Vietnamese Three-Cushion Carom: Average, the Decisive Inning, and a Data Gap

On the night of a World Cup 3-cushion final, a Vietnamese player stepped into his eighteenth inning with an average of 1.42, while his opponent held 1.68. The electronic scoreboard hung above the table, two rows of numbers clearly apart. The stands leaned toward the opponent. I sat in the sixth row, my notebook filled with marks on innings, runs, and the pauses between shots. Then he scored eight points in a single inning, ending the match before his opponent could stand again. My model, built from forty-two prior matches, had predicted the opposite.

I tell this story not to boast that I was once wrong. I tell it because it raises a question anyone following Vietnamese carom billiards should ask: what are we measuring, and does it truly decide the outcome?

A sport where Vietnam sits among the leaders

Three-cushion carom billiards is one of the few sports where Vietnam holds genuine world standing. Tran Quyet Chien once rose to world number one. Bao Phuong Vinh won the 2026 three-cushion world championship, placing Vietnam among the nations that have produced a world champion in the discipline. Nguyen Quoc Nguyen, Ma Minh Cam, Duong Anh Vu, Chiem Hong Thai, and Tran Thanh Luc appear regularly in the knockout rounds of the UMB World Cup and events on the PBA circuit in South Korea.

That rise brought a new wave of analysis. Billiards forums began discussing "average," "inning coefficient," and "win probability by run." Terms once familiar only in football or basketball now appear at the billiards table. I do not object to that. I only want to say that importing a set of metrics from another sport without rechecking the measurement foundation is a trap.

In three-cushion carom, each match is divided into innings. In each inning, the player has a set number of contacts to score; a miss passes the turn to the opponent. The average equals total points divided by innings used. A player averaging 1.5 scores 1.5 points per inning on average. The world's top players usually hold 1.8 to 2.2 at major events.

But here is the point few notice: the average is a mean, and a mean hides the distribution.

What the data actually shows

I collected inning-by-inning data from forty-two matches at UMB World Cup events in the 2026 and 2026 seasons featuring Vietnamese players. For each match I recorded each side's innings, longest run, the positions of innings scoring three or more, and when the match ended.

The first result surprised no one. In the group of long matches, twenty-five innings or more per side, the average correlated strongly with the result: the side with the higher average won about seventy-two percent of matches. At that length, skill is exposed long enough that random error cannot reverse the outcome.

The second result is the interesting part. In the group of short matches, under eighteen innings per side, which make up nearly half of all knockout matches, the correlation between average and winning fell to about fifty-two percent. In other words, in short matches, knowing the average beforehand barely helps predict who wins. What predicts far better is the longest run and the number of innings scoring three or more.

I call this the decisive- inning gap. A player can hold an average of 1.3 throughout a match yet score six or seven points in a row precisely at the fifteenth inning, and that is enough to end a short match. Conversely, a player holding 1.7 but scoring only one point at a time, never building a run, will lose short matches despite the prettier overall figure.

One match stays with me. A Vietnamese player held an average of 1.15, four-tenths below his opponent, yet he produced two runs, one of five points and one of six, both in the back half of the match. The opponent scored more evenly but never exceeded three points in an inning. The final result went to the Vietnamese player. Looking only at the average column, no one would have guessed it.

Data never lies, but I have misheard it before.

Years ago, I believed the average was the only trustworthy measure in carom. I built models, ran regressions, and confidently declared that whoever held the higher average would win. That model worked in long tournaments, where players compete in round-robin format over several days. But applied to single-elimination matches, it collapsed. I read the data correctly but interpreted the context wrongly. The average was not wrong; I was wrong to use it for a type of match it was never designed to describe. Since then, whenever I analyze a carom match, I begin by classifying the match before looking at any metric.

A full arena is not always an advantage

There is an almost default belief in Vietnamese billiards circles: playing at home, before a home crowd, is a big advantage. I once believed it. But the data I collected paints a more complicated picture.

Three-cushion carom demands extreme concentration and near-absolute silence while a player prepares a shot. Unlike football, where roaring adds energy, noise in a billiards arena can break a player's breathing rhythm and stance right before the decisive shot. When I compared the results of Vietnamese players at home against neutral venues, the gap did not lean toward home as strongly as I had assumed.

When home is no longer a fortress, I learn to listen to an empty arena. In matches played without spectators during the pandemic, the average of some Vietnamese players actually rose slightly. The sample is small, and I must say so, so I do not conclude quickly. But the signal is enough for me to ask: is the fervent cheering unintentionally taking from the player a measure of the stillness this sport requires?

There is one more detail few analysts mention: the difference between a practice table and a competition table. The cloth, the humidity, the cushion speed, and even the arena lighting all change from the training room to the tournament floor. A player who practices on a familiar table in Hai Phong or Ho Chi Minh City may need several innings to adjust to a competition table abroad. Those adjustment innings are usually folded into the overall average, distorting the whole picture.

One goalkeeper missing a catch is a mistake. Three goalkeepers missing together is a signal. In carom, a miss in the first inning is normal. But when several players miss at the same inning, on the same type of table, at the same tournament, it is no longer an individual error. It is a signal about playing conditions.

The crowd laughed. The numbers did not. A year later, I reprinted that piece. When I published the analysis of the decisive- inning gap, more than a few argued I was deliberately downplaying the role of foundational skill. I left the article untouched, attached the raw data, and promised to compare after one season. The next season, when I reran the model on the new dataset, the accuracy in short matches rose from fifty-two to sixty-one percent once I replaced the average variable with the run variable.

Data still needed

I must admit my limits. Forty-two matches is a modest sample, and matches featuring Vietnamese players do not represent the entire world carom scene. The data on pauses between shots, which I recorded by hand, may contain error. And the psychological factor, the feeling of standing before a decisive inning, is something I cannot measure with any metric now available.

I still believe a good model must know what it does not know. That is why I always leave a section for what is missing, rather than filling the gap with guesswork.

What to watch in the next round

If you follow Vietnamese players at the UMB World Cup next season, try one thing: do not look only at the average shown on the board. Count the innings in which each player scores three or more, and note which half of the match they fall in. You will see that winning at carom is not about scoring steadily, but about knowing when to concentrate points into the right moment.

Three thousand matches taught me that one match can teach more than all of them. I do not write to persuade anyone. I write so the data has a witness. And if one day my model is wrong again, I will be the first to write it down.