Policy
The taste model
Everything WineHQ can do — recommend, compare, pair, find — rests on one structure. This is that structure, stated openly, because a recommendation engine you cannot inspect is just an opinion with a database behind it.
Ranges, not points
The question “what does Chardonnay taste like” has no single answer. Chablis is lean, taut and unoaked; a warm-climate Chardonnay can be broad, buttery and tropical. Both are honestly Chardonnay. A model that stores one number for Chardonnay’s body is not simplifying — it is asserting something false about half the wines in the category.
So every profile stores three values per dimension: a realistic minimum, a typical value and a realistic maximum, each on a 1–5 ordinal scale. Variability becomes data rather than a caveat bolted on afterwards, which means the site can say “this one varies a great deal” because it can see that it does. Similarity is measured between bands, with overlap discounted, rather than between single points.
These are ordinal positions, not measurements. A tannin value of 4 does not mean twice as much tannin as 2; it means “noticeably grippy” against “soft”. Every dimension has written anchors for each point so the scale means the same thing to every contributor, and no arithmetic is performed on them that would require them to be interval data.
The sixteen dimensions
| Dimension | What it measures | Low | High |
|---|---|---|---|
| Sweetness | Perceived sugar in the finished wine, driven mainly by residual sugar but modified by acidity, alcohol and fruit ripeness. | Bone dry — no perceptible sugar | Lusciously sweet — concentrated, syrupy, sipping sweetness |
| Acidity | Total perceived acidity. Rises in cooler climates and with earlier picking; falls with heat and ripeness. | Low — soft, broad, little refreshment | High — sharp, racy, strongly refreshing |
| Tannin | Phenolic compounds from grape skins, seeds, stems and oak. The principal structural axis of red wine. | Very low — effectively absent, as in most white wines | High — powerfully drying and structural when young |
| Alcohol | Perceived alcoholic strength, which tracks actual ABV but is modified by sweetness, acidity and body. | Low — roughly under 11% ABV, or a deliberately light style | High — above 15%, or fortified |
| Body | The overall weight and viscosity of the wine in the mouth — a composite of alcohol, sugar, extract and glycerol. | Light — delicate and watery in weight | Full — dense, weighty and powerful |
| Fruit intensity | How loudly the fruit flavours speak, independent of whether they are ripe or tart. | Subtle — fruit is a background note | Exuberant — fruit dominates the whole wine |
| Aromatic intensity | How strongly the wine smells before it is tasted. Some grapes are intrinsically aromatic; others express mainly on the palate. | Neutral — very little to smell | Intensely aromatic — leaps from the glass |
| Oak influence | Perceptible impact of barrel or alternative oak contact — vanilla, toast, coconut, smoke, cedar and added texture. | None — no discernible oak character | Dominant — oak is the first thing you notice |
| Savoury character | Non-fruit, umami-adjacent character: meat, leather, olive, soy, broth, cured notes. Often increases with age. | Purely fruit-driven | Strongly savoury and umami-rich |
| Floral character | Blossom, rose, violet, elderflower, acacia and related aromatic notes. | Absent | Perfumed and dominant |
| Herbal / green character | Herbs, grass, blackcurrant leaf, capsicum and other green notes, largely driven by methoxypyrazines and picking decisions. | Absent | Emphatically green and herbaceous |
| Spice | Pepper, clove, cinnamon and related spice notes, whether from the grape (as with rotundone in Syrah) or from oak. | Absent | Emphatically spicy |
| Earthy character | Forest floor, mushroom, wet stone, clay and undergrowth notes. Frequently, though not exclusively, associated with cooler and more traditional styles. | Absent | Emphatically earthy |
| Perceived freshness | The overall impression of liveliness — a composite of acidity, alcohol balance, fruit character and the absence of tiredness. Distinct from raw acidity: a wine can be high in acid yet taste heavy. | Heavy and unrefreshing | Zesty and invigorating |
| Carbonation | Dissolved carbon dioxide, from completely still through lightly prickly to fully sparkling. | Still — no perceptible bubble | Vigorously sparkling with a persistent bead |
| Texture | Mouthfeel independent of weight — from lean and cutting through to creamy, waxy or velvety. Driven by lees contact, malolactic conversion, skin contact and ageing. | Lean and cutting | Velvety, waxy or notably unctuous |
Grape and style are different things
This is the modelling decision most wine sites skip, and skipping it is why so much wine content is vague. A grape is a plant. A style is a finished wine — a grape or blend, grown somewhere specific, made a particular way. Sauvignon Blanc is a grape; Sancerre and Marlborough Sauvignon Blanc are styles, and they taste substantially different.
So taste profiles live primarily on styles, where they can be specific. Grapes carry an envelope: a deliberately wide band describing what the variety can do across its range, with the caveat attached. When you ask WineHQ what to try after Malbec, it knows whether you mean the grape as a category or a particular expression of it, and answers accordingly.
Directions, not just similarity
“Similar to X” is rarely what a reader wants. They want like this but cheaper, or like this but less oaky, or like this but I can drink it on a Tuesday. WineHQ models 28 named directions, each with an explicit definition of which dimensions it shifts and which it must hold constant, so a request for something less tannic cannot return something more tannic that happens to score well on everything else.
Every score decomposes
No engine on this site returns a bare number. A similarity score is a sum of named components — profile distance, shared grapes, shared origin, curated relationship, direction satisfaction, and penalties for colour and category mismatch. A pairing score is a sum of the pairing principles that fired, each naming itself. A finder result carries both its reasons and its compromises.
This is not decoration. It is the constraint that keeps the model honest: a component that cannot be explained in a sentence is a component that should not be in the sum, and several were removed during development for exactly that reason.
Known limitations
- The model describes styles and varieties, not individual bottles. Producer, vintage and site variation are all real and all outside its scope.
- Bands are editorial judgements anchored to written definitions, not laboratory measurements. Two careful people would place some of them differently.
- Aggregating sixteen dimensions into one distance necessarily loses information. That is why the components are always shown alongside the total.
- Preference is not modelled. WineHQ can tell you what is similar and why; it has no view on what is good.