:

How the TrackLab Score works

A plain English explanation of what the number means, where it comes from, and why you can trust it.

Want the full technical detail? It's in collapsible sections marked ⚙️ — open them if you're curious, skip them if you're not.

📋 What is the TrackLab Score?

Every horse in every race gets a number from 0 to 100. That number is our best estimate of the horse's chance of winning, based on data — not gut feeling, not tipster opinion.

A score of 15 means roughly a 15% chance of winning. A score of 40 means about a 40% chance. The numbers are honest — when we say 20%, horses in that range really do win about 20% of the time.

📝 Example
2
River King
Kempton Park · 2:30 PM · Handicap Chase
70
TrackLab Score
TrackLab estimates about a 70% chance of winning — a very strong pick.
💡 The key thing
The score is a probability, not a ranking. A score of 70 means "this horse probably wins." A score of 8 means "probably not, but stranger things have happened." Both are useful information.

Scores are generated fresh every morning at 7 AM using the latest data available — form, jockey stats, trainer records, race conditions, and bookmaker odds. They update throughout the day as odds move.

New to horse racing? Read our beginner's guide →

💎 Value signals

TrackLab compares its score against the bookmaker odds to find horses where the bookmaker may be underestimating the chance of winning. This comparison produces three labels:

Super Value
The bookmaker is significantly underpricing this horse
TrackLab thinks the chance of winning is much higher than the odds suggest. These are the strongest signals — about 9 per day across all races.
44%
Win rate
Value
The odds look a bit generous
TrackLab's estimate is moderately higher than the market price. These are solid picks — about 45 per day.
24%
Win rate
Neutral
TrackLab agrees with the bookmaker
The odds roughly match TrackLab's estimate — or TrackLab rates the horse lower. No particular edge here.
⚙️ Backtested returns (technical)

On our 3-month out-of-sample test set (Oct–Dec 2025), betting £1 on every Super Value selection at best available bookmaker odds produced a +30% return on investment. Value selections returned +23%.

These figures use actual best-available bookmaker odds (accounting for overround), not theoretical fair prices. The value thresholds are automatically calibrated during model training to optimise real-world returns.

+30%
Super Value ROI
+23%
Value ROI
3 months
Test period

Important: Past performance does not guarantee future results. These are backtested figures, not live trading results.

🧪 What goes into the score

The model looks at 67 different pieces of information about each horse, grouped into eight categories:

🐎 The Horse
Age, sex, career record, win rate, where they've raced before, how much weight they're carrying
⭐ Ability Ratings
Official handicap rating and how it compares to the other horses in this race
📅 Recent Form
Last few finishing positions, wins recently, whether they're improving or declining, how long since they last raced
🏇 Jockey & Trainer
Recent win rates, track-specific records, how well this combination has performed together
🏟 Course & Distance
Past results at this specific course, at this distance, on this type of ground — with enough data to be meaningful
🌧 Race Conditions
Ground conditions, race class, distance, surface type, number of runners, prize money
📈 Class Movement
Whether the horse is stepping up or dropping down in class from its last race
👓 Equipment
Blinkers, cheekpieces, tongue strap, visor — equipment changes that can affect performance
It starts with the odds
The model uses bookmaker odds as its starting point — the collective wisdom of the market. Then it learns where the market gets things wrong. This means TrackLab agrees with the bookies most of the time. It's only flagging value when the data says there's a genuine edge.

🔄 When the score updates

Scores aren't static — they refresh through the day as new information arrives:

7:00 AM
Morning scores published
Full scoring pass using stable morning odds. This is TrackLab's considered assessment — the one we're most confident in.
9 AM – 5 PM
Refreshed every 5 minutes
As odds move, scores adjust. Only horses whose scores have actually changed get updated — most stay stable.
Race day
Non-runners & going changes
If a horse is withdrawn or the ground conditions change, affected races are re-scored immediately.

🏷 Best available odds

Every score includes a hint showing which bookmaker is currently offering the best price for that horse. We check across 29 bookmakers to find the most generous odds available.

The value signal (Super Value, Value, Neutral) is calculated against the average market price. But the best-odds hint tells you where to actually place a bet for the highest possible return.

Not sure how odds work? See our plain English explanation →

⚙️ Under the hood

This section is for the technically curious. You don't need any of this to use TrackLab — but we believe in transparency, so here it is.

⚙️ The model

TrackLab uses a gradient-boosted decision tree model (LightGBM) — the industry standard for structured data prediction. It was trained on over 500,000 historical race results spanning 2021–2025.

The raw model output is calibrated using isotonic regression, which ensures the scores are accurate probabilities, not just rankings. When we say 20%, horses in that range win about 20% of the time.

0.79
AUC score
0.008
Calibration error
500k+
Training results
67
Features

An AUC of 0.79 means the model has strong ability to distinguish winners from losers. A calibration error of 0.008 means the probabilities are near-perfectly accurate.

⚙️ The "residual" approach

The model uses market odds as a starting point and learns where the market is wrong. This "residual" approach means TrackLab starts from the collective wisdom of the betting market — millions of pounds of informed opinion — then adjusts based on features the market may underweight or overlook.

This is why TrackLab agrees with the bookmakers most of the time. The value signals are the exceptions — the cases where the data says the market has it wrong.

⚙️ Data freshness & reliability

The model relies on data from an upstream pipeline that runs daily. If jockey or trainer statistics are missing (the pipeline hasn't completed), scoring is held until data is ready — we'd rather delay than publish a score based on incomplete information.

Missing odds alone don't block scoring — the model falls back to a uniform prior (equal chance for all runners) and marks these scores accordingly. You'll see a note when this happens.