Why Our Model Liked Mount Atlas at 33/1, and What 657 Long-Shot Picks Say

Mount Atlas won the British Champions Long Distance Cup at Ascot at 33/1, and our racing model had picked him out on the morning card. He was the pick of Stablebet Edge, the model's value-hunting entrant in the Silicon Tipster League, because no runner in the race showed a bigger gap between the model's view and the market's: the model gave him 7.6%, the morning market 3.2%, a gap of 4.4 points.

That gap came from three things on his card, all of them there for anyone to read.

His rating sat in the middle of the field. Mount Atlas went into the race on an official rating of 112. The field ran from 106 to 117 and averaged 112.8, so he was only 0.8lb off the average. The market priced him as one of the outsiders. On the ratings he was close to the pack.

He had won at Ascot before. In our records since October 2023 he had run at the course 3 times and won 2 of them. Course form is one of the inputs the model weighs, and no other runner in the race had more course wins.

His yard was in form. Andrew Balding's string had sent out 9 winners from 70 runners in the fortnight before the race, and 26 winners from 212 runners at Ascot since October 2023.

His last run counted against him in the market's eyes: third of 8 at York on 21 August, beaten 3¼ lengths. The market let him drift from 25/1 in the morning to 33/1 at the off. Ridden by PJ McDonald, he beat Queenstown by three-quarters of a length, with the 7/4 favourite Highwayman back in third (race review).

One winner is still only one winner. The more useful question is what happens across every horse the model likes this way, and that is the rest of this post.

A flatter view of the field than the market's

Set the model's numbers for the Long Distance Cup beside the market's and a pattern shows. The model's chances for the 10 runners ran from 4.0% to 18.9%. The morning market's, with the bookmakers' margin taken out, ran from 1.2% to 20.8%. The model rated the three horses at the head of the market lower than the market did, and the outsiders higher.

That is how the model reads most races. It works from ratings, form, course and distance records and the stable's recent strike rate, and on those measures the runners in a race are often closer together than the betting suggests. The market knows things the model cannot see: how a horse has been working at home, a change of plan, the weight of money from people who follow the yard. Those pull the prices further apart.

So when Stablebet Edge goes looking for the biggest gap on a card, the gap is usually on an outsider. Across its long-shot picks in the test below, the model's average chance was 10.3% against the market's 2.6%.

There is a second reason to be careful with those gaps. Choosing the single biggest disagreement in every race picks out the runners where the model is most likely to be wrong as well as the ones where it might be right. Every estimate carries some error, and the largest gap on a card is the likeliest place for that error to be pointing upwards. A big gap is a reason to look harder at a horse, and on its own it proves nothing about the horse's chance.

Mount Atlas was one of those gaps that came good. The test below asks how often they do.

What 657 long-shot picks say

Stablebet Edge has bet to its current rule since 15 September 2026. Up to 18 October 2026 it had 1,262 settled bets, each a level £1 win bet at the starting price, with non-runners and void races set aside. We took the ones that look most like Mount Atlas: the market gave the horse less than 5%, and the model at least twice that. There were 657 of them.

Add up the model's chances for those horses and it expected 67.5 winners; the morning market's chances, the same figures quoted in each pick's reason (the morning price as a chance, with the margin taken out), came to 17.1. 28 of them won, a strike rate of 4.3%.

The model expected about 2.4 times as many winners as arrived, which is its flat view of the field showing up in the results. The winners did beat the morning market's count, and yet the bets' profit at SP was -£136.09 on £1 stakes, a return of -20.7%.

That looks like a contradiction until you check which price paid out. The League settles every bet at the starting price, and the starting price is a different number from the morning one. At SP, the same horses add up to 30.9 expected winners with the bookmakers' margin left in, or 25.4 with it taken out. The figure with the margin left in is roughly how many winners the bets needed just to get the stakes back, and 28 fell short of it.

The winners explain most of the gap. Of the 28 winners, 20 started at a shorter price than their morning quote, some of them much shorter, so they paid far less than the morning odds suggested. Mount Atlas, who drifted, was one of only 8 that went off bigger. Across all the long-shot picks, 295 shortened and 320 drifted, so the market did not move towards the model as a rule. The horses that were backed during the day simply won more often.

So the market at the off had these horses about right, the model had them too high, and the margin did the rest. The gap between those two prices is the subject of morning price vs SP, and how big prices fare across every runner, not only the model's picks, is covered in is there any point backing a big price.

The shorter-priced group looks better, and proves little

One slice of the record reads differently. Where the market gave Stablebet Edge's pick a chance of 10% or more, there were 236 bets and 73 winners, a strike rate of 30.9%. The model expected 64.4 winners and the starting prices, margin left in, 66.8. The return at SP was +3.1%.

That is interesting, and it is a long way from a finding. It is 236 bets over five weeks, and the result moves a long way with a handful of winners either side. The picks in between, where the market gave between 5% and 10%, returned -36.8% over 331 bets. Across every bet the Edge has placed since it joined the League, its full record is -15.4% over 3,228 settled bets, and that is the figure every slice has to sit beside.

We will keep watching the shorter-priced picks as the sample grows. If the pattern holds over a much bigger record it will be worth a closer look; if it fades, that will be on the board too. Every bet is published win or lose on the Stablebet Edge page, and the model's full results sit on the track record. How the rest of the League got on at Ascot is in the Champions Day tipster league review.

The study's numbers come from a frozen record built from the League's own day-files, settled at the starting price, and will not change. The full-record figures for the Edge are live and move with every race.

Mount Atlas and the model's long shots: FAQ

Why did our model like Mount Atlas?

On the morning card the model rated him at 7.6% against the market's 3.2%, the biggest gap in the race. His official rating of 112 was close to the field average, he had won 2 of his 3 runs at Ascot in our records, and his trainer's horses were in form.

Does the model find value in long shots?

On this record, no. Across 657 picks where the market gave less than 5% and the model at least twice that, the model expected 67.5 winners and 28 arrived. At the starting price the bets returned -20.7%.

Why did more of those picks win than the market expected, yet still lose money?

The market figure in each pick's reason uses the morning price, and the bets are settled at the starting price. Most of the winners shortened before the off, so they paid less than the morning odds implied. At SP the same horses were expected to produce 30.9 winners with the margin left in, which is about what the bets needed to get the stakes back, and 28 fell short of that.

Should I back the model's shorter-priced picks?

No, and nothing here is a tip. That group returned +3.1% over 236 bets, far too small a sample to call, and the Edge's full record is -15.4%. We publish it as something to keep watching.

Where can I see every pick?

Every Stablebet Edge bet, winners and losers, is on its League page, logged before the off and settled at the starting price.

Bet responsibly. Nothing on this page is betting advice. If you bet, stake only what you can afford to lose, and if it stops being fun, stop. Help is at BeGambleAware.org. 18+.