WineGraph Editorial · Data X-Ray
We ran the published wine lists of the 2026 James Beard winners through the WineGraph engine — no human palate, just the bottles. Here is what the data says about how America's most decorated restaurants actually drink.
By the WineGraph engine · 2026-06-18 · no-LLM, deterministic parse
The finding
The 2026 awards put two very different beverage programs on the same stage. Quince’s list — 1,412 bottles, a median of $685 and a top bottle at $14,250 — is a grand Francophile cellar: France is 67% of it, led by Burgundy (29%) and Champagne (17%), and it skews red (60%). It reads like a vault.
Borgo’s is the opposite instinct: 379 bottles, a median of $138, and an almost even France/Italy split (48% / 47%) tilted to Piedmont (17%) and a conspicuous 8% from the Jura. Where Quince collects, Borgo curates — a tight, Italian-leaning drinking list you could actually order off of.
Both numbers come from the restaurants’ own published lists, parsed the same way, the same week. No human tasted anything; the engine read every bottle line and counted.
The X-ray
Outstanding Chef
San Francisco, CA · Michael Tusk (chef/owner)
Origin — countries
Top regions
Color mix (1390 of 1412 bottles classified)
Outstanding Professional in Beverage Service
New York, NY · Lee Campbell (beverage director)
Origin — countries
Top regions
Color mix (56 of 379 bottles classified)
Method
Each list is the venue's own published document, parsed section-aware with WineGraph's deterministic (no-LLM) wine lexicon. Counts are signal hits over individual bottle lines; a bottle inherits its section's color/origin (so a Barolo under "Italian Red Wine: Piedmont" counts as red/Italy/Piedmont). No human curation.
Of the 24 award categories, full machine-readable bottle lists were found for the venues below. Most winners publish only a food menu; this piece grows as more lists are sourced.
One honest limit: color is read from explicit style words and section headers, so a list that names reds only by appellation under-reports color — note the “classified” base on each panel. Origin and region, which appear in nearly every bottle line, are robust. See the canonical standard and the dataset for how the graph is built.