How it works

Methodology

How MaxEdge finds value

No tips, no black box. Here is exactly what happens between a bookie’s price and a pick on your board — and you can check every step.

01 The problem

A price is not a probability

Bookmakers don’t publish true probabilities — they publish prices designed to make money. Every market carries a built-in margin (the “overround”, or vig), so the implied probabilities of all outcomes add up to more than 100%.

How much? Measured across the 2,588 English league fixtures we hold closing prices for in 2025/26, the margin runs from 5.7% in the Premier League to 9.9% in the National League. It is not a rounding error, and it gets heavier the further down you play.

Implied probability  =  1 / decimal odds
IMPLIED · WITH VIGHome 50.0%Draw 28.6%Away 25.0%Sum 103.6% — everything pastthe dashed line is the book’smargin (3.6%).FAIR · DE-VIGGEDHome 48.3%Draw 27.6%Away 24.1%
Worked from Home 2.00 · Draw 3.50 · Away 4.00 — the figures above are computed from those prices, not typed in.

To compare a price to fair value, that margin has to come out first.

02 Best price

Taking the margin off 15+ books

Every bookmaker’s price contains its margin. Averaging the implied probability across 15+ books and normalising the result to exactly 100% removes it, which gives a margin-free reading of what the market as a whole thinks — and, more usefully, shows where the best available price on each outcome actually is.

Fair P(outcome)  =  mean(1 / odds)  ÷ Σ mean(1 / odds)

Two guards keep the arithmetic clean: a single book pricing an outcome wildly out of line (a “palpable error”) is discarded before averaging, and a market is only read when every outcome in it is priced.

What this is not. Until August 2026 we published picks off this number as though it were a forecast. It is not one, and the reason is in the arithmetic above: take the margin out of the market and what you have left is the market’s own opinion, so it cannot meaningfully disagree with the thing it was derived from. Measured over 193 finished picks it lost 47% of everything staked on it. Finding the best price is a genuinely useful job and it is the only one this does here now — it no longer publishes picks.

03 The bar

What a model must prove first

A model doesn’t go live because it looks clever. It goes live once we have measured how far off its probabilities usually run, and shown that the prices it picks beat where the market finishes. Models that can’t show both stay off the board. Two publish today.

Both of them are on the models page with their full settled records — every pick that has finished, what a level stake returned, and whether the price beat the close. The band-by-band record of where each forecast has been right and wrong is on the calibration page, including the bands where it loses.

04 Comparability

Putting two leagues on one scale

A rating is only worth what it was earned against. Our second forecast is an Elo ladder built from results alone — no prices anywhere in it — and Elo is zero-sum inside a match: what one club gains the other loses. So a set of clubs that only ever plays itself keeps its own total forever and drifts nowhere.

That is not a corner case. Six of the leagues we track play 0.0% of their fixtures against a club from another league — MLS, Liga MX, J1, the Chinese Super League, the Russian Premier League and the Argentine Primera. Every club in them sits near the starting rating by construction. A naive comparison would call a European tie between a 1620 club from one pool and a 1620 club from another a coin flip, and be confidently wrong.

The fix is one offset per league, added to both ratings before they meet:

d  =  (elo_home + offset[league_home] + home_advantage)
   −  (elo_away + offset[league_away])

The offsets are fitted by maximum likelihood of the realised result over cross-league fixtures — European ties and the League Cup, where two pools actually meet. No market number enters the fit at any point, which matters more here than anywhere: fitting a model to agree with the price makes it reproduce the price, and then there is nothing left to disagree with.

Rating points added before two clubs are compared · hairline is the origin

Nothing in that fit knows what a strong league is. It only knows who won. The order it returns is the one you would have written down yourself, and the English pyramid falls out in its own right order from the top division to the fourth — which is the only external check an unpriced fit can be given, and the reason the whole ladder is printed above rather than the top of it.

Where it cannot answer, it says so by name. A league with no cross-league fixture in the entire record has nothing to anchor it, so it gets no offset — and a fixture whose clubs cannot both be placed gets no forecast rather than a guessed one. The alternative is a number that looks exactly like every other number on the site and is made up, which is the failure this whole page exists to argue against.

05 The core number

What your edge actually means

Your edge answers one question: for every £1 you put on this price, how much should you expect to win or lose over the long run? (The textbooks call it Expected Value.)

