This is a Sankey diagram, which is a chart for showing how a quantity splits as it moves. It was invented in 1898 to draw the energy losses in a steam engine, and the most famous one ever made is older still: Charles Minard's 1869 drawing of Napoleon's army marching into Russia and dwindling on the way back. The rule in all of them is conservation. Nothing enters or leaves, it only divides.

A cocktail obeys that rule exactly. Whatever is in the glass is one hundred percent of the glass, and the interesting question is how that hundred percent divides. So the band on the left is the whole drink, the middle column is the ingredient families it splits into, and the right column is the ingredients themselves. Every width is a share of the finished volume, and every column sums back to the same total.

Search any of the 1,857 drinks in the collection, or take whatever the page opened on. Nothing is rounded away: every ingredient in the recipe gets its own band, including the ones measured in dashes.

The Negroni is supposed to be equal parts gin, sweet vermouth and Campari. That is the recipe as everyone repeats it. So I averaged all sixteen Negroni riffs in the collection, including the ones that deliberately break the rule, and the shares landed at gin 21.9%, sweet vermouth 18.1% and Campari 13.7%. The classic ratio is sitting right there in the aggregate, slightly gin heavy, with the remainder given over to whatever each variation swapped in.

I find that more convincing than being told the ratio. Nobody enforced it. It is just what sixteen people independently wrote down, and it reproduces a rule of thumb from the 1920s.

The chart on this page draws one drink at a time rather than an average, because an average is a shape and not a recipe. Look up a few Negroni variations in a row and you can watch that same ratio bend without breaking.

Width is share of volume, nothing else. It is not strength, not cost, and not influence on flavor. Two dashes of Angostura are a rounding error on this chart and can still be the difference between a drink and a mixture, which is exactly the thing a volume chart cannot show you. Read the small bands as presence rather than importance.

Where a source recorded a real quantity, the share is that quantity. Where it said "a dash" or "to taste", the share is estimated the same way the recipe breakdowns do it, so that nothing silently vanishes from the total. Roughly 40% of ingredient rows in the collection carry a real number.

Nothing is pooled or rounded away here. A single recipe is short enough to draw in full, and the smallest pours are frequently the ones doing the most work, so every ingredient gets its own band no matter how thin.

This chart is a port of one I built for food. That one takes 100g of an ingredient from the USDA database and splits it into water, protein, fat, carbohydrate and minerals, then splits the fat into saturated, mono, poly and trans, and the carbohydrate into sugars, fiber and starch. Same rule, same shape, different subject: a whole dividing into parts dividing into smaller parts.

Once you have built one you start seeing the shape everywhere, which is more or less the entire working method here. The connection wheel answers what goes with what. Cocktail search answers what is in this drink. The builder answers what can I make. This one answers what is a drink actually made of, which turns out to be the question the other three were circling.