What does professional blind tasting actually look like, step by step? This preview dataset holds 234 annotated steps across 4 tasting trajectories, recorded as working tasters carried 12-wine sets of white wine from first pour to final score.
We gave tasters from the Oenra network a tasting brief and recorded them carrying it through to a finished, ranked flight: coding and pouring the glasses, reading colour and clarity, swirling and calling the nose, judging attack and mid-palate, checking structure and length, retasting the close pairs, ranking the set, and logging the score. The tasters narrated their reasoning aloud as they worked, so every step's thought is transcribed from the expert's own voice.
Each step pairs a photograph with a first-person thought, a structured action, and reproducible grounding: a glass poured, a swirl, an aroma called out, a descriptor logged, or a point placed on the scale, with a preferred execution path marking the most deterministic option. The format follows the AgentNet trajectory schema, extended with a tasting action taxonomy and multi-path execution. This structure makes the data directly usable in several contexts:
- Palate preference modeling: long-horizon blind-flight trajectories with photographs, actions, and scores in a format interoperable with existing sensory-panel tooling.
- Taster calibration: pair photographs with narration-grounded reasoning to capture how experts actually plan, taste, and correct course mid-flight.
- Sensory scoring systems: every step carries a structured action, so a panel can reach the same verdict through a shared rubric instead of loose prose.
- Benchmarking and evaluation: a human expert baseline for long-horizon tasting work, reproducible end-to-end through the Oenra flight protocol.
The dataset was recorded in July 2026 in the Oenra tasting rooms. It is released under the CC BY 4.0 license on Hugging Face.