for a 12 °C drop in a 750 ml bottle: 55 percent. The stick is part plastic, so the real share is lower. The register says "only half".

## Against the existing add-on schema

`zamana-expertise/data/schema-parts/supplies/kitchen/add-ons/beverage_chiller.jsonc` is empty (`{}`). The leaf composes kitchen-standard (`materials` with roles, `max_heat_temp`, `dishwasher_safe`) and supplies-standard (`high_signal_markdown`, `sizing` with `pack_count`, `pieces_included_text`, `dimensions`, `capacity`, `weight`, `measurements_text`). Proposals only; nothing edited.

| Facet in this register | In schema today | Proposed kind under AGENTS.md | Note |
|---|---|---|---|
| drink_target | no | hard filter, enum [wine_or_sparkling_bottle, drinks_for_a_crowd, single_glass, hot_coffee_or_tea] | Read from the title. Dense and cheap, so worth committing to augment. |
| chill_role | no | hard filter, enum [chills_from_warm, keeps_cold_only] | The fact buyers get wrong. Derivable from cold_source, so if only one of the two is stored, store cold_source and derive this at query time. |
| cold_source | partly, via `materials` (stainless_steel, stone_other) | category quality currency, enum [ice_and_water, frozen_water_or_gel_core, solid_stainless_steel, solid_stone, insulation_only] | `materials` cannot say whether a steel cube is solid or water-filled, which is the difference that matters. Consider adding `soapstone` and `granite` to the kitchen material enum; both currently fall into `stone_other`. |
| prefreeze_time_h | no | high_signal_markdown | One number per product, from the instructions; worth a sentence, not a column. |
| stated_chill_time_min | no | high_signal_markdown | A claim with no standard test; storing it as a field would let the ranker sort on marketing. |
| bottle_fit | no | hard filter, enum [still_750, still_and_sparkling, up_to_1_5_l] | A literal fit fact buyers filter on. Amazon has a "Size" row on some listings. |
| capacity | yes, `sizing.capacity` (fl_oz, qt, l, count_items) | sizing | Already right. Bucket capacity should be normalised to fl_oz; `count_items` fits a stated bottle count. |
| vacuum_insulated | no | high_signal_markdown | AGENTS.md cuts construction booleans; "double-wall vacuum" is a construction claim the markdown pass already keeps. |
| drink_contact | partly, via `materials` roles | high_signal_markdown | Certifications are cut under AGENTS.md; the food-contact claim belongs in the markdown line. Material is already covered by `materials`. |
| ice_equivalent_g | no (derived) | derived at query time from cold_source, `sizing.weight`, `sizing.pack_count` | Not stored. Needs `sizing.weight` to be the weight of the cold part, which holds for stones and cubes only. |

Fields this register considers noise for beverage chillers: `max_heat_temp` (nothing here is heated; Arnold's hot-rock aside is not a buying use); `sizing.dimensions` and `measurements_text` for sleeves and sticks (bottle fit is the real question, and outer dimensions do not answer it). `dishwasher_safe` is worth keeping: HyperChiller sells on it and Corkcicle Air is hand-wash only. `sizing.pack_count` is worth keeping for stone and cube sets.

## Branches

Every leg below names the register sentence it rests on. Answer mapping for each split: split 1's five answers are its five values (a glass of any drink answers `one glass`; drink_target backfill note); split 2's "room temperature" maps to `chills a warm drink` and "already cold" to `keeps a cold drink cold` (chill_role backfill note); split 3's "properly cold" maps to ice and water or a frozen water or gel core, "noticeably cooler" to solid stainless steel, and "just take the edge off" to solid stone (cold_source backfill note; corrected 2026-10-25, see below). Split 3 has no branches.

