The Horse Hasn't Changed. The Game Has.

From Phil Bull's stopwatch to artificial intelligence: how 80 years of data have transformed what punters can know — and who controls the numbers.

HORSE RACINGTECHNOLOGYGAMBLING

Ed Grimshaw

11/8/20269 min read

In 1948 a former schoolmaster from the West Riding put men with stopwatches beside the winning posts of British racecourses, wrote down what they told him, and began a revolution that the betting ring took years to notice and the sport has never quite finished arguing about.

The idea was almost insultingly simple. A racehorse’s merit could be measured rather than merely admired, and the measuring would be done by arithmetic instead of by the loudest voice in the bar.

The man was Phil Bull, and he was the kind of character racing produces once a generation and then spends half a century quoting: a cigar-smoking socialist who drove a Rolls-Royce, a professional punter with a mathematician’s suspicion of opinion, and a man who called racing a triviality and gave it his life. Before Timeform he had sold his time figures by post as the Temple Time Test, under the name William K. Temple, because a teacher could not put his own name to it.

His most famous touch owed nothing to a stopwatch. Beaten in the 2,000 Guineas after a wide draw, an injured eye and a ride that left him too much to do, Dante looked to Bull like a colt who had been unlucky rather than outclassed, so Bull backed him for the Derby at 10-1. Dante went off the 100-30 favourite in the wartime Derby at Newmarket, won by two lengths, and paid Bull a reported £14,000 or so.

The apostle of the clock landed his biggest touch by looking up from the page, which is worth remembering whenever the figures get clever.

A marriage made in Halifax

What made Timeform work was a partnership that barely outlasted its first year. Dick Whitford, the form man, left in 1949, reportedly after Bull declined to give him a cut of the profits.

The marriage of their methods outlived the quarrel. Bull’s time figures told you how fast a race had really been run; Whitford’s form ratings told you who had beaten whom, and by how far; and the two, coordinated and published from Halifax under the Portway Press imprint, produced the number that became the sport’s unofficial currency of merit.

The weekly Black Book became scripture for a generation of form students, and the symbols beside the figures, the squiggle for unreliable and the small p for improvement to come, were read like illuminated manuscripts. In October 2020 Timeform stopped printing the annuals and the Black Book, and the scripture moved to the screen.

The tyranny of the single number

For all its genius, the Timeform era carried a limitation that nobody resented because nobody could imagine an alternative. You were handed a figure.

It was a superb figure, and the services that followed kept the same shape. The Racing Post, launched in April 1986 with Sheikh Mohammed’s money, put Racing Post Ratings and Topspeed in the morning paper, and the tipping lines did the same by telephone: each arrived as a verdict, and the reader’s part was the congregation’s, to receive, to trust and occasionally to grumble from the back pew.

A single figure compresses class, pace, going and distance into one number, and the whole craft of the form student lay in unpicking what had been compressed. You could not reweight it or ask how it behaved on soft ground at Haydock, because the workings lived in somebody else’s office.

The heretic with a compiler

“In the nineties you were handed a number and asked to trust it. I wanted to give the form student the pencil: choose what goes into the figure, then test it.”

– Simon Walton, founder of Proform

He put the question to himself in 1995, a programmer with a lifelong weakness for the form book and a thought that would have earned him a stern letter from the old guard: what if the reader could build the number?

Proform Racing was established that year and shipped its first software in 1999, built on exactly that heresy: speed ratings averaged against variables the user chose, so that the final figure became a function of the user’s own judgement rather than an edict handed down from on high.

It sounds technical. It was philosophical: the handicapper’s pencil had been passed across the counter and the form student had become an author. Proform Racing Limited is even registered in Halifax, where Bull did his sums.E

Why the data is worth having

Behind every rating sits a harder fact: the price is already a forecast, and a formidable one, assembled daily by thousands of people with money riding on being right. William Benter, who ran a betting syndicate in Hong Kong, wrote in 1994 that the public’s odds are a sophisticated estimate of each horse’s chance, and that any model must rival them to profit.

His paper reports significant positive results over five years of live betting, and the method was humility: he blended his model with the public’s implied probabilities rather than trusting it alone. On that evidence, data earns its keep only where it knows something the price does not, so the test of a rating is not how many winners it finds but whether it finds them at prices that pay.

The prices have habits too. Since Griffith first noted it in 1949, the favourite–longshot bias, outsiders overbet and favourites underbet, has turned up at racetracks around the world with very few exceptions, Hong Kong among them (Snowberg and Wolfers, 2010). A bias like that shows itself only to someone with the records to find it.

How the information changed hands

Then came the exchange. Betfair launched in June 2000, founded by a former professional gambler, Andrew Black, and a JP Morgan trader, Edward Wray, and let punters bet one another rather than the bookmaker. A price could now be recorded, and a recorded price is one you can test your opinion against, which is the professional’s whole trade and the subject of part two.

When a race became a shape

The second revolution was the clock itself, sliced finer. For most of racing’s history the stopwatch gave you one thing, the final time, which is rather like judging a symphony by its running length.

Sectional timing – the furlong-by-furlong record of who went too fast, who was given a breather, who was asked an impossible question at halfway and answered it anyway – turned a race from a number into a shape.

