Billiards Needs a Discipline Identification Standard Before Advanced Data

On an internal spreadsheet at a billiards club in Ho Chi Minh City, two rows sit side by side. The first logs a 9-ball match that ended 7-5 after twelve racks, including three break-and-runs. The second logs a three-cushion match that ended 40-38 after thirty-two innings, averaging 1.25 points per inning. Both rows feed the same column: billiards, won. Staff use that column to build a monthly internal ranking. A ranking built from two units that cannot be converted into one another. The error is not in the data entry. It sits in the fact that four different sports are filed under a single word.

Billiards in Vietnam is a family of disciplines. Carom three-cushion is measured in innings and points, where an average above 1.5 points per inning belongs to the world's leading group. Pool is measured in racks, and 9-ball differs from 10-ball in how the final ball is handled. Snooker is measured in frames, with its own index set covering pot success, safety success and century breaks. Chinese 8-ball sits between them, played on a smaller table than snooker while still grouping balls into sets. Players such as Tran Quyet Chien in three-cushion and Duong Quoc Hoang in pool represent two unit systems that do not translate. Spectators call all of it billiards, and that is accurate at the table. It only breaks down inside a data room.

I have been logging three-cushion and pool matches in Ho Chi Minh City since 2026, initially to answer one narrow question: does a player win consecutive matches through skill, or through a favourable sequence of innings. After more than two hundred hand-recorded matches, I realised I had never managed to define a good safety in three-cushion. The five bottlenecks below are the result of stopping to do so.

Discipline identification is the first bottleneck. Before counting anything, a record must state which discipline it belongs to, which format, which table size, which ball set. Any comparison is meaningful only when both sides are measured in the same unit. A snooker player with ten century breaks in a best-of-19 format cannot be placed beside a three-cushion player averaging 1.8 points per inning in a race to 40.

The second bottleneck is that the units do not convert. A century break measures the ability to accumulate points within one frame, and it depends on cue-ball position after every shot. A three-cushion average measures output per inning, and it depends on the quality of the leave and on three-cushion calculation. A break-and-run in 9-ball measures the ability to close a rack from the break. The most beautiful index in one discipline is a meaningless index in another. When a news board places Ronnie O'Sullivan's record beside Dick Jaspers' and asks who is better, that board is comparing temperature with length.

The third bottleneck is the definition of defence. In snooker, a good safety puts the cue ball where the opponent has no legal shot. In three-cushion, defence works by pushing the cue ball into difficult areas, and no standard index is published. In 9-ball, defence means hiding the cue ball behind another ball, and it is measured by how often the opponent is forced into an error on the next visit. Defence in billiards is four different concepts living inside one word. I once merged them into a single defence-efficiency column in my own log. That column predicted nothing.

The fourth bottleneck is the compression of technical advantage in short formats. A player who wins 60 percent of innings over the long run wins roughly 60 percent of innings. In a race to 40, the standard error of the win rate falls with the square root of the number of innings, so results track ability fairly closely. In a race to 15, there are far fewer innings, the confidence interval widens, and a weaker player still wins often enough. The same applies to pool: a race to 5 carries far more variance than a race to 11. Short formats do not produce a more deserving winner; they only produce more variance.

The fifth bottleneck is reliance on touch. Rolling speed depends on humidity, temperature and cloth speed. In Ho Chi Minh City, afternoon humidity can run higher than in the morning, and the same shot with the same power travels differently. A player with a high success rate in the morning session will not necessarily hold that rate at night. Touch is a variable to be measured, not an explanation to be used. When I cross-checked my log against room conditions, the correlation between form and scoring dropped sharply on rainy days. Most of that drop was not in the player's arm.

From these five bottlenecks I tried to build a simple model for three-cushion: use the six-month scoring average as a baseline, compute an expected score for each match from projected innings, then compare expectation with the actual result. The method copies the logic of expected goals in football. Results only stabilised once the sample passed roughly one hundred innings per player. Below that threshold, the overperformance is mostly noise. A player can beat expectation by 30 percent across a seven-match event and return to the previous level at the next one. That is regression to the mean, not a breakthrough.

A player's journey is not an upward arrow, it is a scatter plot. Each point is one event, and the vertical axis is the gap against expectation. A flat trend line over several months is normal. Anyone reading that chart as a straight line will keep mistaking noise for a turning point.

The most easily missed part is that the absence of a shared data system does not mean importing the entire football toolkit. Football has an advantage in event volume per match, enough for an index such as expected goals to stabilise after a few games. A three-cushion match in a race to 40 produces only a few dozen meaningful innings. There are far fewer data points per match, and the signal-to-noise ratio is far lower. Building an expected-points-per-inning index and publishing it after every short-format event manufactures false precision. The reader gets a clean chart and a wrong conclusion.

There is another explanation for a player's scoring average rising over three months, and it has nothing to do with skill: new cloth, a new ball set, or a weaker opponent group. In my data, two events with the same name held at different venues produced field-wide averages far enough apart to reshuffle most of the rankings. That is correlation driven by conditions. Separating it from ability requires recording playing conditions before recording scores. That takes extra work, and almost nobody does it.

A ranking is, in essence, a regression model, but people keep calling it a race. A ranking is meaningful only when its input variables are controlled. In billiards today, the inputs are not controlled, so the standings move more than ability actually does.

What I take away sits elsewhere: the first task is a shared identification standard. Every record must state the discipline, the format, the table size, the ball set, the room conditions and the unit of measurement. Once that standard exists, advanced metrics have something to stand on. The second task is to log playing conditions as a mandatory part of the match record. The third is transparency about sample-size thresholds before publishing any conclusion about form. One three-cushion season supplies far too few innings to talk about turning points. Anyone who wants to know which player is rising should first ask what unit his numbers are in.

Data limitations: this article draws on a hand-recorded log of roughly two hundred three-cushion and pool matches in Ho Chi Minh City between 2026 and 2026, plus published score sheets from a number of international events. Sample sizes per player are uneven, with some cases below one hundred innings, so any individual comparison here illustrates method rather than ability. Room conditions were recorded in full for only about one third of matches. No standard dataset for the three-cushion discipline was published by any federation during this period. Confidence intervals around the gap-versus-expectation estimates are wide, and I do not assert any trend beyond the scope of the stated data.