**Split 1, drink_target ("A wine bottle, beer, a crowd, one glass, or coffee?")**

- when `wine or sparkling bottle`: then_ask chill_role, bottle_fit. (makes_irrelevant capacity dropped 2026-10-25: wine and champagne buckets are bottle products, Corkcicle "Slip in your favorite bottle of wine", and their capacity varies.)
  - chill_role: "Vacu Vin tells you to chill the bottle in the fridge before its insulated Hive, and Corkcicle says its freezer stick keeps chilled whites cold" (chill_role, why experts care); both are bottle products.
  - bottle_fit: "Sparkling bottles often will not go into a still-wine chiller, so makers sell separate models" (bottle_fit, why experts care).
- when `beer bottle or can` (added 2026-10-25): makes_irrelevant bottle_fit, capacity. No then_ask.
  - bottle_fit irrelevant: its values are wine fits (still, sparkling, 1.5 L); Vacu Vin's Beer Cooler Sleeve states its own fit, "fits snugly around 0.33 L beer bottles and half-size wine bottles" (vacuvin_coolers supports line).
  - capacity irrelevant: "Only buckets and tubs vary in this; sleeves, sticks and single-bottle chillers hold one bottle" (capacity, why experts care); a beer sleeve holds one bottle or can.
- when `drinks for a crowd`: then_ask capacity, vacuum_insulated; makes_irrelevant prefreeze_time_h.
  - capacity: "For a crowd it decides whether one bucket is enough" (capacity, why experts care; reworded 2026-10-25 so the prose no longer ranks questions).
  - vacuum_insulated: "An insulated wall keeps ice and bottles cold longer and does not sweat onto the table" (vacuum_insulated, why experts care), with Corkcicle's "triple-walled" bucket as the example (reference line).
  - prefreeze_time_h irrelevant: "An ice bucket needs only ice" (prefreeze_time_h, why experts care) and "Set to 0 for insulation-only holders and ice buckets" (its backfill note). Chain: a crowd is served from buckets and tubs (split 1 rationale names buckets for this answer), and buckets need no frozen part.
- when `one glass` (renamed from `one glass of spirits` 2026-10-25): then_ask cold_source, drink_contact; makes_irrelevant bottle_fit, capacity.
  - cold_source: "Three answers keep steel apart from both, which is the main choice for a glass" (split 3 rationale, rewritten 2026-10-25).
  - drink_contact: "Anything in the drink is a food-contact article" (drink_contact, why experts care); stones and cubes sit in the glass (definition).
  - bottle_fit irrelevant: the facet is "Which bottle shapes and sizes go into the sleeve, chiller or bucket" (definition); a glass chiller holds no bottle.
  - capacity irrelevant: "Only buckets and tubs vary in this" (capacity).
- when `hot coffee or tea`: makes_irrelevant chill_role, bottle_fit, capacity. (then_ask prefreeze_time_h dropped 2026-10-25: HyperChiller, the only coffee chiller cited, states no freeze time on its product page, its other catalog entries or its FAQ, so the facet would be empty on this branch.)
  - chill_role irrelevant: "Coffee and tea start hot, so a coffee chiller always has to chill" (chill_role, why experts care): the value is fixed on this branch.
  - bottle_fit irrelevant: "coffee chillers you brew straight into" (drink_target, why experts care); HyperChiller: "place in your Coffee maker and brew ... in to the HyperChiller and then into your cup" (verbatim above). No bottle is involved.
  - capacity irrelevant: "Only buckets and tubs vary in this" (capacity).

**Split 2, chill_role ("Starting at room temperature, or already cold from the fridge?")**

- when `chills a warm drink`: then_ask prefreeze_time_h, stated_chill_time_min.
  - prefreeze_time_h: "Anything that chills from warm needs ice on hand or a part frozen 90 minutes to 6 hours ahead" (third trade-off).
  - stated_chill_time_min: "It is the number buyers compare and the least reliable one" (stated_chill_time_min, why experts care); it only applies to products that chill.
- when `keeps a cold drink cold`: then_ask vacuum_insulated, bottle_fit.
  - vacuum_insulated: "insulated holders sit on the keeps-cold side of the chill_role split" (vacuum_insulated, why experts care).
  - bottle_fit: Vacu Vin's Hive "Fits bottles from 0.7 to 1.5 L" and Vinglacé's Grandé exists "so you don't have to worry about your favorite bottle of bubbles fitting" (verbatim above), reflected in bottle_fit's why experts care.
  - Dropped leg: makes_irrelevant prefreeze_time_h. A buyer whose bottle is already cold can still pick a gel sleeve, which needs freezing, so the leg could not be traced.

