AI Competition
Our first choice in each race beside EquinEdge's, on the same races, with the record kept.
Three names on the record
Each race, each one names a single horse as its first choice. We check one thing: did that horse win? The colours stay the same in every chart on this page.
Amber
Horse-Oracle
The horse our model ranks highest in the race. It's a paper model, so nothing is staked.
Green
EquinEdge
A racing-analysis product for handicappers. We use the first choice it publishes for each race and describe it only from its own website.
Outline
Morning-line favourite
The favourite the track's handicapper sets before racing. Our yardstick, not a rival, and not the closing price.
How each race is scored
The same six steps, in the same order, for every race on the record.
Same races
A race counts only if both named a first choice. These are the paired races.
Same question
Did the first choice win? Second and third place aren't scored here.
Same answer key
Winners come from the settled official results, not from our notes.
A yardstick
The morning-line favourite is scored on the same races.
The gap, with its range
The difference in winners per hundred races, with the range chance allows at 95%.
The disagreements
Where they picked different horses, we count whose pick won, and test that count exactly.
Only disagreements decide it
When both name the same horse, the race can't tell them apart. The same goes for races where both are wrong.
So the comparison comes down to the two coloured squares: races where they picked different horses and only one of them won. If those two counts are near even, the record can't separate the products, however many races there are.
When we say close, and when we say leads
The word on the board is set by these rules, not chosen by us after seeing the numbers.
- Under 300 racesNothing published
Too few paired races to compare.
- 300 to 4,000Close
Always "close", even when the range clears zero.
- 4,000 or moreLeads in this block
Only when the range clears zero for that block of races.
- Two blocksConsistent lead
The range clears zero in two separate blocks of races.
Reading the range
The dot is the gap we measured. The bar is how far chance alone could move it. If the bar crosses or touches the level line, the record can't separate the two products, and we say close.
These are shapes only. The real range is drawn from the record at the top of the page.
Head to head, on the same races
The full record
AI Competition
Our first choice in each race beside EquinEdge’s, scored on the same races, with the record kept.
Paper model, $0 staked.
Not advice on any race.
- %winners in races
- %winners in races
- Morning-line favourite%winners. Our yardstick, not a rival.
RacesTracksPaired racesUpdated
The record is being rebuilt from a new receipt.
The figures come back once every number in it checks out. Until then this page shows how the comparison works, and nothing else.
The record needs JavaScript to check its numbers before it shows them.
The race-by-race receipt is linked under “Check it yourself”.
Who found more winners?
When they disagreed, who was right?
Has it stayed close?
By field size
Why it says “close”
The word on the board comes from these rules, set before the results were known. It is not chosen on the day.
How the record is kept
Open any part. Nothing here changes a number above.
How each race is scored
- Same races. A race counts only if both named a first choice before post time. These are the paired races.
- Same question. Did the first choice win? Second and third place aren’t scored here.
- Same answer key. Winners come from the settled official results, not from our notes.
- A yardstick. The morning-line favourite is scored on the same races.
- The gap, with its range. The difference in winners per 100 races, with the range chance allows at 95%.
- The disagreements. Where they picked different horses, we count whose pick won, and test that count exactly.
Why only the disagreements decide it
When both name the same horse, the race can’t tell them apart. Nor can a race where both are wrong. The comparison comes down to the races where they picked different horses and only one of them won.
| Horse-Oracle’s pick | EquinEdge won | EquinEdge lost |
|---|---|---|
| Horse-Oracle won | Both right changes nothing | Only Horse-Oracle right counts for Horse-Oracle |
| Horse-Oracle lost | Only EquinEdge right counts for EquinEdge | Both wrong changes nothing |
What this does not show
- How money would have done. It counts winners only, with no prices and no payouts.
- The closing price. The yardstick is the morning line.
- Picks made at different times. A pick made near post time with live prices is not scored against a pick published earlier.
- One unbroken system. When our system changes during a record, we name the change and show both parts.
- A small gap settled. A few thousand races can show a large gap. They cannot settle a small one.
Who is on the record, and how a product joins
Horse-Oracle is the horse our model ranks highest in the race. It’s a paper model, so nothing is staked. EquinEdge is a racing-analysis product for handicappers. We use the first choice it publishes for each race and describe it only from its own website. The morning-line favourite is set by the track’s handicapper before racing. It is our yardstick, not a rival, and not the closing price.
We add a product only when it publishes dated first choices that we can save before post time and score the same way. If you publish dated picks and want them scored like this, write to us.
If you are named here
If you work for a product named on this page and see an error, write to us. We will check it against the receipt, fix the page where it appears, and record the correction on the corrections page. Corrections are shown. They are not quietly removed.
QuestionsWhy compare with EquinEdge?A comparison means little unless someone else’s picks are scored on the same races with the same rules. Naming a product is not a criticism. We say only what its own website says.What if the record shows Horse-Oracle behind?We publish it as it stands. Windows where we trailed stay on the page. Nothing is removed after the fact.Is this a tip sheet?No. Horse-Oracle is a paper model with nothing staked. It gives no advice on how to spend money on races.Will the numbers change?Yes. The record is rebuilt from a new receipt when new races are added. Earlier versions are kept.
Check it yourself
The receipt lists every race with both first choices and the winner. Every figure on this page can be worked out from it.
Fingerprint (first 12 characters of its SHA-256)
If the file you download hashes to something else, it is not the file this page was built from.
Download the race-by-race receipt
The receipt link returns with the record.
First choiceThe horse a product ranks highest for a race.Paired racesRaces where both products published a first choice.Morning lineThe track’s own pre-race estimate of each horse’s chances. It is not the closing price.Range chance allowsThe 95% interval: how far the gap could move if a similar set of races were drawn again.CloseNo clear difference between the two products in these races.ReceiptThe published race-by-race file behind every number on this page.