Chopin competition 2025: detailed scoring

When the prizewinners of the XIX Chopin Competition were announced late on October 20th, my initial reaction was that the result seemed to be one that made the most people the least satisfied. Now that the detailed scores have been released and I've had some time to look at them, we can start to add some colour to the result.

I'll start with a disclaimer: the analysis that follows here isn't qualitative. I didn't get to listen to quite as much of the competition this year as I would have liked, and as far as I've been told by friends who did attend in person, the impressions one might have formed by watching the competition online differed markedly from perspectives from inside the Warsaw Philharmonic Concert Hall. Besides, the Chopin competition, just by virtue of its size and limited repertoire selection (which was even more limited this year!), tends to be one of the worst competitions to spectate. So I'm not going to insist that someone should have placed higher, or tell you that the results were wrong—I didn't hear enough pianists to be able to do that credibly.

But the numbers themselves tell a story.

Colourful scorecards

We can measure how much variation was in the jurors' assessments of each performance by looking at the standard deviation of scores. The standard deviation of the average performance was almost a half point above what it was in 2021:

Average standard deviation across all Stage I–III performances, 2015–2025

Here we plot the standard deviation of each Stage I performance against the adjusted average, or AAVG, which was used to determine the list of successful candidates. I've labelled the top four competitors by standard deviation:

Note how recognizable some of the names at the top are—and also how they come from every range of the scoring table. Officially, Pedro López Salas is at the bottom of the Stage I table, but gets a 22 from Yulianna Avdeeva anyway:

Pedro López Salas, Stage I scorecard (2025)

This might be one of the wildest scorecards ever at the Chopin competition, featuring close to every integer between 8 and 22, and a total of twelve distinct scores from among the seventeen jury members. That is the highest number of distinct scores on any scorecard since the Chopin competition moved to a 25-scoring system in 2015. Nine Stage I performances got at least ten unique scores from the jury in 2025, a mark only reached twice in the past two competitions.

Performances receiving the most distinct scores, 2015–2025

Right in front of López Salas for variance (and right behind him for number of unique scores) is Xiaoyu Hu, who has a whopping 11 scores clamped to a raw average of 19.65:

Xiaoyu Hu, Stage I scorecard (2025)

Hu's adjusted average for Stage I is virtually unchanged at 19.60, just enough to scrape into Stage II in 39th position, and he finishes bottom of the Stage II table at 40th. Meanwhile, Piotr Pawlak and Vincent Ong have high STDEV largely by way of an extreme outlier score each:

Vincent Ong, Stage I scorecard (2025)

Piotr Pawlak, Stage I scorecard (2025)

Disproportionate effects

Of course, we can't really discuss outlier scores without mentioning David Khrikuli. Khrikuli ticks all the boxes for controversy: he has the highest standard deviation in Stage II and Stage III; if you count only the number of perfect or near-perfect scores (24s and 25s), Khrikuli is fifth in Stage II and third in Stage III; and the cutoff for prizes is just in front of him.

He ends up just outside the prizes, because—as was identified in the discourse almost immediately following the release of the scores—he also has the lowest single score in both Stage II and III. Both are from Ewa Pobłocka. So how much did Pobłocka's low scores hurt his placing, in the end? Well, as it turns out, just enough. The final table looked like this:

Final ranking table with prizes (David Khrikuli highlighted)

Let's make one change to one score in the entire competition: replace Pobłocka's 12 for Khrikuli in Stage II with an "s" vote, so that it doesn't count anymore. Here's the table now:

Khrikuli moves just ahead of William Yang, the winner of the sixth prize. What if we do the same with Pobłocka's Stage III score for Khrikuli (a 14)?

Khrikuli ends up just ahead of Piotr Alexewicz and Vincent Ong and presumably shares the fifth prize along with them. And if you do the same for Pobłocka's other Khrikuli scores, Khrikuli ends up comfortably in front of Alexewicz and somewhat behind Tianyao Lyu.

