Wine is rated out of 100.

It uses about eight of them.

Across roughly 130,000 Wine Enthusiast reviews, nothing at all is published below 80, and 86% of every wine reviewed lands between 84 and 92. The bottom four fifths of the scale are decorative. What is left is a narrow band in which price buys points at a steeply diminishing rate, and in which the name of the reviewer matters about as much as a tenfold change in price.

Read the short version or the code and data

the hard floor. Not a single published review sits below it, so the usable scale starts four fifths of the way up.

86%

of all wines score between 84 and 92. Nine points hold almost the entire catalogue.

r = 0.61

points against log price. Real, but logarithmic: $10 buys about 85 points, $100 buys about 92.

3.1 pts

between the most and least generous reviewer. On an eight-point scale that is not a rounding error.

Start · the question

What does 92 points actually mean?

Wine scores are treated as an absolute measure. A 92 on the shelf tag is read as "92 out of 100," which implies there is a 62 out there somewhere and that this wine is comfortably above the middle. Both implications are wrong, and the size of how wrong is worth measuring.

The public winemag dataset makes it checkable: about 130,000 reviews, each with a score, a price, a vintage, and the taster who wrote it. That is enough to ask what the scale actually does, what price buys, and how much of a score is the wine versus the person drinking it.

↓ the shape of the scale

A floor at 80 and a spike in the middle

8 pts

the effective width of a scale advertised as 100 points.

The distribution is not a bell curve spread across a hundred points. It is a narrow hump sitting on a cliff. Nothing below 80 is published at all, which is a publishing decision rather than a fact about wine, and the mass piles up between 84 and 92.

This is the structural version of "inflation." No year-over-year creep is needed for the scale to be useless at the bottom; it simply never uses the bottom.

Histogram of wine scores showing a hard left edge at 80, a tall peak between 86 and 90, and a thin tail extending toward 100. — The published distribution. A cliff at 80, almost everything in the high 80s.

↓ what money buys

Price works, logarithmically

Expensive wine does score better, and the relationship is strong for something this noisy: r = 0.61 against log price. But the log is the story. Going from $10 to $100 — a tenfold increase — moves the expected score from about 85 to about 92, roughly seven points. The next tenfold buys far less.

Read the other way, the cheap end of the market is where the points-per-dollar lives, and nothing in this data suggests a $400 bottle is meaningfully better rated than a $100 one.

Scatter of price against points on a logarithmic price axis, with a rising curve that flattens noticeably above about one hundred dollars. — Points against price on a log axis. The curve flattens hard past $100.

↓ two things that should matter and don't

Vintage barely moves. The taster moves a lot.

Average score by vintage is essentially flat, drifting about 0.09 points per decade. Whatever people mean by good and bad years, it does not show up in the mean of a large catalogue.

Reviewers, by contrast, differ by 3.1 points between the most and least generous. On a scale that is genuinely about eight points wide, that gap is a third of the whole usable range, from nothing but a change of author.

Mean score by vintage year, an almost horizontal line with minor year-to-year wobble and no visible trend. — **Vintage** — flat, about 0.09 points per decade.

Mean score by individual taster, showing a spread of about three points between the most and least generous reviewers. — **Taster** — a 3.1-point spread, person to person.

↓ what this cannot tell you

One publication, one snapshot

Everything here comes from a single snapshot of a single magazine. That matters most for the word "inflation." Because the data carries no review date, this cannot measure scores creeping upward year over year, which is what most people mean by the term.

What it measures instead is the structural version: an 80 floor, a compressed band, and a scale whose advertised range is mostly fiction. That is a real and separate finding, and the page says which one it is rather than letting the headline imply the other.

Vintage is parsed out of each wine's title, so a small number of titles that do not carry a year are simply absent from that chart rather than guessed at.

Finish · how it was built

Where the numbers come from

Data: The public winemag-data-130k-v2 Wine Enthusiast dataset, via a GitHub mirror
Sample: ~130,000 reviews with score, price, title and taster
Price model: Points against log price, since the raw price distribution is extremely skewed
Vintage: Parsed from the wine title; rows without a parseable year are dropped from that view only
Built with: Python, pandas, matplotlib
Related: The same 130k reviews become a retrieval index in Wine Sommelier RAG and the search tool inside the Wine Pairing Agent