Fair A/E compares how often a horse, trainer or jockey actually won with how often a margin-free reading of the betting market implied they should have. Above 1.00 means more wins than the market expected; below 1.00 means fewer.

Why does KickTheBookies strip the bookmaker margin out of A/E?

Because a starting-price book sums to more than certainty, standard A/E divides wins by an inflated expectation and reads low for everybody. Measured across the KickTheBookies archive the effect runs 13 to 15 per cent cold, uniformly enough that nothing about the figure looks wrong.

How is fair probability calculated?

Each runner's implied probability is one divided by its decimal starting price. That figure is then divided by the overround of its own race — the sum of every priced runner's implied probability — so the field's probabilities sum to 1.00.

When does KickTheBookies refuse to calculate Fair A/E?

When fewer than 2 runners carry a usable price, when under 90 per cent of the field is priced, or when the book sums below 1.00 or above 3.00. Those races contribute nothing to any Fair A/E figure, which is why patchily-priced racing shows no Fair A/E at all.

How large a sample does KickTheBookies require before trusting a figure?

Any sample below 30 runs is flagged on the page as a small sample rather than hidden, and no comparative claim — best course, strongest angle — is made on one.

What period does the KickTheBookies archive cover?

Individual race records run from 2 January 2005 onwards, and aggregate figures for trainers, jockeys, courses, sires and owners run from 1 January 2008 so that every horse inside the window has a complete captured career behind it.

Methodology

Fair A/E

A/E asks whether results beat the market rather than merely beating the field. Almost every published version of it divides by an expectation that still has the bookmaker's margin in it, and reads low for everyone as a result. This page defines what we do instead, and the races where we refuse to compute it at all.

Definitions last changed 27 August 2026. Thresholds on this page are read live from the code that computes them.

What A/E measures

A/E is actual wins divided by expected wins. Expectation comes from the betting market: a runner sent off at 4/1 is implied to win one time in five, so it contributes 0.2 of a win to expectation. Sum that across every run in a sample and compare it with the wins that actually happened.

Above 1.00 means the sample won more often than the market implied. Below means less often. The reason to use it rather than strike rate is that it corrects for class of opportunity: a yard sending out short-priced favourites should win more often, and a high strike rate on its own does not tell you whether it won more often than it was supposed to.

Why the standard version reads cold

A bookmaker's book does not sum to certainty. Add up the implied probability of every runner in a real race and you get roughly 1.10 to 1.45 depending on field size. That excess is the margin, and it is the bookmaker's entire business model.

Feed those raw implied probabilities into A/E and every expectation is inflated by the whole margin, so every A/E figure in the sport comes out low. Across our archive the effect runs 13 to 15 per cent cold.

Uniform error is worse than noisy error. A figure that is wrong by a stable amount looks like a number, behaves like a number, and never flags as broken — which is exactly the kind of plausible wrongness this site exists to remove. A trainer genuinely running at 1.00 reads about 0.86, and nothing on the page tells you so.

How we strip the margin out

Four steps, computed once per runner when a race is written into the archive, so every surface on the site divides the same numbers.

Implied probability, per runner
implied = 1 ÷ decimal SP
A runner with no usable starting price, or a price below 1.01, contributes nothing and is not counted.
The race's overround
overround = sum of implied, across the field
One figure per race, stored on every runner in it. This is the margin, measured rather than assumed.
Margin-free probability
fair probability = implied ÷ overround
The field's fair probabilities now sum to 1.00 — one winner's worth of expectation per race, which is what actually happens.
Fair A/E
Fair A/E = wins ÷ sum of fair probability
Computed only over runs that have a fair probability. That basis is reported separately and never blended with the others.

When we refuse to compute it

A race's book has to be admissible before it can contribute expectation. Four conditions. Fail any one and the overround is left undefined, every fair probability in that race is left undefined, and the race contributes nothing to any Fair A/E figure anywhere on the site.

At least 2 priced runners. One price is not a market.
At least 90% of the field priced. Below that the book is a sample of a market, not the market.
The book sums to at least 1.00. This one is structural, not statistical. A starting-price book cannot be under-round — that would be free money, and it does not occur. So a book summing below certainty is not an unusually generous market, it is proof the prices we hold are incomplete, whatever the coverage figure says.
The book sums to no more than 3.00. Beyond that the input is implausible and is treated as bad data rather than an extreme market.

The cost of this is real and we would rather state it than hide it. Patchily-priced racing — much older racing, and French racing in particular — returns no Fair A/E at all on this site. That is the honest answer rather than a gap in our data. A/E built on a partial book is not a slightly noisy figure; it is a wrong one that looks fine.

Three sample sizes, never blended

The three metrics we publish qualify different rows of the same slice, so they have different sample sizes. All three are shown on every surface, because a figure quoted against the wrong sample size is the commonest way a page like this misleads.

Sample	Counts	Basis for
n	Every run in the slice.	Strike rate, place rate
n at SP	Runs with a parseable starting price.	ROI and profit or loss
n fair	Runs whose race produced an admissible book.	Fair A/E

These can legitimately differ by a wide margin, and the gap between n and n fair is itself information: it tells you how much of a record was run in races we can price properly.

The companion metrics

Defined here rather than elsewhere because they come out of the same computation, and a definition with two homes eventually has two meanings.

Strike rate

strike rate = wins ÷ n × 100

Every run counts, including runs with no usable price.