Now, of course, competition scoring doesn't work like that; you don't get to just throw out marks that you don't like, or at least if you do that, you have to apply that procedure to everybody. (One could make a similar argument about, for example, Hyo Lee, who was close enough that even a single raw point in Stage III would have put him through to the final at the expense of both Alexewicz and Ong.) However, the point remains that one juror's set of scores was enough to knock Khrikuli off the list of prizewinners.

But wait: isn't the procedure of adjustment to the mean supposed to prevent outlier scores from unfairly affecting results like this? Well, yes—but here we get to see one of its flaws in action: the outlier score is included in the average, and therefore affects the range to which all scores get clamped. Let's dive into Khrikuli's Stage II scores:

David Khrikuli, Stage II scorecard (2025)

Khrikuli's raw average is an even (or odd?) 21 points, which means scores are clamped to the range between 19 and 23 points. Now, let's jettison Pobłocka's score:

Notice what happens? The raw average moves up to 21.56, so high scores are clamped less—the 24s are now only clamped 0.44 points instead of a full point—and the low scores are clamped to 19.56. In fact, the 19s are now clamped where they weren't before, because they're now out of range. In other words, by removing Pobłocka's score, we've effectively given Khrikuli an extra half point from nine other jury members, far in excess of the two points to which the clamping was supposed to limit a juror's dissent. The combination of Pobłocka's scores for Khrikuli ultimately sets him back 0.30 points in the final standings, which is enough to make the difference between a sole fifth prize and finishing out of the prizes entirely.

Interestingly enough, the Chopin competition itself previously employed a procedure to partially address this effect of outliers on the clamped range: in 2010, a separate "auxiliary average" was calculated from all non-outlier scores, and outliers were clamped a range around the auxiliary average instead. Using this system, the final table would have looked like this:¹

A divided jury

Then again, one of the reasons such sabotage was even possible in the first place was that this was the most contentious jury since at least 2010. (We don't have any data for 2005, but it can be safely assumed that the 2005 jury was also not particularly contentious.)

Normally, one would expect a relatively robust correlation between scores given by an individual jury member and the average of all scores given by the jury; after all, the individual score forms part of the input into the raw average. Only five times between 2010 and 2021 was the Pearson correlation coefficient between a jury member's scores and the raw average for a stage under 0.40. This happened twenty times in 2025:

Stages where the Pearson correlation coefficient between a juror's scores and the raw average was less than 0.40, 2010–2025

The low correlation scores suggest a significant breakdown in jury cohesion, which explains the enormous delay in arriving at a final verdict (since confirmed anecdotally by several members of the jury) and supports the idea of the final verdict being largely the result of compromise. Contrast 2015—where almost every jury member's scores have at least a moderately strong correlation with those of every other juror—with 2025, where by Stage III, some correlations between jury members even skew negative: the more Michel Beroff prefers a candidate, the less Garrick Ohlsson prefers the same candidate, and vice versa!

Pearson correlation coefficient between individual jury members, all performances, 2015 (top) vs. 2025 (bottom)

Pearson correlation coefficient between individual jury members, Stage III, 2015 (top) vs. 2025 (bottom)

Another indicator that points to disagreement between jury members is high scores given to competitors who were nevertheless eliminated: it means that other jurors found the same performance disappointing enough that it didn't pass in spite of the high score.

In 2015 and 2021, this happened seldom. Performances that received a 24+ score from any jury member were 40-for-43 (93.0%) in advancing to the next stage in 2015, and 41-for-49 (83.7%) in 2021. In 2025, that proportion drops to 70.9% (56-for-79), with twenty-three performances receiving a 24+ score and getting eliminated anyway. That means that twenty-three times during this competition, a jury member assessed a performance as being either literally perfect, or as close to perfect without being perfect—and then had a bit of a shock when that candidate didn't even advance to the next stage. And this happened to almost a third of such performances!

