How to read a racing record

Almost every service publishes a record. Almost none of them publish one you can actually judge. Here is how to tell the difference, and the same test applied to ours.

Win rate tells you close to nothing

"We hit 34% of our plays" sounds like a number until you ask what it paid. Hitting a third of your bets at even money loses money. Hitting one in six at 8/1 prints. A win rate with no prices attached is a statistic chosen because it was the flattering one, and a service that leads with it is telling you which number it could not lead with.

The question is never how often. It is how often, at what price, against what the race was worth.

Flat-bet return is the honest number

Bet the same amount on every play. Add up what came back. That is flat-bet return, and it is the only figure that cannot be dressed up by bet sizing. The moment a record uses variable stakes, "3 units on this one, half a unit on that one," you are reading a number that depends on decisions made after the fact, and you have no way to audit them.

We publish flat $2 win bets on every flagged play. No parlays, no staking plan, no place or show cushion to soften a losing month.

A hundred bets is a rounding error

This is the one that separates people who have done this from people who have not. Horse racing variance is enormous. A 20% win rate over 100 bets can come from a genuine edge or from nothing at all, and the arithmetic cannot separate them. You need something in the range of a thousand settled bets before a return starts carrying information, and most services quietly restart the count long before they get there.

So when you read any record, ours included, find the sample size first. If it is under a few hundred, the return is a story, not evidence. We say that plainly on our own record page even when the number is good, because it would be dishonest to only mention sample size in the months it excuses us.

Takeout is the opponent

Every dollar in the win pool has roughly 15 to 18 cents taken out before anyone is paid, and exotics run higher. You are not trying to beat the other bettors. You are trying to beat them by more than the house takes off the top, every single race, forever. That is why a break-even record is genuinely respectable and why "we are up slightly" is a claim worth more scrutiny than it usually gets.

We print the takeout on every card so the bar is visible next to the plays.

The price you were quoted is not the price you got

Pari-mutuel odds are final. Whatever the board showed when you bet is irrelevant; you are paid the closing price. So a record that grades itself at morning-line odds, or at the price showing when the pick was posted, is grading a bet nobody could have made.

Ours is graded at the closing price, and we publish the drift between the price we called and the price the race closed at. When our number gets bet down, that shows up as a worse result, which is what should happen.

How to read a flag

Model

Our estimated win chance for the horse.

Fair

The odds that chance implies: the price where a bet breaks even.

Board

The morning line, the track’s early guess at the odds.

Win bet

Bet the horse to win. Pari-mutuel pays the closing price no matter when you bet, so you are backing the horse, not a fixed price.

Hold / bet down

Whether the price should stay long or get bet down by post: how much of the edge is likely to survive. Long prices the crowd leaves alone tend to hold, short ones get bet down. It is a heuristic on a small sample, and every call is graded on the record either way.

The rest of the card

Most races. We still show our top pick and a read on it, but there is no price worth taking.

Odds note: 7/2 is the same as +350.

What to actually check

• Was it published before the race? A pick you cannot timestamp is a memory. Look for a hash, an archive commit, a post with a time on it.
• Are the losses there? Count them. A record showing only winners is not a record.
• Can you recount it yourself? Every bet, every price, every result, downloadable. If you have to trust their arithmetic, you are not reading evidence.
• Does anything get edited after the fact? Corrections should be appended and dated, never quietly patched.
• Do they tell you when the sample is too small? The ones who do are the ones who understand what they are publishing.

Our own answers are on the record and in corrections, and every card ever published is in the archive, hashed before post time.