Smarter bottle label scans

When the scanner isn’t 100% confident about a wine, it now shows a list of likely matches with three little dots per result — one for the winery, one for the wine, one for the vintage — so you can see at a glance why each option is being suggested.

The scanner was confidently wrong too often. We’d see a Trisaetum label, the AI would pick out “J Fray Wines” from small text at the bottom, and the scan would resolve to that — a real winery in our database, but not what’s actually in your hand. Or it would see “Pangloss Cellars” and match to “Pangloss” instead because of a duplicate listing. The old single-score match was forced into one verdict; if the verdict was wrong, you had to scan again and hope it landed differently.

Now the scanner shows you the top candidates. Each card has three small dots — green / yellow / gray — for winery confidence, wine confidence, and vintage confidence. You can see which candidates we’re sure about and why. Tap the one that matches. The auto-match still kicks in when there’s a clear winner, but you can always tap “See alternatives” to second-guess us. The same picker works on the “remove from cellar” scan flow — snap a label, see candidates, pick the right one to remove.

The matcher now thinks about three things independently rather than collapsing to one score. Listing confidence answers “which winery is this?” — wine confidence answers “which of that winery’s wines is this?” — vintage confidence answers “does the year on this label exist in our system?”. They don’t always align, and that’s the point: you can have a strong wine match for a winery we’re only 70% sure about, which is genuinely useful information for picking the right answer.

We also added a label-prominence signal. Big block letters on a wine label are almost always the brand or wine name; small marginal text is usually credit lines. The matcher now uses the visual prominence of each text fragment when ranking candidates, which fixed cases like “Evolution Pinot Noir by Sokol Blosser” — Evolution dominates the label, so we score that as the primary signal rather than the small “by Sokol Blosser” credit.