The Thursday Take is our weekly column: one racing talking point, put to the test against the Proform database — thirty years of British racing, every price, every result. Including, as this week, when the popular story does not survive contact with the numbers.
The short version
- British turf racing ran 1,438 consecutive races without a single one on soft ground or worse — from Ayr on 20 June 2026 to Carlisle on 19 August, a run of sixty days. It is the longest such run since 2010.
- Between 1 July and 18 August 2026, not one British turf race was run on soft or heavy going. In thirty years, no other summer has managed that — the nearest is 2006, which still had thirteen. The measured going allowance, +1.91 seconds per furlong, is a thirty-year record.
- It was not, however, the fastest ground on record. 70.8% of races came on good-to-firm or quicker, which is second to 2006's 74.4%. The remarkable thing about 2026 is not how firm the peak was — it is that the rain never came at all.
- "Firm ground favours front-runners" fails. Horses that led won 21.3% on fast ground against 18.7% on soft, but around 40% of that gap is fast-ground fields being smaller — and front-runners returned an A/E of 1.399 on fast against 1.377 on soft, with the highest figure of all on good. The ground does not change what a front-runner is worth.
- "Fast ground means fewer upsets" is almost entirely field size too. The share of winners returning 20/1 or bigger looks 0.92 points lower on fast ground; hold field size constant and the gap shrinks to 0.18 points, which is nothing.
- The one thing that genuinely pays is the one thing you cannot have. Knowing in advance which horse will lead is worth about 30% over the market (A/E 1.295 across 901,164 runs). The best pre-race prediction of it is right 50.4% of the time, finds horses that win at 20.6% against 11.5% for the rest — and returns 1.006. Par, in every going band.
- Build it: System Builder — Race Type = Turf, Official Going = Good To Firm + Firm, Pace Rating 8 or more, Pace Rating % of Race Total 30% or more. Then switch the going to Soft and Heavy and watch the answer refuse to move.
This week's talking point: the ground
British racing has spent August talking about water. The Racing Post reported racecourses being spared usage restrictions while roughly three-quarters of England sat in drought, Newmarket managing its watered gallop around reservoir levels, and trainers losing grass gallops entirely. Clerks have been irrigating flat out since June.
That is the input. We are interested in the output — the going, which is what a punter actually bets into. And the output was remarkable, though not quite in the way the headlines suggest.
What actually happened
Here is every British turf race run between 1 July and 18 August, in each of the last thirty years. The same calendar window every time, so nothing is being flattered by a part-season.
| Year | Races | % good-to-firm or faster | Soft or worse | Avg going allowance |
|---|---|---|---|---|
| 1997 | 1,053 | 60.3% | 56 | +0.53 |
| 1998 | 1,111 | 61.1% | 24 | +0.98 |
| 1999 | 1,128 | 65.6% | 50 | +1.39 |
| 2000 | 1,214 | 62.6% | 21 | +1.25 |
| 2001 | 1,240 | 54.2% | 40 | +1.03 |
| 2002 | 1,203 | 45.8% | 111 | +0.31 |
| 2003 | 1,236 | 68.3% | 22 | +1.02 |
| 2004 | 1,250 | 49.8% | 64 | +0.44 |
| 2005 | 1,242 | 57.7% | 54 | +1.11 |
| 2006 | 1,266 | 74.4% | 13 | +1.64 |
| 2007 | 1,146 | 35.5% | 238 | +0.35 |
| 2008 | 1,305 | 36.9% | 139 | +0.30 |
| 2009 | 1,362 | 35.6% | 107 | +0.63 |
| 2010 | 1,416 | 56.8% | 39 | +0.93 |
| 2011 | 1,419 | 40.0% | 73 | +0.46 |
| 2012 | 1,281 | 31.1% | 312 | −0.26 |
| 2013 | 1,457 | 60.9% | 70 | +1.73 |
| 2014 | 1,405 | 52.8% | 89 | +1.02 |
| 2015 | 1,381 | 38.2% | 102 | +0.52 |
| 2016 | 1,431 | 47.6% | 123 | +0.95 |
| 2017 | 1,476 | 20.7% | 236 | +0.02 |
| 2018 | 1,470 | 55.3% | 35 | +1.13 |
| 2019 | 1,442 | 41.3% | 170 | +0.69 |
| 2020 | 1,242 | 32.3% | 116 | +1.36 |
| 2021 | 1,361 | 31.7% | 141 | +0.63 |
| 2022 | 1,287 | 53.1% | 30 | +1.32 |
| 2023 | 1,264 | 16.1% | 270 | −0.17 |
| 2024 | 1,204 | 38.7% | 81 | +1.15 |
| 2025 | 1,277 | 35.8% | 58 | +0.98 |
| 2026 | 1,189 | 70.8% | 0 | +1.91 |
Three things in that table, and it is worth being precise about which is which.
