Match Score: which rum suits you?

The more you rate, the closer it gets: for people who have rated more than 100 rums, the prediction is on average less than half a point away from the rating they give afterwards, and with five ratings it is still about a full point off. Below five ratings the app deliberately shows no percentage at all, because at that point the community rating is simply the better guess. Those are the figures of the evaluation from September 2026 in the table below.

How do you check something like that? You take ratings the app has never seen, let it guess, and compare the guess with the real rating. Hundreds of thousands of times. It is the same method Netflix used in 2006, when it put up a million dollars for whoever could best predict how a member would rate their next film from a hundred million ratings. The starting point for everyone was the average per film, exactly our one rating for all. Netflix itself was ten percent better than that average, and after three years the winners were another ten percent better than Netflix. One of their most important tricks: taste changes over time. That is why recent ratings count more with us too.

The app also keeps score on itself. Every time it shows you a prediction and you rate the bottle afterwards, it stores both numbers. By mid-September 2026 that had happened about 19,600 times. Across those real cases the picture is honest rather than flattering: for people with more than 100 ratings the prediction is about 4 points off on the 100-point scale, and above 90 percent the app is still a good 3 points too optimistic. That is where the tuning work goes.

It helps most if you are just getting into rum, and least if you already know your shelf. As a beginner you are exactly the group for which the community rating is a full point off in four out of ten cases, and you are probably still looking for your own taste, which is completely normal. Rate the five rums you already know, including the ones you did not like, and from then on the whole community works for you.

If you already have two hundred bottles on the shelf, you usually know yourself whether a cask-strength Hampden is your thing. For you the Match Score is more of a cross-check, and on brand-new bottles with one or two ratings it cannot show its potential yet. I could tune the score sharper for us geeks, distillery, vintage, cask, all of it in the calculation. I built that and tested it, and the result was classic overtuning: better for a few special cases, clearly worse for everyone else. So no, the Match Score is built for the many.

The community rating tells you how good a bottle is in everyone's eyes, the Rum Finder recommends three bottles without an account, and the Match Score predicts in the app whether a specific bottle fits you. The community rating is a median, it starts at the fourth rating and it stands on every rum page on this website. The Rum Finder answers which bottle to pick when you cannot or do not want to rate first: four questions about how you drink, which aroma world you lean to, how strong it may be and what you want to spend, then three bottles that ship to your country today. It needs no account and no ratings and also works as a gift path. The Match Score does not ask; it learns from your ratings and, from five ratings of your own, puts a personal number on every rum in the database.

Honestly and completely: rate what you did not like as well, because one honest 4 usually tells the app more than ten further 8s, and set the tasting tags, not only the points. Someone who only ever gives 7s and 8s has a flat profile for the app; an honest 4 weighs more because there is a separate dislike profile that actively pulls the Match Score down for bottles in that direction. Your ticks count twice, once for your own profile and once for the profile of the bottle.

Because the community rating is the middle of everyone, while taste varies: held against 245,789 individual ratings, it sits at least a full point away from what the taster gives in every 6th case. For eight years RumX had exactly one number per rum, the community rating, the median of all ratings. Nobody can buy it, no brand, no shop, it belongs to the community, and it remains the best answer to the question how good a bottle is. But rum is the most diverse spirit there is. Among beginners the full-point gap shows up in 4 out of 10 ratings, among connoisseurs with 50 or more ratings only in every 9th. And three out of four ratings come from those connoisseurs, so the community rating is above all a connoisseur rating. If you deviate from it, you are not wrong. You are just not the middle.

The differences start with the nose. In 2014 researchers compared the odorant receptor genes of more than a thousand people. Each of us carries about 400 of these genes, and any two people differ functionally in over 30 percent of them (Mainland et al., Nature Neuroscience 2014). Experience adds the second layer: an analysis of 4,300 professional whiskey reviews found that experts describe expensive bottles with leather, tobacco and oak while caramel stands for cheap ones, yet when 520 consumers were asked what they would buy, leather and tobacco came last and caramel first (Hamilton, Neill and Lahne, Food Quality and Preference 2023). Taste is not a question with one right answer. The Match Score is the attempt to put that personal factor next to the rating instead of into it.

In April 2026 the app showed six out of ten people more than 90 percent Match Score for a bottle almost nobody had tasted, and the RumX forum spotted it before I did. The Match Score had just shipped in an app update when someone posted a list there: five bottles, all above 90 percent, and he would not buy a single one of them. Others chimed in, 93 percent here, the same list there. So I looked. One of the bottles had exactly one rating, a perfect 10 out of 10 from a single person, plus thirteen people who had it in their collection without a rating. Collection data is a weak signal for the Match Score, but together with that one perfect rating the app concluded this must be the best rum in the world and handed that bonus to everyone.

That was my mistake, not the community's and not the bottle's. Two days later it was fixed, nobody saw the 90 percent on that bottle any more, and since then the third rule applies: below three community ratings there is no Match Score at all. Rather an honest "I don't know yet" than a confident wrong number. The hits came from the same thread: one member had seven ratings in, the app said 91 percent, he opened the bottle and gave it 92 out of 100. That is how it is supposed to work. The thread is still open, and the misses interest me more than the hits.