How we hand-transcribed fifty pre-Prohibition cocktails.

A working methodology page for the editors, historians, and fact-checkers who will read the Codex and want to know what we did and did not check.

Why hand-transcription, not OCR

The first attempt at the Codex used Tesseract OCR plus a Claude vision pass to clean up errors. We threw the output away after sampling fifty random recipes and finding errors in eleven. The 1908 Boothby World's Drinks page that prints the Sazerac has a column-break mid-recipe; OCR scrambled the gomme syrup volume across two paragraphs. The 1917 Ensslin Aviation entry uses an old-style long-S that read as "f," which turned simple syrup into fimple fyrup. Most damaging, the gum-syrup abbreviation g.s. read as g.f. in three places, which an unaware reader would interpret as a different ingredient.

Modern reprints have the same problem. The 1928 facsimile reprint of Jerry Thomas's How to Mix Drinks normalized "wine-glass" measurements to "ounces," which appears tidy until you realize a wine-glass in 1862 was 2 oz, not 5 oz. Several recipes in modern cocktail books trace their wrong proportions back to that single normalization. The 1880 Manhattan was 1:1 whiskey-to-vermouth in Engel's printing; modern books call for 2:1 because the modern books were written from the wartime drift, not from Engel.

Hand-transcription is slow and not glamorous. It also catches the things that automated tools propagate.

The five source manuscripts

All five are public domain. We worked from scans held by the EUVS Library (euvslibrary.com), the Internet Archive, and HathiTrust. Where multiple printings exist (Johnson 1882, 1888, 1900), we noted the printing in the source line and resolved discrepancies by majority printing or, where one printing was clearly the corrected version, by the corrected printing. Page numbers in the Codex citations refer to the EUVS facsimile edition unless otherwise noted.

Transcription rules

1. Preserve the original wording verbatim, including quirky capitalization and the period dashes. Where an em-dash exists in a 1908 page, we left it. Where the printer used a long-S, we silently substituted the modern s. Where a recipe name was set in display type and the body text was set in a different type, we transcribed the body and noted the display variant.
2. Resolve ambiguous abbreviations explicitly. G.S. = gum syrup. P.B. = Peychaud's Bitters. O.B. = orange bitters. We expanded each abbreviation in a footnote on first appearance per recipe, but kept the original in the body so the page reads the way it was printed.
3. Note printer's errors as printer's errors, not as recipe history. Boothby's first 1908 printing has a typo in the Sazerac proportions that was corrected in the second printing later that year. The Codex prints the corrected version and notes the typo in the footnote.
4. Provide a measured modern build alongside the original. Wine-glass = 2 oz. Pony = 1 oz. Dash = roughly 1/8 teaspoon for bitters, more for syrups. Where the original specifies a non-standard glassware (e.g., "in a goblet"), we kept the goblet reference but added the modern glass equivalent. Where a recipe specifies a discontinued ingredient (e.g., Old Tom gin, calisaya, Crème Yvette), we note the closest modern substitute and flag the substitution as a substitution, not a continuity.
5. Cite the source line on every recipe. Author, title, year, page. Historians read the page first.
6. Catch the discrepancies the modern reprints carried forward, and flag them. See below.

What we caught

The Codex prints the original specs of these famous cocktails as the source manuscripts wrote them. The modern reprints we cross-checked against frequently print a different version, and in many cases the modern version is a wartime drift or a post-Prohibition reconstruction. We do not say "modern is wrong"; we say "modern is different, and here is when the difference began."

The Old-Fashioned (Thomas 1862)

Thomas's 1862 entry calls for whiskey, sugar, two dashes Boker's bitters, lemon peel. No orange. No muddled cherry. The orange entered the recipe in the 1900s, possibly through Harry Johnson's 1882 manual where the lemon-or-orange option appears as a footnote. The muddled-fruit version is a 1940s Wisconsin invention and is not in any source manuscript before that.