Not one race on soft or heavy. In thirty years, no other summer has managed that. The nearest is 2006, and 2006 still had thirteen. After that it is 2000 with 21, 2003 with 22, 1998 with 24. Half the years in the series are in three figures; 2012 had 312 and 2023 had 270.
The clock agrees, and sets a record. Going allowance is measured from race times rather than judged by eye, and 2026's +1.91 seconds per furlong is the highest in the thirty years — ahead of 2013 (+1.73) and 2006 (+1.64). Two independent measures, the same answer.
But this was not the fastest ground on record, and we should say so plainly. 70.8% of races came on good-to-firm or quicker, and 2006 beat that with 74.4%. Anyone who remembers the summer of 2006 remembers ground that rode quicker than this one at its peak, and they are right.
So the story is not that 2026 was the firmest summer. It is that 2026 was the summer without rain events. 2006 had a faster peak, but it still got wet thirteen times. 2026 did not get wet once. That is the thing that has never happened before in our data, and it is a different claim from a record for firmness — a more interesting one, and one that survives contact with anybody's memory.
The run itself: the last British turf race on soft ground before the dry spell was at Ayr on 20 June. The next was at Carlisle on 19 August. In between, Britain staged 1,438 turf races — 986 of them on good-to-firm or faster, eight on good-to-soft, and not one on soft or heavy. That is the longest gap between soft-ground races since 2010, when a 77-day run covered 2,326 races. (A longer gap in 2020 was the racing suspension, not the weather, and only 258 races fall inside it.)
Two terms, in plain English.
A/E — add up what the betting market thought each horse's chance was, then compare it with how many actually won. 1.00 means the market had it exactly right. Below 1.00 means you were paying over the odds; above means the group won more often than its prices implied. It is a fairness score, not a profit figure, and one huge-priced winner cannot distort it the way it distorts ROI.
Going allowance — the clock's opinion, in seconds per furlong, of how much quicker or slower the ground was running than standard. The going description is a human judgement made by eye; the allowance is measured afterwards from the times.
One honest qualification before we go further, because it cuts against the easy story. This was not a dry year from end to end. June 2026 was actually softer than a normal June — 89 of its 850 British turf races were run on soft or heavy, 10.5%, against a long-run average of 6.8%. The dryness arrived late, and then it arrived completely.
Three things everybody knows about fast ground
Ground like that is supposed to change racing. Ask around and you will get the same three answers: it favours front-runners, it produces fewer upsets, and it makes favourites more reliable. All three are checkable claims about a knowable population, which is exactly what this column exists for.
So we tested them on every British Flat turf race since 2008, up to the day the run ended — 658,265 individual runs, split into three going bands: fast (good-to-firm or quicker), good, and soft (good-to-soft or slower).
Why 2008 and not 1997? The going table above needs only the going, which we hold back to 1997. These three tests need a price for every runner, and Betfair Starting Prices cover our data in full from 2008. So the climate question gets thirty years and the betting question gets eighteen — which is the right way round, and we would rather show you the seam than quietly split the difference. Across those eighteen years the whole cohort returns between 0.996 and 1.000 in every single year, so par really is par throughout.
Test 1 — "Firm ground favours front-runners"
On the face of it, yes. Horses that made the running won 21.3% of the time on fast ground against 18.7% on soft — a gap of 2.58 points, which sounds like a lot.
But fast-ground fields are smaller. The average British Flat turf field in this data is 10.2 runners on fast ground and 10.7 on soft, and a front-runner in a small field has fewer horses to hold off. Standardise both bands to the same field-size mix and roughly 40% of the gap goes.