## Trimmed detail

- **summary**: "steel a little more" stands for the Alcohol Professor finding that Kollea steel cubes were "within range of many of the actual ice variations" at two minutes.
- **mechanism.plain**: the per-volume reason steel beats soapstone (3.76 versus 2.2 to 3.0 J/cm³·K) is in the data points above, not the mechanism. Arnold's point that ice is "a very good conductor of heat – four times better than stationary water" was cut; it explains why super-cold ice warms to 0 °C fast but is one source.
- **cold_source.why_experts_care**: Drinkhacker's numbers (stones 56.5 °F at 1 minute to 63.7 °F at 20; ice 49.5 °F; Tilt steel 46.1 °F) are here and in the reference line.
- **stated_chill_time_min.why_experts_care**: Corkcicle Air's 15-minute claim is for reds to "cellar temperature", which is not a fixed number; no brand in this run states a start and end temperature.
- **drink_contact.why_experts_care**: Teroforma's "Do not chew or ingest. Not intended for use with large volumes of liquid." The second sentence also supports the stones-are-weak judgement.
- **tradeoffs[1]**: Samsara's spinning method (2 to 3 minutes in heavily salted ice water) is the winemaker's claim and is not in the register.
- **judgements[2].note**: the stick arithmetic (55 percent upper bound) is in the data points above.

## Corrections

- 2026-10-25, before the first build: removed "This is the commonest wrong buy" (chill_role) as an unsourced opinion; replaced "Sparkling bottles are wider" with what the brand pages actually show (separate models); removed "a few degrees" for soapstone, because Drinkhacker measured a 13 °F drop at one minute before it warmed back; replaced "ice in water beats ice alone", which no read source compares, with "an ice-water bath beats a freezer"; restated the gel-sleeve claim disagreement as Vacu Vin's site versus its Amazon Rapid Ice listings, since they may be different product generations; set `vertical` to `other` (see the top of these notes).
- 2026-10-25, second pass: split 2's rationale no longer says the makers state their products "cannot" chill a warm white; it now says, as the brand pages do, that they are sold to keep chilled bottles cold. "Under 1 J/g" restated as "at most about 1 J/g" because Greenstone gives soapstone up to 1.0. ETB's ice value at 0 °C corrected to its printed 2.09 kJ/kg·K.
- Build: validate, render and viewer ran clean. The skill's `tests/viewer-browser.py`, run on a scratchpad copy, passed at 1380, 768 and 390 px in light and dark (6 of 6). Script note: `schema.json` has no `supplies` vertical (see the top of these notes); no script misbehaved otherwise.

## Corrections after the blind check (2026-10-25)

Sources re-read for these fixes: Teroforma ICON product JSON (curl, verbatim: "particularly effective at retaining temperature for extended periods of time", "Do not chew or ingest"); HyperChiller V2 product JSON, its full products.json and /pages/faq (no freeze time anywhere); Corkcicle Air page ("Weight: 0.11lbs", "Maintains chilled white wines for up to one hour and cools down room temperature reds to cellar temperature in 15 minutes"); Engineering ToolBox metal densities ("Stainless Steel 7480 - 8000"). WebSearch was unavailable (session budget spent), so the Kollea cubes' and Tilt sphere's construction is still unknown and is stated as unknown.

