StableBet
Professor Furlong and Pascal at the AI Lab
THE AI LAB
Pascal turns up with a new system every single week. So we test every one of them, properly, against the market. Come and see how his ideas have actually held up.
And one of them is going to land any day now. Any day, I'm telling you.
PASCAL'S PANEL · WEEK 4

Pascal's Panel - Week 4: Four-fold on random horses vs The AI's most-confident picks

← All issues·The standings

This week on the Panel

Welcome to the Panel, where every betting idea gets a proper test. Two strategies in front of the Professor this week, both run over real races with the method laid bare, so you can see for yourself how each one holds up.

First, Pascal, the eternal optimist. He decided his careful doubles were too timid and that the real money was in a big-priced four-fold, so this week he is throwing four picked-at-a-glance runners into one accumulator for the life-changing return: Four-fold on random horses, Four random horses in an accumulator. He is certain he has cracked it. He usually is, and that is exactly why his ideas are worth testing.

Second, a new idea. It comes from Rishi, one of our AI punter personas: the AI-truster, the sort who would back the computer's number one, backed blind. The persona is modelled, not a real reader (more on that just below), but the strategy it suggests is a genuine, testable one: The AI's most-confident picks, follow only the model's most-confident picks, on the theory that a clever computer must know more than a punter squinting at the form.

The Professor tested both the only way that counts: against 26,839 real British races, flat stakes to Starting Price, with the fallers and pulled-up horses counted honestly as losses. Here is how each one actually performs.

Pascal
Pascal
the eternal optimist
−62.1p / £1
This week's pick Four-fold on random horses

“Stick four together and the odds multiply, so a tenner turns into a life-changer. One slip, one big day, and I never have to back singles again. The little stake is nothing and the payout is everything.”

Meet Pascal

Pascal is the Lab's resident optimist and Professor Furlong's foil, certain every single week that he has finally cracked it. He turns up with a fresh angle, usually an accumulator or a way to chase last week's loss back, pitches it with total confidence, and watches the Professor test it properly on real races. He never wins and never learns, and he is the warmest reason to keep coming back.

See the full experiment →
Rishi
the AI-truster
−0.8p / £1
This week's pick The AI's most-confident picks

“If the Professor's machine is this sure about a horse, that has to be where the money is. Wait for the big confident picks, pile in, and let the clever computer do the hard work for me.”

Meet Rishi

Rishi trusts the machine over any human hunch, so he follows the model's most confident picks, sure that a clever computer must know. The honest twist is that our own model, backed blind, loses about as much as the favourite, because naming the likely winner is not the same as beating the price.

Their tell: the computer's number one, backed blind

See the full experiment →

Professor Furlong's verdict

Professor Furlong
On Pascal's pick It loses, and it is the worst slip on this whole page: over the sample it returned about -62.14% to SP, so for every £100 staked you got back roughly £37.86. Here is why. A single random horse already loses about 21.6p in the pound, because every price carries the bookie's margin and a blind pick has no judgement to offset it. Stack four of those legs together and you do not add that leak, you multiply it, four bets at a bit over three-quarters back compounds down to about £37.86 back, and all four horses must win or the whole thing dies. The rare big winner is real, but it is far rarer than the multiplied price makes it feel, which is exactly why the random four-fold is the bookmaker's favourite slip and the fastest way on this page to empty your pocket.
Professor Furlong
On Rishi's pick No. Backing only the picks the model rates 35 percent or better still lost money, about -0.8% to starting price over 361 bets. The reason is plain: the model being confident is not the same as the horse being a good price. By the time the model and the market both agree a horse should win, the bookies have already shortened it to match, so there is no spare value left to take and the overround quietly does its work. We land this close to break-even only because the sample is small and jumps-heavy, which is well within the swings of luck, not because anything has beaten the book. It is not a profit and it is not an edge, and the moment you take real-world prices and pay commission it slides back into a clear loss.

The standings

Both of this week's strategies sit on the leaderboard already. Here is where everything we have tested stands right now, ranked by real-world return, the least-negative at the top. Every system is a net loss, so the top row is the best real performer, not a winner.

24 systems tested · 26,839real races · none beat the bookmaker's margin. The best real return is −0.8pin the pound, so even the strongest system is a net loss. The bookmaker's margin is the reason.

#StrategyWho proposed itReal return (per £1)SampleStrike£200/week → today
1The AI's most-confident picksThe ProfessorGuest: RishiAI persona, modelled−0.8p361——
2Odds-on favourites onlyThe systemsGuest: BarryAI persona, modelled−4.8p4,637wins 60%−£879
3Favourite in a big fieldThe systemsTested system−7.1p612wins 24%—
4Favourite in a small fieldThe systemsTested system−7.6p3,876wins 46%—
5Favourite over jumpsThe systemsTested system−7.9p10,262wins 37%−£2,157
6Favourite on soft groundThe systemsTested system−8.8p10,672wins 35%—
7Back the favouriteThe basicsTested system−8.8p27,892wins 35%−£2,041
8Festival favouritesThe systemsTested system−8.8p2,145wins 32%—
9Favourite in handicapsThe systemsTested system−8.9p18,700wins 30%—
10Favourite on the FlatThe systemsTested system−9.3p17,630wins 34%—
11Follow the AIThe ProfessorTested system−11.4p6,601——
12Back the second favouriteThe basicsGuest: SandraAI persona, modelled−11.9p27,890wins 21%−£2,971
13Top-rated horseThe basicsGuest: FrankAI persona, modelled−15.8p24,654wins 19%−£4,167
14Top-rated in handicapsThe systemsTested system−16.3p18,684wins 16%—
15Lucky 15 on favouritesThe multiplesPascal, the punter−17.6p———
16A random horseThe basicsTested system−21.6p27,892wins 13%−£5,260
17Back the lowest drawThe systemsTested system−24.6p17,611wins 12%—
18Back the old stagersThe systemsTested system−27.0p8,643wins 8%—
19Each-way an outsiderThe basicsTested system−30.0p26,234wins 7%—
20Four-fold on favouritesThe multiplesPascal, the punter−30.8p———
21Back the outsiderThe basicsTested system−34.7p27,892wins 3%−£4,833
22Four-fold on random horsesThe multiplesPascal, the punter−62.1p———

24 systems tested · 26,839 real races · none beat the bookmaker's margin. The best real return is -4p in the pound. The bookmaker's margin, the over-round, the slice baked into every price so the odds add up to more than 100%, sits in all of them, and no selection rule we have tested clears it. That is the genuinely useful part: the leaderboard shows exactly where the edge goes.

See the full Betting Systems Leaderboard →

What we're testing next week

Pascal will be back with another idea he is sure cracks it, and the Professor will test it the same way. And another of the personas has floated another testable angle from one of the personas, and we will run it the same way, against real races, with the method shown and the real result reported.

If you take one thing from the Panel, take this: picking a likely winner is the easy part. Getting paid more than the true risk is the hard part, and the over-round is what stands in the way. That is the insight every row on the leaderboard is really measuring. Bet for fun, with money you can afford to lose, and read our track record for whether anyone, us included, holds a real edge at all.