Then ask the only question that matters — were you paid for it?
| Going band | Horses that led | Strike rate | Same-field-size strike rate | A/E |
|---|---|---|---|---|
| Fast (good-to-firm or quicker) | 28,906 | 21.32% | 20.66% | 1.399 |
| Good | 23,260 | 19.14% | 19.85% | 1.401 |
| Soft (good-to-soft or slower) | 23,252 | 18.74% | 19.09% | 1.377 |
Front-runners are underbet everywhere — that 1.40 is a real and very large number, and we will come back to it, because it is the most misunderstood figure in racing. But it is the same size in every going band. Fast 1.399, soft 1.377, and the highest of the three is good. The spread across the three bands is smaller than the margin of error on any one of them.
The ground does not change how much front-runners are worth. It changes how many rivals they have to hold off.
Test 2 — "Fast ground means fewer upsets"
Also field size, and almost all of it.
| Measure (British Flat turf) | Fast | Soft | Raw gap | Gap at the same field size |
|---|---|---|---|---|
| Favourite strike rate | 35.22% | 32.26% | +2.96pp | +1.82pp |
| Average winning Betfair SP | 7.81 | 8.45 | −0.64 | −0.10 |
| Winners returned 20/1 or bigger | 6.95% | 7.87% | −0.92pp | −0.18pp |
Raw, fast ground looks like it produces meaningfully fewer shock results — 6.95% of winners at 20/1 or bigger, against 7.87% on soft. Compare small fields with small fields and large with large, and four-fifths of that gap evaporates, leaving 0.18 of a percentage point. On 49,000-odd winners, that is not something you could ever bet into.
The average winning price tells the same story even more starkly: an apparent gap of 0.64 points of odds shrinks to 0.10 once field size is held constant. The compression everyone attributes to the ground is very largely field size wearing a going costume.
Test 3 — "Favourites are more reliable on fast ground"
This one is true, and it is the only claim of the three that does not die outright.
Look only at 8-to-14 runner Flat fields, where the average field size is near-identical across all three bands, and favourites win 32.07% on fast ground against 29.47% on soft. That 2.6-point gap is real and it survives the field-size control.
Their A/E on fast ground is 1.036, against 0.998 on good and 0.996 on soft. Now, we are going to be straight about this one rather than waving it away: on a naive count that 1.036 sits about two standard errors above par, which is not nothing. Two things temper it. The runs are not independent of one another, so the true margin is wider than a simple count suggests. And 3.6% gross, before you have paid any commission, is not a business — it is a rounding error that happens to point upwards.
So: real, small, and about as close to fully priced as makes no difference. Which is a more honest verdict than the one we would have given you on a shorter run of data, where the same figure looked like comfortable noise.
Three claims, and not one of them is worth money. The front-runner effect is partly field size and identical in every going band. The upset effect is four-fifths field size and the remainder is a rounding error. The favourite effect is genuine, tiny, and gone by the time you have paid commission. The most extreme ground in thirty years of British racing changed how the sport looked and next to nothing about how it paid.
The number everybody quotes, and nobody can use
Go back to that 1.40 for a moment, because it is where fast-ground analysis usually ends up. Front-runners return an A/E around 1.30 — pace pieces quote it constantly, and it is real. Across all British turf racing since 2008, horses that made the running returned 1.295 against their Betfair prices over 901,164 runs. A 30% edge over the market. Enormous.
There is one problem. You cannot know before a race which horse will lead. That 1.295 is calculated from how horses actually ran, which is information that arrives at the same time as the result.
So we re-ran it using only what is knowable in advance: Proform's pre-race Pace Rating, applying our own Bet Finder's definition of a likely leader — a Pace Rating of 8 or more, and at least 30% of the race's combined pace ratings. Same races, same prices, same window. The only difference is that this version could actually have been bet.
| Going band | A/E if you KNEW who would lead | Predicted leaders | Their strike rate | How often they led | A/E on the prediction |
|---|---|---|---|---|---|
| Fast | 1.350 | 8,590 | 20.33% | 49.7% | 1.001 |
| Good | 1.281 | 13,389 | 20.43% | 50.3% | 1.004 |
| Soft | 1.277 | 16,526 | 20.78% | 50.8% | 1.010 |
| All going | 1.295 | 38,505 | 20.56% | 50.4% | 1.006 |
Three readings, and all three are worth having.
