Recommendations from what you actually smoke

Two things happen, and neither of them is a hidden model. First, the cigars you rated well get matched against the catalogue on their attributes — wrapper, origin, strength, body and brand — and cigars that share those attributes rise to the top. Second, your ratings get compared against other members' ratings, and cigars that members with overlapping taste rated well get surfaced too.

That's the whole mechanism: attribute matching plus a comparison against similarly-rated members. No magic, nothing you can't reason about, and no part of it that is not built from something you did.

A top-ten list is the same for every visitor, and it describes the average smoker — who does not actually exist. If you consistently rate Connecticut shades poorly and full-bodied maduros highly, a popularity list has no way to know that. It just keeps recommending the same generic middle of the catalogue.

Your ratings are specific to you. The more of them there are, the further the recommendations move away from generic and toward what you've actually shown you like.

The engine looks at a defined set of inputs from your journal, not your entire account history.

Cigars you have already logged are taken out. A recommendation you have smoked is not a recommendation, whichever way you rated it.

So are one-off customs. A shop roll somebody logged is not something you could buy, so it is never suggested to you.

And on the similar-taste half, only other members' public entries count. A private entry never surfaces its author to anyone, directly or through a cigar it caused to be recommended.

The picks come in three: For You, from the attributes you rate well; Similar Users, from members whose ratings track with yours; and Explore, which does the opposite of the other two on purpose — a wrapper you have never tried, a strength outside your usual — because a profile built only from what you already like narrows until nothing new gets in.

Separately, every cigar's page lists the cigars closest to it, which works whether or not you have logged anything.

In the app, each pick carries the short reason it was picked — the wrapper, the origin, the brand, or how many members with similar taste rated it well.

The first few entries move it further than the hundredth will. A handful of ratings is already enough for the attribute half to have something to match on, and the picks visibly stop being generic.

From there it keeps sharpening. We would rather say that plainly than promise pinpoint accuracy on your first session, which would not be true of any system built on data it does not have yet.

Recommendations are a starting point, not the only way to find a cigar. Browsing, filtering and searching the catalogue directly is always available, whether or not you've logged a single entry.