The Sazerac (Boothby 1908)

Originally cognac, not rye. The 1870s phylloxera blight in France killed brandy supply; American bartenders swapped to rye and the swap stuck. By 1908 Boothby is printing both versions; the Sazerac House in New Orleans was using rye by then. Both are correct period spec. The Codex prints the cognac version as primary and the rye as the historical pivot.

The Manhattan (Engel 1878, Johnson 1882)

The 1880 Manhattan was 1:1 whiskey-to-vermouth, not 2:1. Engel 1878 spells it out: half a wine-glass of each. Johnson 1882 prints the same proportions. The 2:1 modern spec is a wartime drift that crystallized post-Prohibition. The 1:1 is sweeter, softer, more Italian, and reads as the Manhattan its inventors actually drank.

The Aviation (Ensslin 1917)

Ensslin's 1917 entry includes Crème de Violette as the second modifier. The ingredient was effectively unavailable in the United States from 1960 (when the last importer stopped) to 2007 (when Rothman & Winter relaunched it). Every Aviation poured for nearly fifty years was missing the ingredient that gave the drink its lavender color and its name. The colorless gray-Aviation common in 1980s bar manuals is the wartime version, not the original.

The Daiquirí (multiple)

The Hemingway daiquirí, mixed by Constantino at El Floridita in Havana from the 1930s, has no sugar by request. Hemingway was diabetic. The grapefruit-and-maraschino addition is also Constantino's. Modern recipes usually print the sugar back in.

The Mint Julep (Thomas 1862)

Thomas's 1862 julep specifies cognac, not bourbon, with a Jamaican rum float. The bourbon-only version is a Reconstruction-era American adaptation; the original is colonial-era and substantially different.

What we did not do

• We did not generate any recipes. Every recipe in the Codex is from a printed manuscript, transcribed by hand. There are no AI-written cocktails in the Codex.
• We did not normalize jiggers. A 2-oz wine-glass is a 2-oz wine-glass. The modern build column converts; the original column does not.
• We did not silently correct historical sourcing claims. Where Boothby attributes a recipe to a person who almost certainly did not invent it (the 1908 attribution of the Bronx to Joseph Sormani at Waldorf-Astoria, for example), we kept the attribution and noted the contested history in a footnote.
• We did not include any post-Prohibition recipes in the pre-Prohibition section. The Mai Tai (1944), Singapore Sling (post-1915 attribution disputed), and Margarita (mid-century) are all out of scope for the Codex.

Source verification

Every recipe in the Codex was cross-checked against at least two of: the EUVS facsimile, the Internet Archive scan, the HathiTrust scan, and an independent secondary source (most often David Wondrich's Imbibe!, PUNCH magazine's deep dives, Robert F. Moss's spirits writing, or Camper English's Alcademics). Where two sources disagreed, we cited both and printed the version that the older or more authoritative source carried.

We are not professional historians. We will be wrong somewhere. If you find an error, email [email protected] with the page reference and the source you used, and we will correct the next edition. The Codex is a 2026 first edition; the second edition will fold in reader corrections.

Tools and time

The transcription took six months of weekend work. Each manuscript took roughly three weekends to read straight through and another two to cross-check against modern reprints. We used:

• A 27-inch monitor with the manuscript scan on the left half and a plain-text editor on the right.
• The EUVS Library, Internet Archive, and HathiTrust as primary sources.
• David Wondrich's Imbibe! (2007, revised 2015), PUNCH magazine's pre-Prohibition coverage, and Robert F. Moss's writing for cross-checking.
• A spreadsheet of the proportions, with one column per source, so discrepancies surfaced by sorting.
• No AI-written prose in the Codex itself. The methodology footnotes (this page) and the Codex book copy were drafted by Kyle, edited and tightened with a Claude assist, and re-checked against the source manuscripts before publication.

How to use this for your work

If you're a writer, journalist, or historian researching pre-Prohibition cocktails, the Codex is free to cite. The recipe text is in the public domain (we transcribed it; we do not own the source recipes). The methodology footnotes and the side-by-side modern builds are released for editorial use under a non-exclusive license; please credit Speakeater and link speakeater.com/codex.

For commercial reuse outside editorial coverage, contact [email protected].