- The prediction is genuinely good at picking horses. Runners flagged as likely leaders win at 20.56% against 11.52% for everyone else — nearly double the strike rate, from a filter you can apply the night before.
- It is right about half the time. 50.4% of predicted leaders actually made the running. That is the best available answer to the question, and it is a coin toss.
- It is worth almost precisely nothing. A/E 1.006 across 38,505 predictions, and 1.001 / 1.004 / 1.010 across the three going bands. The market has already paid for the half you can predict.
The 1.30 that pace content quotes is the price of a crystal ball, not an edge. And note where the ground sits in that table: nowhere. 1.350 / 1.281 / 1.277 looking backwards, 1.001 / 1.004 / 1.010 looking forwards. Flat both ways.
This is not an argument against pace analysis. A filter that doubles a strike rate is extremely useful for building a shortlist, and the horses it finds do win. It is an argument against one specific claim — that front-runners are underbet in a way you can collect on.
The bonus: the drought stories had the withdrawals backwards
A recurring line all summer was that trainers pull horses out when the ground goes firm. It is intuitive, it is repeated by people who know far more about horses than we ever will — and in the data it is the wrong way round.
| Going band (British Flat turf, 2014–2026) | Races | Declared | Ran | % withdrawn |
|---|---|---|---|---|
| Fast | 17,355 | 163,638 | 148,923 | 8.99% |
| Good | 14,724 | 148,037 | 136,781 | 7.60% |
| Soft | 14,760 | 152,915 | 136,463 | 10.76% |
Trainers withdraw more on soft ground than on firm, by nearly two percentage points. (Declared-runner counts start in 2014 in our data, so this table covers thirteen years rather than thirty.)
And the summer of 2026 makes the point sharply: British Flat turf races between 1 July and 18 August lost 5.97% of their declared runners, the lowest figure in the thirteen years we can measure. Second lowest? 2022, at 6.74% — the other drought summer in the series. The two driest summers we have on record are also the two in which fewest horses were taken out.
Fields were smaller this summer — an average of 8.15 declared and 7.66 that ran, both lows for the period. But that happened at the declaration stage, not on the morning of the race. The horses were not being pulled out. They were never put in. Which is a different problem, and one for a different week.
One warning before you go hunting
The obvious next move after a summer like this is to look for sires whose runners go particularly well on fast ground. We tested it properly, and it is a null result — a solid one.
Take every sire with 300 or more fast-ground runners in Britain since 2008 — there are 226 of them — and measure how far each one's record sits from what its runners' prices implied. The spread across those 226 sires is 0.984 times what pure chance alone would produce, where 1.000 is pure chance. Their average is 0.995. Sires' fast-ground records are scattered every bit as randomly as shuffling the runners would scatter them, and if anything marginally less.
Nor does it repeat. Take the sires that stood out on fast ground over the first half of the period, follow them into the second, and they came back at 1.017 on fast ground — against 1.020 for the very same sires on every other kind of going. Whatever survived is not a fast-ground effect at all, and 1.7% gross is a losing bet once commission is paid.
If you meet a fast-ground sire angle built on four hundred runs, it is almost certainly noise wearing a name.
And how the summer itself paid
One last check. If firm ground really did tilt the sport, the summer of 2026 is where the tilt should show up in the results. It does not. Across British turf racing between 1 July and 18 August: favourites returned an A/E of 1.003 — dead par — and the horses that made the running returned 1.183, which is lower than the front-runner premium's long-run figure, not higher.
The most rainless summer in thirty years of data produced completely unremarkable betting outcomes.
Build it yourself
Everything above is reproducible in the System Builder in about two minutes, and this is a version worth saving, because it also flags qualifying runners on today's and tomorrow's cards.