1. cold_source (split 3): steel no longer maps to "properly cold". Question now "Properly cold, noticeably cooler, or just take the edge off?"; properly cold = ice and water or frozen core, noticeably cooler = solid stainless steel, edge off = solid stone. Rationale rewritten to the three-way reading of both tests; backfill note says neither tested steel is known to be solid; backfill_cost raised from low to medium for the steel case.
2. drink_target: added the value `beer bottle or can` (Vacu Vin Beer Cooler Sleeve, verbatim above; a vacuvin_coolers supports line added) with a branch that makes bottle_fit and capacity irrelevant; renamed `one glass of spirits` to `one glass` and the backfill note now says a glass of any drink answers it and that pour-through chillers file under coffee. Question now "A wine bottle, beer, a crowd, one glass, or coffee?". Title-versus-bullet rule added for keyword-stuffed titles (B00CNFTOWA). Kollea beer sticks not added: no Kollea page was read.
3. Judgement 3 (Corkcicle Air): claim now states only what Corkcicle says (keeps chilled whites, brings warm reds to cellar temperature, claims nothing for warm whites); the 50 g arithmetic is presented as conditional, with Amazon's 0.1 kg shown to reverse it; sources_disagree set to true. "Corkcicle says as much" removed. Seed `kept` line updated to match.
4. chill_role: rationale, backfill rule and Corkcicle claim now agree. Freezer sticks sit on the "chills a warm drink" side (they carry a frozen core and Corkcicle sells the Air for warm reds); the rationale says a warm start removes insulated holders only; the backfill note says to show the stick's own cellar-temperature claim. why_experts_care reworded from "only taking reds" to Corkcicle's wording.
5. Ice-bath yardstick: "which is the honest yardstick" removed; stated_chill_time_min, trade-off 2 and judgement 5 now call 11 to 20 minutes unmeasured blog claims, one from a chiller seller. Judgement 5 lowered from high to medium confidence, since its minutes and order rest on those claims.
6. Wine branch: makes_irrelevant capacity removed (Branches trace updated).
7. Coffee branch: then_ask prefreeze_time_h removed; no coffee chiller with a stated freeze time was found (HyperChiller pages checked above). prefreeze_time_h backfill note now says to leave it empty where no time is given. hyperchiller_v2 supports line added.
8. capacity: source changed from listing to researched (only a brand page backs it; no retail bucket listing was read, and the hint now says so). Definition and backfill note now store ounces and any stated count, match the buyer's bottle count only against stated counts, and show ounce-only buckets without filtering them out. why_experts_care "first fit question" reworded (ordering prose).
9. Judgement 4 (Vacu Vin): restated as different Vacu Vin product lines (current Cooler Sleeves 15 min, Rapid Ice listings 5, one titled a can cooler), not "even within one brand"; sources_disagree set to false because these are different products. stated_chill_time_min why_experts_care reworded to match.
10. Judgement 8: limited to ice and frozen water cores; stone and steel now said to depend entirely on freezer temperature (ours, from ice_equivalent_g). "Cooking Issues measured" changed to "says", since Arnold states it without a reported measurement in what was read.
11. bottle_fit: values made exclusive ("still wine only", "still and sparkling", "up to 1.5 L, sparkling too") with a largest-stated-fit rule; a fit line that does not mention sparkling is now left empty as unknown, not recorded as still wine only.
12. ice_equivalent_g: formula now volume × volumetric heat capacity, for solid stone or steel only; depends_on now cold_source, sizing.dimensions, sizing.pack_count, materials; why gives soapstone 2.2 to 3.0 J/cm³·K (Greenstone and Traditional Oven) and steel 3.7 to 3.9 (Wikipedia 3.756; ETB 0.49 × 7.48 to 8.0); says frozen cores get no figure. New reference etb_metal_density. Ice-equivalent bar grammar now covers stone and cube sets only.
13. drink_contact: 21 CFR 174.5(a)(2) quoted as "of a purity suitable for its intended use" (also in judgement 6 and the fda_21cfr_174_5 supports line); Teroforma's "Do not chew or ingest" added; the fish-odour sentence moved out (it stays in judgement 7) to stay under 80 words, and drinkhacker_2013 dropped from this facet's evidence.
14. Judgement 1: sources_disagree set to true; the note quotes Teroforma's "particularly effective at retaining temperature" claim and says we side with the two timed tests. Teroforma supports line added.
15. Summary: "Ask first" removed; it now says the drink and whether it starts warm decide which chiller can work, and steel chills a little further than stone. Optional part taken for vacuum_insulated only: source relabelled researched to derived (keyword match on title and bullets), its question changed from "Mind drips and a wet ring on the table?" to "Should the bucket or holder stay dry outside?" (gel sleeves sweat too), and the answer mapping written down. prefreeze_time_h stays researched: two Amazon listings are too few to call it a listing field; its answer mapping is now written down.
- Not from the check: `vertical` changed from `other` to `supplies`, since schema.json now allows it and the brief asked for it (see the top of these notes).
- Rejected findings: none. Not done: a coffee chiller's own volume still has no facet (checker's aside, not a required fix); insulated party tubs against `coolers` and insulated bottle holders against `bottles` stay unruled until those siblings have registers.
- Build after these fixes: validate, render and viewer all exit 0.