The Weight Next to the Name

Hear it: the podcast episode “One Scale for Every Horse” tells this story.

In a handicap race, the racing secretary studies the field and assigns each horse a weight to carry, deliberately loading the best horses with the most — the idea being that if the weights are right, every horse crosses the wire together. The weight is the opinion, expressed in pounds. Our card borrows that language. For every starter we compute the weight it would carry in a handicap leveled by our own ratings, and we print that weight next to its name.

How we compute it

The raw material is the horse's own past performances — the same lines printed in the program. For each of its past races we take the speed figure it actually earned. Our first version of this rating measured that figure against a par — a standard for the class of the race — and rated the horse on the difference. When we rebuilt the rating in October 2026 we put that par term on trial in the fit, and it earned exactly nothing once the raw figure itself was in the model: the difference-from-par carried no information the figure did not already carry. So the par subtraction is gone. The figure itself is the evidence.

Since October 29, 2026, that figure is our own. We compute it from the running lines themselves — every track, distance and surface carries its own constant, and each line is projected the way the horse itself runs — instead of taking a published figure as an input. The speed number printed on our sheets comes from the same figure, so the weight beside a horse's name and the SPD on its line are two readings of one number.

A horse's figures are read against the company it keeps. Each is centred on the average figure of its class segment — claiming horses against the claiming average, stakes horses against the stakes average — because the same figure earned in rich company and the same figure earned cheaply are not the same afternoon. The races are then combined by a rule we call a corroborated best: the horse's median race — its typical afternoon — sets the base, and its single best race is allowed to lift that base, but only a quarter of the way toward itself. One brilliant afternoon does not make a rating; a brilliant afternoon the rest of the record points toward can move one.

Recent races count more than old ones: a past race's pull halves about every two years, and a horse coming back from a layoff of more than 90 days is docked half a point for each further month away. The combined figure is then pulled back toward its class average — no horse is taken entirely at its own word. The rules for that pull were set when the rating was rebuilt in late October 2026, and the current version of the rating, v4.1 — in production since October 31, 2026 — keeps them exactly. The standard pull was measured from the spread of our own data rather than assumed as a round number. A horse with only one to three past races behind it gets a gentler pull than that standard: its few races are allowed to move its weight further, because when we tested the sterner pull on lightly raced horses it dragged them too close to the average and threw away what their own races were saying. But when a horse's recent races disagree with each other — when the combined figure sits more than five points from the middle value of its last three races — the pull is doubled, because a record that argues with itself has earned less trust. And whatever the pull, a horse with only a race or two behind it cannot be carried further from its class average than the most extreme figures of horses with long records actually reach. What the current version changed is not the rating at all — it is the ruler the rating is read on: how many pounds each point of it is worth. Finally the rating is turned into a weight on a fixed anchor:

So a horse at 126 lb has been running a little over one rating point better than the average horse, and a horse at 114 lb a little over one point worse. The version before this one counted each point at a pound and a half, which squeezed a whole field's weights into a band only a few pounds wide — real differences between horses all but vanished, and a scale nobody can read is no use as a measuring stick. At about five and a half pounds to the point, the weights run correspondingly wider: on the November 10, 2026 card they ran from 68 lb to 137 lb. The scale is entirely ours — it is built from the figures in the past performances, and no published rating (ours is not a Beyer, a Timeform, or a Longines number, and it will not match them) is an input to it.

Every starter on the card carries a weight — no blanks. A horse with only one past race to measure is still held on a short leash: the cap above keeps its weight inside the range that horses with long records have actually earned, so a single afternoon, good or bad, moves it only so far. And a first-time starter, with no past races at all, is shown at the average itself — 120 lb — marked with a †. That marked weight is a placeholder, not a measurement: it says only that the horse has no starts to rate, and we never let it stand in for a real rating anywhere the weights are used.

Reading a field by its weights

• Top weight is our highest-rated horse in the race — on its figures measured against the company it has been keeping, which is not the same thing as the fastest horse or the classiest name.
• The spread is the story. If the field runs from 126 lb down to 124 lb, our numbers see a near-level race. If it runs from 127 lb down to 108 lb, one part of the field has been performing far above the standard of the other part's races.
• A pound or two is a tie. Two horses within a pound of each other are, on this scale, effectively level — which is exactly how the weight is allowed to be used in our selections, below.

What it is not

It is not the weight the horse carries today. Today's actual impost is set by the conditions of the race and printed in the program. Ours is a hypothetical handicap weight, a measuring stick — the two numbers will sometimes be close and often will not.

It is not a pick, and it does not rank our selections. Our picks are made by a published protocol anchored on the morning line. The weight's role inside that protocol is deliberately small and defined: it settles contests that the rest of the analysis leaves effectively tied, and a clear weight advantage (four pounds or more) counts as one named signal among several when a challenger argues against an incumbent pick. It never orders a field by itself. We hold it to that limited role because we keep testing the larger one. In the October 2026 rebuild, on 548 held-out races, the new rating's ordering beat our old rating head-to-head — where the two disagreed on the top horse, the new rating's pick won 46 times to the old one's 32 — but it still finished well behind the market's own prices. A measurement can be honest and useful without being a better oracle than the market — and we would rather publish the test than oversell the tool.

It is not borrowed. Other scales exist — officials and publications rate horses their own way, on their own anchors. A 124 on our scale and a 124 on another are different statements, because the average horse underneath each anchor is different. Weights are only ever comparable inside the one scale that produced them.

The same idea, across 33 years

The weight on the daily card is built from past-performance lines. We also maintain a second, chart-based scale built the other way — from official race charts and beaten lengths, solved as one connected web of shared opponents. Its proving ground was the Breeders' Cup Classic: all 34 runnings from 1989 to 2022, plus the 2021 Whitney, solved as a single web of 323 horses. Inside that web — where the 120 lb anchor is an average Classic starter, a far richer horse than the daily web's average — Cigar rates top of the multi-time winners at 126 lb, Flightline carries 123 lb, and Sunday Silence 122 lb, about a third of a length between the two of them across 33 years. The full story, including how the 1989 chart was recovered after the official archive proved to start in 1991, is told in the podcast episode "One Scale for Every Horse."

That is the whole idea, both scales of it: one number, in the sport's oldest unit, computed from evidence we hold and can show — printed next to the name, with its limits stated in the same place.