Mark Selby's 147 and the Data Void of Professional Billiards

On 30 April 2026, at the Crucible Theatre in Sheffield, Mark Selby did something unprecedented in the 46-year history of the World Snooker Championship: he made a maximum 147 break inside a final. Fifteen reds, fifteen blacks, not a single error. In any statistical table, that is the most perfect frame this sport can produce — one hundred per cent pot success, the longest possible scoring sequence, a table state controlled from the opening break to the final ball.

Selby lost that final 15-18 to Luca Brecel.

I sat with that frame for a long time afterwards. What stopped me had nothing to do with feeling sorry for a player. It was a paradox: the most perfect data point billiards can generate sits inside a defeat, and no statistical table I have ever seen explains it. Data never lies, but I have misheard it before — and this time I misheard it in a different way.

Professional billiards does not lack numbers. The World Championship carries a prize fund of roughly 2.4 million pounds, with 500,000 pounds for the champion. World Snooker Tour broadcasts pot success rate, safety success rate and average shot time in seconds. CueTracker archives the result of every frame, every century, every 147 across more than three decades. Ronnie O'Sullivan leads the professional maximum-break list with 15, far clear of the field; he was also the first man past one thousand career centuries, a milestone he reached in March 2026.

Set beside football's data ecosystem — where every pass, every pressing action, every shot is assigned a probability value — billiards looks like a library full of books with almost no index. We know which player pots more balls. We do not know how difficult those pots were.

That is where this piece begins: why does a sport with a cleaner mathematical structure than football hold many times less analytical data than football does?

What billiards measures, and what those metrics conceal

During the 2026/24 season I hand-recorded forty frames involving twelve different players. For each frame I logged three things: the number of shots taken from beyond 1.5 metres, the number of safeties that forced the opponent to hand the table back, and the points scored immediately after an opponent left a favourable table. Based on my experience of tracking matches, this is the only way to verify what the eye sees and the electronic scoreboard does not display.

Take pot success rate, the most quoted figure of all. Suppose Player A takes on ten difficult pots and makes seven. Player B takes on ten easy pots and makes nine. The table concludes that B is more accurate. But if all seven of A's pots opened up scoring positions, while B's nine were merely safe finishes that opened nothing, then the metric has inverted the truth. Pot success rate is a systematically biased metric: it punishes aggressive players and rewards those who hang on. Football solved the equivalent problem with expected goals, weighting each shot by the quality of the chance. Billiards has no publicly published equivalent.

Average shot time sits in the same trap. It is a real number, machine-measured, technically beyond dispute. People use it to label one player slow and another fast. Yet the same man will have a completely different shot time when leading by 60 points than when trailing by 60. A match average blends two opposing psychological states into a single cell, which is then read as a fixed human attribute. I once made exactly this mistake, using an average metric to describe a conditional phenomenon, and it took me a long while to notice.

A sport that could be computed more thoroughly than football

There is something rarely said out loud. In theory, billiards has far greater modelling potential than football. In football, a goal is a terminal event, impossible to derive from the preceding state. In billiards, the score is a state variable: at any moment, the points remaining on the table can be calculated exactly from the number of reds left and the positions of the colours. The points required to overturn a deficit are an arithmetic result, not a guess.

Billiards belongs to the group of head-to-head sports closest to determinism, and simultaneously to the group least measured. Record the coordinates of all 22 balls on every shot and you could reconstruct almost an entire frame as data. No professional tour has published such a dataset.

In other disciplines the gap is wider still. Professional 9-ball has break-and-run percentages and run-out counts, but collection remains manual and inconsistent between tours. Chinese 8-ball, where players such as Zheng Yubo and Gareth Potts compete, has almost no standardised public dataset at all. I once tried to assemble stable numbers for a single season of that circuit and gave up after several weeks.

Why the technology is sufficient but the data is not

Many analysts will tell you billiards lacks data because it lacks technology. I do not believe that. Cameras at the Crucible have had enough resolution to read ball positions for more than a decade. Motion-tracking systems used in other sports have operated reliably under far harsher lighting and speed conditions.

The problem is economic incentive. Betting liquidity in billiards is several orders of magnitude below football, the media market is narrower, and no data company has a financial reason to build a high-grade metrics system and then give it away. Football has expected goals because millions of people are willing to pay to read it every week. Billiards does not have that readership, and I have to admit I am one of the few willing to sit and count by hand.

Even if that dataset appeared, I would still have to state its limits. I once built an expected-goals model for Vietnamese football and had it destroyed by a single goalkeeper, because the metric ignored shot-stopping form. Billiards has its own version of the problem: a safety that produces no visible error can still be the finest shot of a frame, and it evaporates from every success-rate table. Some things sit beyond measurement — arena noise, a tip that needs replacing mid-match, the long silence before a decisive black. Models cannot capture them. And a model that does not know what it fails to capture is a dangerous model.

I do not write to persuade anyone. I write so that the data has a witness.

Sample size, variance and the trap of the aggregate figure

O'Sullivan's figure of 15 maximums is a real fact, but it is also a metric distorted by the number of events he enters. A player who competes in more tournaments has more chances to reach a 147, even with an identical cue action. Comparing two players by raw maximum-break counts is like comparing two strikers by goals without dividing by minutes played.

Sample size in billiards is harder at a deeper level. A professional plays roughly thirty to forty matches a season, but each match contains many frames, and each frame is a small sample with enormous variance. One contact a few millimetres off on the break changes the entire table. That means any conclusion drawn from a single frame is close to statistically meaningless, even when that frame contains a 147.

I learned this the expensive way. Three thousand matches taught me that one match can teach more than all of them — but only when that match is placed back into the longer sequence it belongs to. That is why I still keep my handwritten ledger instead of relying entirely on the aggregate numbers handed to me.

Signals for next season

I do not intend to sit and wait for a complete data system to appear. What I can do next season is specific: extend the ledger to a metric nobody has named — the number of times a player forces an opponent to surrender the table after three consecutive safeties. Provisionally, this correlates with match outcome more strongly than pot success rate. That correlation may be pure coincidence, and I do not yet have enough data to rule the possibility out.

The crowd laughed. The numbers did not. A year from now, I will rewrite this piece and check every figure I wrote down.