What does a professionally written tasting note actually look like, across bottlings and regions? This dataset holds 35 finished tasting notes, one for each of 35 unique, synthetic bottlings across 4 wine regions, the coastal south, the limestone north, the alpine terraces, and the river valley, all delivered as scored notes. We gave tasters from the Oenra network a simulated cellar brief for each bottling, including:
- Estate name
- House style guidelines
- Appellation
- Bottle sample
From there, we asked wine professionals to write a final, on-style note that an importer would actually put in front of a buying team. Notes were scored on the 100-point scale and incorporated all house style features and inputs.
The published dataset is the cellar input and final output. Every bottling is represented once, so the set spans 35 distinct house identities and use cases, distributed across coastal south, limestone north, alpine terrace, and river valley bottlings.
The range of sensory conventions, including fruit-forward coastal whites, mineral-driven limestone reds, high-acid alpine bottlings, and dense river-valley blends represents the range of tasting work being commissioned by real producers to be carried out by real tasters. This structure makes the data directly usable in several contexts:
- Palate preference modeling: a curated corpus of real, on-brief scored notes for conditioning preference systems on professional structure, vocabulary, and house style
- Sensory scoring systems: real examples of descriptor hierarchy, fault flagging, and score integration
- Benchmarking & evals: a human-written reference set for measuring automated notes against professional output, by region
The dataset is released under the CC BY 4.0 license on Hugging Face.