● How it works — open book
TBredIQ is a field-ordering engine: it reads the past performances, ranks the field, and hands you a clear read — a Score, a confidence level, and a short list of the horses that matter. This page explains how that read is built, what every number means, and — plainly — what the engine does and doesn’t do. Every figure here is measured against picking blindly, never against the market.
01What the engine is
TBredIQ reads the same past-performance data a sharp handicapper studies — speed, pace, form, class, connections, pedigree — and ranks the horses by how likely each is to finish well in today’s specific field. The output is three things you can read in seconds: the engine’s top-rated horse with a confidence level, a three-horse shortlist, and a grade for the race as a whole.
What it is not: a market-beating oracle. It doesn’t predict exact payoffs, and it doesn’t know what the pool will look like at post time. What it does is order a field better than chance — consistently, across thousands of races — with documented limits we spell out below.
02How the engine reads a race
No single number decides a race. The engine weighs these signals against each other and against the rest of the field — the same way a careful handicapper builds an opinion.
How fast each horse has actually run, adjusted for the track and the day. The backbone of the read — but never the whole story.
Who wants the early lead, whether one horse clears cleanly or several grind into each other, and how that sets up the closers behind them.
Whether a horse is rounding into top form or coming off a peak it may not repeat. Recent, legible form counts more than an old line.
The quality of company a horse has faced, and whether today’s level is a step up, a drop, or a fair spot for it.
Rolling win-rate patterns for the barn and the rider — a supporting signal when both are running hot, never a stand-alone reason.
Sire and damside cues for surface, distance, and first-time situations — most useful where the form lines are thin, like debut runners.
Signals computed from daily past-performance data, morning-line odds, pedigree, and rolling 30-day trainer/jockey patterns. See “Data sources” below.
03What you actually see
Three pieces of the read show up on every race. Here is exactly what each one means — and what it doesn’t.
Our clearest read of the day. Pace is defined, form is legible, and the top pick stands apart from the field. Worth your full attention.
A genuine opinion with a reason behind it — less separation than a FOCUS, but the engine has a real read.
No clean edge here — too tight, too thin, or too unpredictable. The reason is always stated. A PASS is one of the most honest things we can offer.
04Is our confidence honest?
That is the whole promise of a confidence level — and we can show it holds. The engine’s confidence tiers map cleanly, step by step, to how often the top pick actually wins, across nearly 5,000 races.
In the sharpest spots of all: about 4 in 10. That’s our highest-confidence tier — about 3× what picking blindly returns in a typical field. The gradient is real; the more confident we are, the more often the pick delivers.
The yardstick we hold ourselves to. To check that our confidence is honest, we measure the gap between the confidence we claim and the win rate we actually post — a standard measure called Expected Calibration Error (ECE). Ours is 0.045; the common threshold for “well-calibrated” is below 0.05, so we land just inside it.
One wrinkle we don’t hide. At the very top of the scale the engine is a touch over-confident — when it rates a horse extremely highly, the real win rate runs roughly 10 points below that. It doesn’t flip the signal: the top tier genuinely outperforms the lower ones. But read the very top band as “high confidence,” not as a precise number. We’re working on a recalibration and will update when it ships.
13% · 24% · 33% top-pick win rate by confidence tier; sharpest spots ~40%, vs ~13% picking blindly across all races. ECE = 0.045 (threshold 0.05); documented tail over-confidence ~10 points at the high end. N=4,933 settled races.
05What it does — and what it doesn’t
What it does. It orders the field better than chance, race after race. Our top pick finishes in the money — first, second, or third — in about 62% of races, against 38% for a horse picked blindly. The actual winner sits inside our three-horse shortlist nearly 2 of 3 times. Our top pick wins outright about 29% of the time, against 13% picking blindly. Those gaps are real and consistent.
What it doesn’t do. It does not out-predict the closing market. The morning-line favorite wins about 33% of the time — ahead of our roughly 29% on the same set of races. The closing market aggregates everything the public knows by post time; our engine tracks that benchmark closely and does not uniformly beat it on raw win prediction. We say so plainly because it matters.
What we tested and didn’t find. We back-tested straight bets and exotic structures across thousands of races, filtered by field size, confidence tier, odds band, and surface. After correcting an early base-unit error that had made one configuration look promising, the result was consistent: these approaches returned less than was wagered — roughly the track’s takeout. Our shortlist hits exotic combinations more often than chance, but takeout plus the cost of covering combinations erases that edge. “Hits more often than chance” and “returns more than it costs” are separate statements; they did not reconcile in our data. That finding is the line between an honest analysis tool and a sales pitch — and we’re the former.
Sources: scorecard in-the-money 61.7% vs 37.8% random; winner-in-shortlist 63.7% vs 39.9%; top-pick win 29.2% vs 12.7%; chalk comparison ML-favorite 33.2% vs engine 28.4% on the runners-available subset; back-tested straight-bet and exotic returns below the wagered amount across every tier tested. ops/scorecard_output.txt (2026-06-27), N=4,933.
06The one structural edge
There is one honest exception, and it’s structural — not a prediction. When a jackpot pool (a Pick 5, Pick 6, or Super Hi-5) goes unsolved, the unawarded money rolls forward into the next card. That carried-over money is dead money: dollars already in the pool that a winning ticket collects without the new buyer having put them there. A large carryover changes the pool arithmetic in the player’s favor.
Our carryover tracker surfaces these spots, shows the pool size, and flags the dead-money component. The arithmetic is yours to evaluate — we show it; we never tell you what to play or how much. See today’s carryovers →
A vocabulary note. A Pick 5 or Pick 6 is a pari-mutuel multi-race wager — one pool, split among winning tickets after the track’s takeout. It is not a parlay. A parlay is a sportsbook product with multiplying fixed odds; the mechanics are different. We use the right words because getting them wrong changes how you think about value.
07What we publish — and what we deliberately don’t
A PASS is not a failure. Most handicapping services never advise inaction — every race gets a “selection.” We think that’s dishonest. Racing has bad spots: priced-up chalk with no value, unreadable first-out fields, exotic races where the pool structure removes any analytical edge. When the engine can’t separate the field, it says PASS — out loud, with the reason and a note on what would flip it. Knowing which race to skip is one of the most valuable reads we give you.
08Data sources
The engine’s performance figures are forward-log metrics — measured on settled races as they resolved, not in-sample training scores. All statistics on this page trace to ops/scorecard_output.txt (generated 2026-06-27), N=4,933 settled North American thoroughbred races, April–June 2026.
The honest summary
Our top pick lands in the money in about 62% of races, and the winner sits in our three-horse shortlist nearly 2 of 3 times. Our most confident reads win above 40%. The engine does not out-predict the morning line on raw win prediction, and back-tested betting strategies returned less than was wagered — about the track’s takeout. The one structural edge in our data is the dead-money carryover dynamic, and that’s a pool mechanic, not a model output. These numbers describe field-reading quality — not a payday, not a market-beating claim, and not advice on what to wager or how much.