Performances receiving at least one 24+ score, 2015–2025

Five competitors even got eliminated after receiving multiple 24+ scores, which didn't happen even once in 2015 or 2021. Jack Gao got eliminated after a stage in which he got two perfect scores. Only once in the previous two competitions combined did a competitor get eliminated after receiving even one perfect score!²

Performances receiving at least two 24+ scores, 2015–2025

Gao's elimination can be explained by the fact that of all the performances since 2010 to earn two perfect scores (a list that, up until this year, contained only those of eventual prizewinners), his is the only one to receive no other marks above 22. With no other "supporting" scores to keep the raw average up, the adjustment procedure works against him: his raw average is an even 20—the lowest average of anybody with two perfect scores, by over a whole point—and Garrick Ohlsson's and John Rink's perfect scores are adjusted to 22.

Performances receiving at least two perfect scores, 2010–2025

Yang (Jack) Gao, Stage III scorecard (2025)

Two ways to win

And that brings us to the winner of the first prize of the XIX Chopin Competition, Eric Lu. With a jury as divided as this one, there's room to come up the middle, and Lu manages to do exactly that. To illustrate this, let's consider each jury member's maximum awarded score for each stage to be a metaphorical vote for "first place"—they didn't rate anybody else more highly during that stage. Lu wins only 13 possible first place votes out of 60 (four stages multiplied by fifteen jury members, given that Dang Thai Son and Robert McDonald don't assess him), or 21.7%. In Stage II—a stage in which Lu topped the table—it was William Yang who had the most jury members assign him their top score; he garnered seven top scores to Lu's four.

Lu's 13 individual maximums are by far the fewest achieved by a first prize winner since 2010. Scores weren't released for the final in 2021, but Bruce Liu already outdoes Lu's total in his first three rounds, being assigned a juror's top score in a stage 22 out of a possible 46 times (47.8%). Seong-Jin Cho went 37-for-67 (55.2%) in 2015, and Yulianna Avdeeva five years before that captured 52.1% of the top scores given by a twelve-member jury (25-for-48).

Number of times winners received the highest score from a jury member in each stage, 2010–2025

Almost everybody else struggles to command jury support as well, however. Kevin Chen has 19 individual maximums, but six of them are in Stage I, mathematically the least consequential; through the last three stages, he's tied with Lu on 13. (Also keep in mind that Chen, with no abstentions on the jury, has eight more opportunities to score individual maximums.)

With no consensus favourite, then, let's repeat this analysis, except this time counting a juror's highest and second highest score, exclusive of ties. Lu picks up a whopping 16 "second place" scores to go 29-for-60 (48.3%) in achieving the highest or second highest score given by each juror in each stage. Ironically, he's even second place in second placements: Shiori Kuwahara picks up seventeen "second place" scores, but she only has six "first" placements to begin with.

Number of times finalists received the highest or second highest score from a jury member in each stage, 2025

There's two ways to win the Chopin competition. One is by acclamation: everybody votes you first place, then you have higher marks than everybody else and you win. This year features the other method: when the jury is divided as this one, pick up a healthy share of the first votes, pick up a healthy share of second place votes, and receive as few low scores as possible. Lu receives only five scores lower than 20 in the entire competition, the fewest of any finalist:

Number of scores < 20 received during each stage, 2025 finalists

One might argue that Eric Lu won the XIX Chopin Competition because his competitors were simply too divisive. So does that mean the standard of competition was lower in 2025 than in previous editions? Well, we could perform a straightforward comparison of averages across competitions. There are many reasons not to, however: it's a different jury, there's cumulative scoring in 2025 where there wasn't in 2015, the yes/no system was abolished for 2025, and the final in 2015 was scored relatively from 1 to 10. But if you're willing to assume that scores after each stage of previous competitions were cumulative, and you compare the adjusted averages after 2015's Stage III against the weighted average after 2025's Stage III...

... 2025 Eric Lu would have finished fourth in 2015.

You can check out the detailed scores of the 19th Chopin Competition here.

¹ Of course, given the separation that emerges between Ong and Alexewicz under this system, they could have decided to place Alexewicz sixth and again leave Khrikuli on the outside looking in.
² Natalie Schwamová in Stage I, 2015, who received a 25 from Philippe Entremont.