How to Use the Racing Terminal
Tekbred runs two independent scoring systems on every race. Understanding the difference — and when to use each — is the core skill.
System Overview
Tekbred is a data-driven recommendation system. It does not tell you who will win — it identifies which horses have a statistically higher probability of finishing in the top positions, based on machine learning models trained on historical US race data.
Three base models analyze each horse independently: QUANT, VECTOR, and ALPHA. Their outputs are combined into a single blended model, and the terminal shows you one number per horse: its chance of winning the race.
Those percentages are calibrated, which is the important part. Across 30,491 races in the past year, horses shown at 60%+ won 68% of the time and horses shown at 20–28% won 23% of the time. When the terminal says 32%, it means 32%.
Race Confidence — the number that matters
Every race card carries a THIS RACE badge. It is the model's probability that the winner is among its top 3 ranked horses, for that specific race.
This race
87%
The model expects to cover this race's winner 87% of the time. Specific to the field in front of you.
Track · 1yr
63.8%
How the model has done at this track over the past year, across every field size. Background, not a pick rule.
Read THIS RACE. That is the whole rule.
You do not need to compare the two numbers. Two races both showing 77% perform identically, whether one sits above its track average and the other below it. Races that beat their track average do hit more often — but only because they tend to be small fields, and THIS RACE is already high for exactly that reason. Comparing the two adds nothing.
Field size drives everything. A 4-horse field is far easier to call than a 10-horse field, so THIS RACE moves with the number of runners — shown next to the badge. A low percentage means a hard race, not a weak read.
Interactive Demo
Same race, two modes. Toggle score mode and TOP depth — notice how rankings, HR, and model highlights all update together.
| # | Horse | WIN % |
|---|---|---|
6 | ①Munnings Express | 34.2% |
2 | ②Paula's A Star | 27.8% |
5 | ③Geez Eloise | 21.5% |
3 | Intentious | 16.5% |
What to notice
The percentages add to 100% across the runners. When the top two sit close together the race is genuinely open; when one horse is far clear, that is the strongest signal the system produces.
HR + TOP depth
Try switching TOP from 3 → 1. Both numbers drop, because the bar is stricter — the winner must be the exact top-ranked horse, not merely in the top three.
The Three Base Models
These three models run independently on every horse entry, then blend into the single win percentage the terminal shows. They are kept deliberately different from one another — one is a neural network, two are tree ensembles — because models that disagree in useful ways blend better than models that agree. The weakest of the three standalone still earns its place for exactly that reason.
Specializes in speed and class patterns — raw speed figures, historical finishing times, and class level transitions.
Most influential at tracks where pace and class are the deciding factor.
Specializes in conditions analysis — surface, distance, and how each horse’s profile aligns with today’s specific race conditions.
Strongest when surface or distance is a key differentiator in the field.
Specializes in recent form and trainer intent — last race cycles, workout patterns, and stable signals that indicate a horse is being pointed for a top finish.
Often the first to flag an improving horse before the public recognizes it.
Reading WIN %
The WIN % column is each horse's probability of winning. Down a card the percentages add to 100% across the runners still in the race — so it reads the way odds do, but without the bookmaker's margin baked in.
The circled numbers ①②③ mark the top 3 — the same depth the TOP buttons control. In the example the first three are close together (32.6 / 31.1 / 23.4), which is the model saying this is a genuinely open race rather than a standout.
SCR — scratched
A horse withdrawn after entries were published drops to the foot of the card with no percentage. When that happens the race is re-run through the model, because removing a runner changes every other horse's chances — the remaining percentages are recalculated, not just rescaled.
Track Record
The second badge on each card shows how the model has performed at that track over the past year, at whichever depth the TOP 3 button is set to. It is measured on real results across every race at that track, not on a sample.
Tracks with too few races in the window show N/A rather than a percentage. Below 150 races the number carries a swing of several points either way, which is not worth showing.
Why tracks differ — it is mostly field size
Saratoga is not scored worse because the model struggles there — it runs the biggest fields on the circuit, and a nine-horse race is simply harder than a six-horse one. Across every track in the past year, the top three cover the winner 91% of the time in four-horse fields and 60% of the time in ten-horse fields.
Use THIS RACE to choose, the track number for context
A small field at Saratoga can read 87% while the track sits at 63.8%, and it will perform like 87%. The track number tells you what kind of racing to expect there — it does not tell you which races to trust.
Analysis Guide
High THIS RACE (roughly 75%+)
- Usually a smaller field — the model has a clear read
- The top three cover the winner about three times in four
- Worth your attention first
Low THIS RACE (roughly 55% or below)
- Usually a large or evenly matched field
- A hard race, not a weak model — treat it as genuinely open
- Check the runner count shown beside the badge