The catch, as anyone who has worked with them knows, is that a closing split means little until you know what preceded it. A fast last three furlongs after a dawdle is a sprint from a near-standing start rather than a fast-run race, and a finishing-speed figure will flatter whichever horse the slow pace suited.

Going compounds it: the description is a clerk’s judgement, the clock a measurement, and when they disagree the clock is the harder evidence.

Britain was slow to take the idea up. TurfTrax covered most all-weather races from 2006 to early 2008 and worked with Channel 4 from 2004 to 2008, but its managing director, Mike Maher, has conceded that the business model was not quite right.

The practical change was cheaper kit. In December 2015 Arena Racing Company began using Total Performance Data’s system, which needs only GPS boxes on the runners, the first of its kind in Europe by the company’s own account, starting at Lingfield.

Sectionals have been freely available on Racing TV, at the courses it covers, since March 2023, which democratised the raw ore. A column of split times is still a puzzle with the picture missing from the box until something sets it against ratings, pace, draw and going.

Elo comes racing

Which brings us to February 2026 and RaceMetrics, the web-native platform Proform built on a quarter of a century of carefully kept data.

The core RaceMetrics Ratings borrow their logic from chess. Arpad Elo, a Hungarian-born physics professor, devised the original for American chess in 1960, and the world federation adopted it in 1970; the racing version updates every morning, adjusting for the quality of the opposition and where a horse finished against expectation, on a scale where 1500 is average and 1600 marks the elite.

The twist is that the Combined Score leaves the horse out altogether and rates the six sets of connections around it – trainers, jockeys, owners, sires, dams and damsires – with the weights chosen on races held back for testing. The horse’s own rating sits beside it as a separate figure: one asks how good the horse is, the other how strong the team behind it is.

On the company’s own figures the top-rated runner on the Combined Score has won 23.04 per cent of nearly 50,000 races since 2023, nearly twice what chance would manage. A runner picked at random in each race wins about 12.5 per cent of the time, a shade above the 10 to 11 per cent that dividing winners by runners suggests, because a small field counts for as much as a big one.

The fairer yardstick is the market favourite, which wins roughly one race in three, but because the Combined Score leaves the horse out, its worth lies in being an independent opinion to set beside the form and the market rather than a selection method on its own.

Unusually for the trade, the method is published, weights and all, and the company describes its figure as a statistical estimator to sit beside the official handicapper’s judgement rather than to replace it. A free account shows one meeting’s racecard a day, ratings included, so anyone curious can watch them at work before spending a penny.

What the old formbook could never tell you

The real departure from Timeform is a set of questions the old books could not pose. Take a horse that finished fourth, beaten three lengths, in a moderate handicap, a run the conventional form line dismisses in four words.

The Future Form Score asks whether the horses it chased home have since gone and won, turning the old pub boast that “the form’s been franked” into a figure from 0 to 100. Collateral form links it through shared rivals to horses it has never met, converting lengths into pounds with a confidence grade attached, which form students once did by candlelight with three back issues of the Black Book.

The sectionals then ask how the fourth place was achieved. Finishing speed is set against a par for how an evenly run race would have finished, so it shows how the race was run as well as who ran it; a Gate Speed Score ranks how smartly the horse leaves the stalls; stride length and cadence show whether it quickens by lengthening or by rattling its feet; and MaxU estimates the pounds it gave away by spending its energy in the wrong furlongs. Fourth, beaten three lengths, can be the best run in the race.

The context is quantified too. Draw and Pace figures test a stall and a running style against five years of course-and-distance history, so the Chester draw stops being folklore and becomes a number. The Trip Predictor refuses to be seduced by one eye-catching run at a new distance, pulling its estimate back towards the career average until the evidence earns more trust, which is shrinkage in the statistician’s sense.

The artificial intelligence that earns its keep

Then comes the part that would have had Bull reaching for a fresh cigar. AI Race Cards analyse every declared runner, and Ask RaceMetrics answers racing questions in plain English, so a query that once took a spreadsheet and an evening takes as long as the kettle.

Each morning the RMP model strikes a fair price for every runner before the market has moved, as a book of exactly 100 per cent with no bookmaker’s margin. The System Builder tests ideas against real Betfair prices on every British and Irish race since 2018 rather than against a starting price carrying the bookmakers’ margin.

The RaceMetrics ratings themselves begin only in 2023, so tests built on them have a shorter history. Entries now carry their ratings from the five-day stage, so Saturday’s big handicap can be worked on from Monday.

The machine, and the hand on it

Nearly eighty years separate Bull’s stopwatches from a platform that rates every horse and six sets of its connections before breakfast, yet the line between them is straighter than it looks. Both rest on a refusal to accept that a horse’s merit is a matter of opinion, and both carry a Halifax postcode, a town better known for wool than for winners.

What has changed is who holds the pencil. Bull’s readers received a verdict; today’s form student can build, test and argue with their own.

The machine is ready. Whether the hand on it knows what to do with it is the business of part two.

Cutting Comment uses data provided by Proform Racing. RaceMetrics provides analysis and data, not betting services. Please gamble responsibly. 18+. BeGambleAware.org