Recipe: the likely leader on fast ground
| Filter | Setting |
|---|---|
| Race Type | Turf |
| Official Going | Good To Firm, Firm |
| Pace Rating | 8 or more |
| Pace Rating: % of Race Total | 30% or more |
Run it, note the strike rate and the A/E, then change Official Going to Soft and Heavy and run it again. The strike rate will move a little. The A/E will not move at all. That is this whole article in two clicks, on your own screen, with your own eyes.
If you would rather reproduce the going table above instead, build the simplest system there is — Race Type = Turf, Course Country = GB, no other rules — and put Going on the breakdown. Every row of our first table is in there.
One note on what we did not use. Going Adjustment — the measured clock figure in our first table — is a post-race number. It is superb for a back-test and useless for tomorrow's card, and the System Builder says so plainly rather than quietly dropping the rule. Pace Rating and Pace % are both genuinely pre-race and are both computed live for your qualifiers, which is why the recipe above is built on those two.
Not a Proform Premium member? The System Builder answers questions like this one in about ninety seconds, and runs your saved systems against every declared runner each morning. See what's included.
FAQ
Was the summer of 2026 the firmest on record in British racing?
No, and the distinction matters. 70.8% of British turf races between 1 July and 18 August 2026 were run on good-to-firm ground or faster, which is second to 2006's 74.4% over the last thirty years. What was unprecedented is the complete absence of soft ground: not one British turf race in that window was run on soft or heavy, where 2006 still had thirteen and no other summer in thirty years had fewer than 21. The measured going allowance, +1.91 seconds per furlong, is a thirty-year record.
How long did British racing go without soft ground in 2026?
Sixty days and 1,438 turf races, from Ayr on 20 June to Carlisle on 19 August. That is the longest gap between soft-ground races since 2010, when 77 days and 2,326 races separated them. A longer 2020 gap was the racing suspension rather than the weather.
Do front-runners really do better on fast ground?
They win more often, but not because of the ground. Across 658,265 British Flat turf runs since 2008, horses that led won 21.3% on fast ground against 18.7% on soft — but around 40% of that gap is explained by fast-ground fields being smaller (10.2 runners against 10.7). What remains is fully priced: front-runners returned an A/E of 1.399 on fast ground, 1.401 on good and 1.377 on soft. The spread across the bands is smaller than the error on any one of them.
Does fast ground produce fewer big-priced winners?
Barely. It looks that way — 6.95% of winners returned 20/1 or bigger on fast ground against 7.87% on soft — but four-fifths of that is a field-size effect. Compare like-for-like field sizes and the gap falls to 0.18 of a percentage point. The same correction shrinks the apparent gap in average winning price from 0.64 points of odds to 0.10.
Why is the front-runner A/E of 1.30 not a betting edge?
Because it is measured from how horses actually ran, which you only learn after the race. Using the best pre-race prediction available — Proform's Pace Rating at 8 or more, holding 30% or more of the race's combined pace ratings — those horses win at 20.56% against 11.52% for the rest, but they lead only 50.4% of the time and return an A/E of 1.006 across 38,505 predictions. The market has already priced the predictable half.
Do trainers withdraw more horses when the ground is firm?
The data says the opposite. In British Flat turf racing since 2014, 10.76% of declared runners were withdrawn on soft ground against 8.99% on fast. The summer of 2026 lost just 5.97% of its declared runners, the lowest in the thirteen years we can measure, and the second lowest was the other drought summer, 2022, at 6.74%.
What is A/E in horse racing?
A/E — actual over expected — compares how many horses in a group actually won with how many the betting market implied would win. 1.00 means the market priced the group perfectly; below 1.00 means you were paying over the odds; above 1.00 means the group beat its prices. Unlike ROI it is not distorted by one or two very big-priced winners, which makes it the right measure for questions like this one.
The Thursday Take runs every Thursday morning, with a special whenever the big meetings come around. All figures from the Proform database, verified 26 August 2026. Going table: British turf racing, resulted, 1 July – 18 August in each year 1997–2026. Betting tests: British Flat turf racing, 1 January 2008 to 18 August 2026, 658,265 runs holding a Betfair Starting Price; pace figures cover all British turf racing over the same window, 901,164 runs. Withdrawal figures begin in 2014, when declared-runner counts start in our data. A/E calculated at Betfair Starting Price across all starters, non-finishers included.
