Billiards and the Data Void: Why Nobody Can Model a Break Shot

The 2026 World Snooker Championship final ended close to midnight Liverpool time. Zhao Xintong beat Mark Williams 18-12 at the Crucible, becoming the first Chinese world champion, after serving a 20-month suspension over betting-related matters. On my screen was a spreadsheet: 35 frames, four column headers, and not a single row of data. The first column I labelled "cloth speed". The second: "arena humidity". The third: "average duration of a safety exchange". The fourth: "number of forced risks taken from a bad break-off position". Four variables anyone who has spent long enough inside the Crucible can feel through their fingertips, and not one of them exists in any public database. Eight years of watching professional billiards, and I still cannot build a minimum viable model for a single match. Last week I received an assignment that ran to one line: "billiards". No discipline, no tournament, no player. Whoever sent it presumably believed billiards is one thing. It is not. The first reason sits in the name itself. "Billiards" covers at least four sports with entirely different rules, tour structures and audiences: snooker on a 12-foot table with 15 reds; American pool in 9-ball and 10-ball under Matchroom's World Nineball Tour; Chinese 8-ball on Joy tables under the CBSA system; and three-cushion carom, dominant across Europe, Turkey and South Korea. The scoring units are incompatible. Snooker is counted in frames, each frame an almost unbroken run by a single player. Pool is counted in racks, where control of the table shifts constantly and a rack can end in three shots. Chinese 8-ball carries the rack structure of pool alongside the called-pocket requirement of carom. Carom is counted in points within an inning of unlimited duration. The technical consequence is straightforward: a "pot success rate" defined for snooker is meaningless when applied to 10-ball, and vice versa. Football does not face this problem at tour level. The Premier League, La Liga and the World Cup all share a 90-minute unit, an 11-player structure and an event-tracking system run by Opta or StatsBomb under identical protocols. Billiards has no equivalent. Broadcast billiards data is produced by the host broadcaster, not an independent third party. Every tournament, every market, every television cycle can define "pot success" its own way. With no common benchmark, no model can be trained across events. The next problem is the unit of analysis and the leverage of a single shot. A professional football match contains roughly 800 to 1,200 passes, each carrying a small weight in the overall result. A 19-frame snooker match contains perhaps 300 to 400 shots from both players. The average weight of each shot is three times higher, and the distribution of that weight is severely skewed. A missed safety at 4-4 in a best-of-19 can turn the entire match; a misplaced pass in the 34th minute leaves almost no trace in the model. Based on my experience watching qualifying rounds in Sheffield and Leicester across recent seasons, most decisive matches are settled not by century breaks but by long safety exchanges after the scores level. That is precisely the thinnest category of data television provides. The third problem is sample size. Ranking-event qualifiers are usually best-of-7. In that format a player produces only 30 to 40 shots of genuine analytical value, and the sampling error is large enough that any conclusion about form sits inside the noise band. A season holds around 20 ranking events, but most of the data concentrates among a few dozen players who go deep. The rest is a thin sea of numbers, insufficient for a model to separate skill from luck. The fourth problem is prize structure. The World Championship pays 500,000 pounds to the winner, while a first-round loser receives around 20,000 pounds. That is an extraordinarily concentrated distribution, and it is not behaviourally neutral. Players at the top have an incentive to take risks because the marginal reward is so large; players ranked around 60th have an incentive to play safe to protect their tour card and income. Two groups operating under two different utility functions, and any model using one parameter for both is modelling the wrong subject. The fifth problem, and in my view the most important, is the physical environment. Cloth speed, humidity, arena temperature and the age of the balls directly shape how the cue ball rolls. A table in a dry, cold hall in Sheffield responds differently from one in a packed arena in Asia. The historical data we have does not record these variables, which means every cross-tournament comparison is comparing things that are not alike. The sixth problem is the shift toward Asia. The 2026-25 calendar carried a denser run of professional events in China and Hong Kong than any previous period. In parallel, in June 2026 the WPBSA announced sanctions against ten Chinese players over betting and match-fixing, with Liang Wenbo and Li Hang banned for life. That is a governance signal landing in the sport's fastest-growing market. No risk model can price an event like that, because it does not live in any match-data table. Finally there is the paradox of the individual record. Ronnie O'Sullivan holds seven world titles and 15 competitive maximum 147s, the most in history. Those are the two most quoted numbers in the sport, and they do not explain why at 48 he still wins more than he did at 38. What is missing is not the volume of data but the type of it. Error is where reality signs its name, and in billiards, error shows up in places the scoreboard never touches. The instinct of any analyst is to import football's toolkit: probability models, composite indices, heat maps of table zones. I tried. Most of that effort, I now think, points the wrong way. The reason lies in a detail from the empty-stadium period. In 2026, when English football grounds closed, long-pass success fell by around 12 percent, contrary to what most models predicted. The variable removed was not a player or a tactic but crowd noise. The noise we lost is the variable the model forgot. In billiards that noise was never recorded to begin with, so it was never removed, and therefore never recognised as a variable at all. Moreover, the failure to identify "billiards" as a single discipline is not an analyst's error. It is the commercial structure of the sport: four rulesets, four audiences, four data regimes, one shared name. Building a model on the shared name is building on sand. There is an occupational risk I have lived through myself. Once you own a model, you start believing everything that matters can be measured. Absolute faith in a model is exactly as dangerous as absolute faith in a tactic: it makes people stop watching. What the industry calls a Tactical Wizard is really just someone who has kept the habit of looking where nobody has placed a ruler yet. Next time you watch a billiards match, instead of counting centuries, count the safety exchanges after the score reaches 4-4. That is where matches are genuinely decided, and where the data is thinnest. Watch the Asian swing of the season, where the sport's capital and its governance risk now sit on the same table. Every break shot is a hypothesis waiting to be disproved by reality, and most of those hypotheses will be disproved by variables nobody has bothered to measure.