Guide — How to read HorsePal PRO

The numbers on every race

When you open a race on HorsePal PRO, every runner has three numbers next to their name: a rank, a win probability, and — for the top pick — an edge. Here's what each of them means and how to read them.

Win probability

Every horse in a race gets a win probability from our model, shown as a percentage. Because exactly one horse wins each race, all the probabilities in a single race add up to 100%. This is what "normalised" means — we're not giving each horse an independent score, we're distributing our confidence across the full field.

If the top pick is at 35%, the model is saying that horse's chance is about 35 in every 100 races that look like this one. The other 65 are split across everything else in the field.

Rank

The runners are listed from most-likely to least-likely according to the model. Rank 1 is the top pick. Rank doesn't care about market odds or bookmaker favourites — it's purely the model's view, based on the horse's form, the field, the going, and around 80 other factors.

Edge

Edge is the top pick's probability minus what a random guess would give you. In a six-runner race, a random guess would be one in six, or about 16.7%. If our top pick is at 32%, that's an edge of +15.3 percentage points over random.

Edge is the number that decides whether a pick becomes a Best Prediction. It accounts for field size in a way that raw win probability doesn't — a 28% pick in a 6-runner race is a stronger signal than a 35% pick in a 12-runner race, because the smaller field means every horse starts with a higher random-guess baseline.

Reading the shape of a race

The headline top pick tells you part of the story. The full ranked list tells you the rest. Two races with the same top-pick percentage can be completely different propositions once you look at what's behind the favourite. Here are the four shapes worth recognising.

Pattern A — Clear favourite

One horse sits well ahead of everything else. The model has concentrated its confidence on a single runner. Races like this are most common in smaller-field conditions races, novice events, and maidens where one horse's form genuinely stands out. This is the shape where a top pick is doing exactly what you'd want it to do.

Pattern B — Two-horse race

Two horses clearly stand above the rest — say 30% and 27% each while nothing else is above 12%. Neither horse is a standout on its own, but together they account for well over half the model's confidence in the race. Historically, races that look like this see the winner come from one of those two far more often than from the pack behind them. The model is telling you this looks like a head-to-head, and what you do with that is up to you — but it's a different kind of race from one with a single dominant favourite.

Pattern C — Contested field

Three to five runners sit within a few percentage points of each other at the top of the list. There's no clear standout — the model genuinely can't separate the front of the field. These races are harder to predict and historically have lower top-pick hit rates than races with a clear favourite. Be selective: a Solid classification on a Pattern C race is a weaker signal than a Solid on a Pattern A race, even though the tier is the same.

Pattern D — Wide open

The top pick is below 18%, and nothing has any real separation from anything else. This is almost always a big-field handicap — the model's inputs don't give it enough to work with when 14+ runners all have plausible claims. Information value is low, and races like this rarely qualify as Best Predictions for that reason. There's no shame in skipping them.

Tiers and the honesty behind them

Every Best Prediction carries a tier — Banker, Strong, or Solid. These are based on edge, not raw confidence, and each tier has a historical hit rate that you can use to set your expectations.

The honest way to show you what the tiers mean is to show you the data. We took every race our model has been tested on — 37,913 races in total — and measured how often the top pick actually won, broken down by tier and by the probability band the pick fell into. These numbers aren't projections. They're what actually happened.

Read this as a calibration of your expectations. A Banker at 50%+ has won about two out of three times historically. A Solid at 20–25% has won about one in four. The tier tells you the strength of the signal, and the rate tells you what strength of signal has historically delivered.

Why calibration matters

A good prediction model is one where the stated probability roughly matches what actually happens. When our model says 25%, it wins about 24% of the time. When it says 40%, it wins about 50% of the time. That's not a coincidence — that's what calibration means. Here's the full picture:

The closer these two columns are to each other, the more you can trust the probability shown on any individual race. In our case, they're within a point or two across every band — which is about as calibrated as a racing model gets.

Day risk — when to be selective

Every morning we look at the day's card — how many races, how many are handicaps, how big the fields are, how strong the top picks look — and we score it against our historical data. The result is the day risk label you see on the homepage each morning: Risky, Thin, Mixed, or Good.

The percentage underneath each label is the model's expected top-pick hit rate for that day, based on the composition of the card. A Good day is one where the card is full of the race types and field sizes where the model historically performs best — smaller-field novice and maiden races, clear favourites. A Risky day is one where the card is dominated by big-field handicaps, the hardest races in the sport to call.

Think of day risk as your strategy setting for the morning. On a Good day there are plenty of Best Predictions to choose from and the model is playing to its strengths. On a Risky day the qualifying picks are fewer and further between — so narrow your focus to the strongest signals and let the marginal ones pass.

Everything the model factors in

HorsePal PRO is built on a dataset of over 1.7 million historical race records going back a decade. Every morning, every runner on the card is evaluated against over 80 features drawn from that dataset. Here's what goes into the prediction you see.

Every morning, the model runs the full day's racecard through every feature for every runner, cross-references them against more than a decade of historical data, and produces a calibrated probability for each horse. That's a scale of number-crunching no human could match in time for first race — and it's what you're paying for. HorsePal does the heavy statistical work so you can focus on backing the picks you like.

Track record and what to expect

Our model has been tested across 37,556 races from 2024 onwards. Here's how it's performed.

In almost two out of every three races, our top pick finishes first, second, or third. And the top pick wins outright 30% of the time — roughly three times what a random guess would give you in the average UK field. Across nearly 38,000 races, that's a consistent edge that very few models in this space can claim.

Where the model is at its strongest

Every model has conditions it excels in, and HorsePal is no exception. The sweet spot is small-field novice and maiden races, where the top pick hits over 40% of the time:

Small fields and races where class tells — novices, maidens, and conditions races — are where the model is at its sharpest. Big-field handicaps are naturally the most competitive races in the sport for everyone involved, and the day risk label will flag them for you so you can plan accordingly.

Use the tiers, check the shape of each race, and let the day risk guide your strategy. The subscribers who get the most from HorsePal are the ones who treat it as a lens — a way to see the day's racing through the data first, then apply their own judgement on top.