Return on investment

ROI = (returns − stakes) ÷ stakes × 100

A notional £1 win stake on every qualifying run, settled at starting price, over n at SP. Returns include the stake, so an even-money winner returns £2 and shows £1 of profit. Level stakes only: no staking plan, no compounding, and no attempt to model what a real punter would have got on the exchange.

Sample thresholds

Below these, a figure is flagged rather than suppressed — you still see the number and you also see that it is thin. Nothing on the site makes a comparative claim off a sample under the significance gate.

Gate	Minimum	What it governs
Significance gate	30	Runs. Below this a slice is flagged as a small sample on the page, and no comparative claim is built on it.
Recent-form gate	15	Runs inside the recent-form window. Below this we publish that the sample is too thin to read rather than reading it.
Fair A/E divergence sample	50	Runs. A Fair A/E figure only qualifies a pattern as an angle on a sample this size or larger.
Value-sentence sample	100	Runs. The written read only calls a pattern good value on a sample this size or larger.
Horse Fair A/E floor	5	Runs. An individual horse's Fair A/E card is hidden below this — a career is a small sample by nature.

The archive window

Individual race records begin on 2 January 2005. Anything earlier is not available from our data provider, so a horse that raced before then has an archived record on this site, not a complete one, and we say so on the page rather than implying otherwise.

Aggregate figures for trainers, jockeys, courses, sires and owners begin on 1 January 2008. The later floor is deliberate: it means every horse inside the aggregate window has a complete captured career behind it, so a horse's earlier form is never silently missing from a trainer's numbers.

Trainer and jockey profiles show the last 5 seasons by default, with the full archived record one click away. The default is the shorter window because a yard's methods change and a fifteen-year average hides that.

How to read a Fair A/E figure

Fair A/E compares wins with a margin-stripped market expectation. Favourite-heavy samples — top yards, leading riders — naturally read a few points warm, so 1.05 means "beat the margin-free market read", not "the market underrates this by 5%".

So treat it as a comparison against a corrected market read, not as a measure of how badly the market is wrong. It is at its most useful in the difference between two slices of the same record — a trainer's figure on turf against the same trainer's on all-weather — where whatever bias applies to the sample applies to both sides of it.

Definitions

Fair A/E: Fair A/E is actual wins divided by margin-free expected wins: each runner's starting-price implied probability is divided by its own race's overround before expectation is summed, so the bookmaker's margin is removed from the denominator. A figure of 1.00 means the sample won exactly as often as a margin-free reading of the market implied.
Overround: The overround is the sum of the implied probabilities of every priced runner in a race. A theoretically fair book sums to 1.00; a real starting-price book sums to roughly 1.10 to 1.45 depending on field size, and the excess is the bookmaker's margin.
Fair probability: A runner's fair probability is its starting-price implied probability divided by its race's overround, so the field's fair probabilities sum to 1.00. It is the denominator of Fair A/E, and it is left undefined for any race whose book is not admissible.
Strike rate: Strike rate is wins divided by total runs, expressed as a percentage. It counts every run in the sample, including runs with no usable price, so its sample size is the largest of the three KTB reports.
Return on investment (level stakes at SP): ROI is the profit or loss from a notional one-pound win stake on every qualifying run, settled at starting price and expressed as a percentage of total stakes. It is computed only over runs with a parseable starting price, and returns include the stake.
Admissible book: A race's book is admissible for Fair A/E only when at least 2 runners are priced, at least 90% of the field is priced, and the overround falls between 1.00 and 3.00. A race failing any of these contributes nothing to any Fair A/E figure on the site.

Quick answers

What is Fair A/E?: Fair A/E compares how often a horse, trainer or jockey actually won with how often a margin-free reading of the betting market implied they should have. Above 1.00 means more wins than the market expected; below 1.00 means fewer.
Why does KickTheBookies strip the bookmaker margin out of A/E?: Because a starting-price book sums to more than certainty, standard A/E divides wins by an inflated expectation and reads low for everybody. Measured across the KickTheBookies archive the effect runs 13 to 15 per cent cold, uniformly enough that nothing about the figure looks wrong.
How is fair probability calculated?: Each runner's implied probability is one divided by its decimal starting price. That figure is then divided by the overround of its own race — the sum of every priced runner's implied probability — so the field's probabilities sum to 1.00.
When does KickTheBookies refuse to calculate Fair A/E?: When fewer than 2 runners carry a usable price, when under 90 per cent of the field is priced, or when the book sums below 1.00 or above 3.00. Those races contribute nothing to any Fair A/E figure, which is why patchily-priced racing shows no Fair A/E at all.
How large a sample does KickTheBookies require before trusting a figure?: Any sample below 30 runs is flagged on the page as a small sample rather than hidden, and no comparative claim — best course, strongest angle — is made on one.
What period does the KickTheBookies archive cover?: Individual race records run from 2 January 2005 onwards, and aggregate figures for trainers, jockeys, courses, sires and owners run from 1 January 2008 so that every horse inside the window has a complete captured career behind it.

How the written horse summaries are produced

The short written summary at the top of a horse profile is phrased by a language model from that horse's archived facts — the same runs, wins, dates and official ratings shown on the page. The model computes nothing. Every number in a draft is checked against those facts, a draft that fails the check is never shown, and a summary appears only while it describes the same number of runs as the page. Where no checked summary exists, the profile shows one written by fixed rules instead.

The figures and tables on a horse profile are computed from the archive directly and never pass through a model.