The Blank Cell in Billiards Data, and the Cost of Filling It With a Story

Last week, at my desk in Liverpool, I reopened my season tracking file and counted 2,184 empty cells across 5,430 rows. The column for cue-ball resting position after a safety shot was blank 62 percent of the time. The column for the gap between shots was blank nearly half the time. I know how many frames I watched this season. I also know that most of what I saw never made it into the file.

What stopped me was something else. Twice I put my hands on the keyboard to fill a blank cell with a story. The first was a safety shot in a qualifier that I believed was better than that player's usual standard. The second was a frame I believed the player lost rhythm in because of noise from the stands. I remember both clearly. I have not one shred of evidence for either.

That is the permanent condition of billiards analysis, and it deserves to be said plainly rather than dressed up in terminology.

A sport recorded meticulously at the result layer and barely at all at the process layer

Billiards, and snooker specifically, has a data paradox. CueTracker, the database maintained by Ron Florax, holds near-complete records of century counts, maximum 147s, head-to-head records and ranking movement for almost every professional player across decades. Want a qualifier from 2026? You have it. Want John Higgins' career century total? You have it.

What you do not have is position. No optical tracking system exists at the level of Opta or StatsBomb in football. Nobody publishes shot-level cue-ball position data in raw form for public use. The atomic unit of this sport is one shot, and one shot contains at least four variables: cue-ball position before the strike, power, spin, and the player's choice among available options. No open dataset records all four.

The consequence is that television footage becomes the only detailed source. That creates the crudest form of selection bias: only frames that reached broadcast exist. Qualifiers in Sheffield or Burton happen in front of a handful of people, and nobody counts. A model trained on that data learns from players who won enough to be televised, then uses that knowledge to predict people who have never appeared on television.

The official ranking system has the same problem in a different shape. A player's ranking is a two-year rolling average. It is stable, easy to look up, and months behind actual form. A rolling average cannot show that someone just changed how they approach the table.

In Britain, where I work, readers follow every frame. They care about the fight to keep a professional tour card, decided by a place in the top 64 after a two-year cycle. That fight happens in exactly the qualifiers nobody counts. Based on my experience tracking matches, most ranking movement comes from frames with no shot-level data at all, which is why the rankings always surprise the people watching.

Four mechanisms that turn a blank cell into a dangerous variable

The most common fix for a blank cell is to fill it with the mean. In snooker, that erases precisely the data that decides outcomes. At professional level, most shots are patterns: familiar lines, familiar stances, familiar pace. What decides a frame is a small group of unusual shots: the safety under pressure, the escape when the cue ball is frozen to a cushion, the shot when only one line exists. That is the tail of the distribution. Filling the tail with the mean amounts to claiming that decisive moments look like ordinary ones. No player wins a frame with an average shot.

The second problem is the unit of measurement. Football settled this long ago: the unit is the possession, and every metric reduces to it. Snooker never did. A visit to the table can be one shot or forty. The century rate, the most quoted figure in the sport, depends on table conditions, the quality of the opponent's safety play, and whether the player had to play a morning qualifier that day. Comparing century rates between two players at two different events means comparing things measured with two different rulers and calling them by the same name.

Tactics are not on the scoreboard; they live in the gap between two shots. That gap is what nobody records. Everyone records the final shot of a frame. Nobody records the eleventh shot, the one where the player decided not to attack.

The third problem is the variable models forget: noise. In July 2026, the World Championship at the Crucible admitted a pilot crowd on its first day, then closed the doors entirely after health guidance changed. The final between Ronnie O'Sullivan and Kyren Wilson finished 18-8. The record preserved the score, the centuries, the duration. The record did not preserve the sound of a chair dragged across a wooden floor, or the fact that a player could hear his own breathing between shots. The crowdless season deleted a variable no model can encode: noise. That same year, when English football returned, I spent six months rewriting a research paper after finding that long-pass completion fell 12 percent against expectation. Snooker has no equivalent, simply because snooker has no metric that could fall.

The fourth problem is the one I want to dwell on. Error is where reality signs its name. A player who enters the shot four-tenths of a second later than his own daily rhythm in the eighteenth frame has not made a technical mistake. He is signing his name to a state. Official data records the missed pot and calls it a missed pot. The four-tenths of a second, the number with no column in my file, is what tells the story.

And there is a fifth problem, rarely raised at data conferences. A complete dataset is not the same as a correct one. In June 2026, the WPBSA announced sanctions against ten Chinese players, with Liang Wenbo and Li Hang banned for life, following a match-fixing investigation. Look at the record: no blank cells. Results there, scores there, timings there, head-to-heads there. Every column filled. The only thing missing was the reason a player missed a shot he had never missed before. When the match ends, the number lies more elegantly than the player.

The blind spot is human, not optical

The industry's default response to all of this is more data: more cameras, more sensors, more metrics. I do not believe that is the bottleneck.

The blind spot sits with the writer. The worst kind of analysis is not the kind that says "we do not know". The worst kind is confident: a tidy argument built on three blank cells, wrapped in enough jargon that nobody bothers to check. I have written pieces like that. The reader was not wrong to believe them; the fault lay with the writer who would not leave the blank cell alone.

The same mechanism governs how the sport evaluates young players. A nineteen-year-old who wins two televised qualifiers gets a ranking, a highlight reel, a narrative. The model sees those two matches. The model does not see the two hundred hours of practice, the coach behind him, or the fact that he just changed his cue. Conclusions drawn from two matches are presented as though drawn from a career. A player's ranking is just a story the market repeats until it believes it.

Some things are revealed more clearly by time than by any metric. O'Sullivan, Higgins and Williams, all born in 2026, are still competing at the top level past fifty. A model trained on 2026 data would have retired all three fifteen years ago. The model was not wrong. Its input was missing a variable nobody had named.

I still keep the file with those 2,184 blank cells, and I have no intention of filling them from memory. A blank cell marked honestly is a statement about your own limits. A blank cell filled with a story is a lie with a nice interface.

The test for next season is concrete: watch whether anyone publishes shot-level data for qualifying rounds, alongside televised frames. If that happens, ranking models will shift in a measurable direction within eighteen months. If it does not, we will keep reading confident stories built on exactly those 2,184 blank cells.