Horse-Oracle

Paper model · $0 staked · Not betting advice

Racehorses running in front of a grandstand

How it works

What we know before a race, what we check after it, and how we keep the two apart.

Before and after, kept apart

Our read of a race is saved before it runs. The check comes after. Nothing learned from the result goes back into the read.

Before the cutoff

What was known

  1. The card, as it stoodTrack, date, race number, field, morning line and scratches, each with its source.
  2. Our read, savedThe horses we chose, their chances, the race context, and what the evidence can't tell us.

After the race

What happened

  1. The official resultFrom the official result or an accepted source, matched to the same race we saved.
  2. The race autopsyDid our top choice win? If not, why not, and what can we fairly learn?

The review never pretends the result was known beforehand, never replaces the saved read, and never treats one race as proof.

Six labels on every study

Every pre-race study carries the same six labels, so you can see where it came from and what it can't show.

Pre-race studySix labels
Source
Where each fact or number came from.
As of
The latest time the information was available.
Model
Which model or rule produced the read.
Comparison
What it's measured against, like the morning-line favourite.
Outcome
The result used for the later check.
Limits
What the evidence doesn't prove.

The comparison is always named exactly. The morning line is the track's own estimate, so we never call it the market. If tote odds are used, the study says which tote and at what time.

When our top choice loses

Every miss is sorted into one of five kinds. Naming the kind keeps us from explaining a loss away.

Did our top choice win?If it did, the autopsy says so and records who else finished where. If it didn't, the miss goes in one box.
  • OrderingThe winner was in our read, but ranked lower.
  • Race shapeThe pace or the trip didn't go the way the read expected.
  • Field constructionWho was entered, and how they matched up, decided it.
  • A scratchA horse came out and the race changed.
  • Missing informationSomething that mattered wasn't in what we had.

The lesson. We write down only what the evidence supports. If it supports nothing, we say that.

Every line of an autopsy carries a tag

  • VerifiedConfirmed by the official record, with the source named.
  • Video-observedSeen on the replay, with the time in the clip.
  • ClaimedSomeone else's statement, credited to them, not checked by us.
  • InferredOur reading, shown with what it rests on.
  • UnknownWhat the evidence can't show. We don't guess.
  • RetractedSomething we got wrong, left visible and struck through, with the correction beside it.

What the numbers mean

A chance is an estimate, not a promise

Say our read gives a horse a 30% chance. Over 100 races like that, about 30 of those horses should win. One race can't tell you whether 30% was right. Checking that across many races is called calibration.

WonDidn't winFor example, 100 races at 30%

A fair comparison is like for like

Two sides can only be compared when they're scored the same way.

  • Our readSame racesThe yardstick
  • Our readSame cutoffThe yardstick
  • Our readSame answer keyThe yardstick

When we confirm a result, we use later races that weren't used to build the model, and keep them apart.

What this doesn't prove

  • That a horse will winA chance is never a promise.
  • That a past result will repeatRaces change. The record shows what happened, not what will.
  • That any model stays ahead for goodIncluding ours. The record keeps running.