Where the recipes come from

Source	Recipes	Licence	What it contributes
Bar Assistant	657	MIT	The primary corpus. Real numeric amounts, ABV, method, garnish, and the ingredient tree.
IBA Official Cocktails	2	MIT	The official specs, and the flag for which drinks are official.
TheCocktailDB	356	Free/attribution	Crowdsourced breadth. Modern drinks, popular drinks, and a lot of novelty ones.
Mr. Boston Official Bartender's Guide	842	CC0 (TidyTuesday)	Classic American recipes from the guide first published in 1935.

Bar Assistant

The primary corpus, and the reason this collection can do anything beyond listing ingredient names. Bar Assistant is an open source bar management app, and this is the seed data it ships to new installations. Alone among the four, it records quantities as numbers with units instead of as prose. 60 ml rather than "1 1/2 oz", which means proportions can be computed instead of parsed out of strings. It also carries alcohol by volume per finished drink, preparation method, garnish, descriptive notes, riff lineage linking a variation back to its parent drink, and a genuine ingredient hierarchy with categories and strengths. MIT licensed and actively maintained.

IBA Official Cocktails

The International Bartenders Association maintains the closest thing the drinks world has to a standards body. 90 official specs across three categories: The Unforgettables, Contemporary Classics and New Era Drinks. Almost all of them already appear in Bar Assistant, so this source adds very few new recipes. Its job is flagging which drinks in the collection are official, and settling the spec when sources disagree. Scraped and published as CSV and JSON under the MIT license.

TheCocktailDB

A long running crowdsourced database with a free public API, and the original source for this project. Its strength is breadth across the modern and the popular, including a great many drinks no curated set would ever admit. Its weakness is that measurements arrive as free text in mixed units, ingredient names are spelled inconsistently, and quality varies with whoever typed the entry. The free test key is documented for development and educational use, and production use of the API is expected to run on a supporter key. This collection ships a snapshot rather than calling the API from the browser.

Mr. Boston Official Bartender's Guide

First published in 1935 and reissued for decades after, the Mr. Boston guide sits behind a substantial share of twentieth century American bartending. This version comes from the TidyTuesday data project as a tidy CSV, one row per ingredient. It contributes the most recipes after Bar Assistant, and it is the reason the collection skews toward drinking from before 1970, back when powdered sugar, blended whiskey and grenadine did far more work than they do now.

Sources get consulted in the order above. Each recipe name gets reduced to a comparison key, with accents stripped, punctuation removed, a leading "the" and a trailing "cocktail" dropped, so that "The Old Fashioned Cocktail" and "Old Fashioned" are recognized as the same drink. The first source to claim a key supplies the recipe body. Later sources only add provenance: a note that the drink also appears there, and the IBA flag if it is on the official list.

That ordering is deliberate rather than alphabetical. Bar Assistant goes first because its records are the most complete, so a drink existing in all four sources shows up with numeric measures, ABV and a method rather than with vague quantities. It costs some coverage, since the same drink written two meaningfully different ways collapses into one entry, and I think that trade is worth making for a collection meant to be read rather than catalogued.

Ingredient names get normalized the same way everywhere. Lowercased, accents folded, a short alias table catching the common variants so that "fresh lime juice" and "lime juice" are one ingredient. Glassware names collapse to a canonical list, because four sources produce four spellings of a highball glass. Nothing else gets rewritten, and ingredient names are presented the way their source wrote them.

The wheel on the home page is not computed from this collection. It is drawn from a separate ingredient association matrix of 300 ingredients and 1,902 connections, published as a CSV and kept as its own artifact so the diagram stays stable while the recipe collection changes underneath it. The recipe collection powers the cocktail search and the ingredient builder. The matrix powers the wheel.

The merge is a single Python script in the repository. It fetches all four sources, applies the ordering above, regenerates the ingredient taxonomy from Bar Assistant's tree, and writes both of the files that ship with the page. It refuses to overwrite the collection if any source comes back with less data than expected, so a failed download degrades to a stale snapshot rather than a broken one.

Back to the cocktail chords, or read how the wheel is built.