Edge (EV)   =  modelProb × decimal odds − 1
modelProb   =  (edge + 1) / decimal odds

Model gives 55% on a 2.10 price:
Edge = 0.55 × 2.10 − 1 = +0.155  →  +15.5% EV

A positive edge means the price is bigger than it should be. It is the same number everywhere on the site, because every page uses the same formula.

The value ladder — the same component you meet on the board

gap +6.1pp vs fair
Implied %
The raw price read as a probability, margin included — 1 ÷ odds. A market’s three implied percentages always sum to more than 100.
Fair %
The same market with the vig removed, normalised to exactly 100. This is what the market really thinks, and it is the baseline an edge is measured against.
Model %
Our own probability for the outcome, produced independently of the price.
Edge
The EV, shown next to the gap between Model and Fair in probability points (e.g. +6.1 points vs fair) — the intuitive read of how far our number sits from the fair market price.
06 Ranking

How we score every pick

The score answers one question: how unusual is this disagreement with the market? Raw edge alone cannot answer it — a +9% edge at 2.06 is a four-point disagreement, and the same +9% at 12.0 is barely half a point. So the score is built on the probability gap, measured against how far off this model’s probabilities usually run:

fair       =  the Fair % beside the pick     the price with the margin taken out
ours       =  (1 + edge) / odds             what our edge at that price implies
error bar  =  how far off this model’s probabilities usually run — MEASURED, not chosen

gap in error bars  =  (ours − fair) / error bar
score              =  100 × (1 − 0.35 ^ gap in error bars)

That makes a score readable as a sentence: 65 is a disagreement exactly one measured error bar wide, 88 is two. Nothing picks those numbers — invert the formula and every cutoff on the ladder falls on a whole number of error bars. A model whose error has never been measured gets no score at all rather than a flattering one.

The market side is the Fair % you can see — the price with the bookie’s margin taken out, which is the same number the ladder beside every pick draws. So the gap the score measures is the gap you are shown: one number doing one job, not two that agree when you are lucky.

There is one score. Until August 2026 the board also carried a Kelly-shaped rating, which is the right shape for deciding how much to stake and the wrong shape for deciding how sure we are — it contains no error bar, so the ladder’s rungs meant nothing when applied to it. Staking lives on its own axis now, in units, and the score answers one question only.

What the score is not. It measures how unusual a disagreement is, not how likely it is to pay. Step 10 has the measurement.

07 Coverage

What we price, and what we signal

Priced and signalled are not the same thing, and it matters which one you are looking at. We read five markets from 15+ books on every tracked fixture, so the board shows you the best available price on all of them. Publishing a pick on a market needs a model that has cleared step 03, and today one has.

Priced · all five markets
Match Odds, Over/Under, Both Teams to Score, Corners and Cards — de-vigged across the book panel, so you can see the best price and the fair line side by side. This is arithmetic on the market’s own numbers and needs no model to be true.
Signalled · goals only
The Dixon-Coles sheet publishes on Over/Under and Both Teams to Score, and nothing else. It is a goals model, so those are the markets it has a forecast for.
Not signalled · 1X2, corners, cards
No model of ours has earned the right to disagree with the market on the result, and the corners and cards models are still paper-trading (step 03). You get the prices; you do not get a pick we cannot evidence.

One page per match holds all of it — the prices, what the engine found, and the model’s read where there is one — on any fixture we track.

08 The proof

Did we beat the closing price?

Win rates over small samples are mostly noise. The honest measure of whether an edge was real is Closing Line Value — did our signal price beat the market’s final price at kick-off? Consistently beating the closing line is the best available predictor of long-term profit, and we track it on every settled signal.

CLV  =  ln( signal odds / closing odds )

Signalled 2.10, closed 1.90  →  ln(2.10/1.90) = +10.0%  beat the close

But that closing price still has the margin in it. A book that closes 1.90 is not saying the true chance is 52.6% — it is saying 52.6% plus its cut. So beating the raw close is easier than it sounds, and the number we actually judge ourselves on takes the margin out of the closing line the same way step 02 takes it out of the opening one.

The difference is not cosmetic. Over the settled book those two measures read +3.88% and +0.43% — the same bets, the same closes. The second is the one on the performance page and the one a model has to clear +0.5% of to publish again, because it is the one that is hard.

Every signal is settled against the real result and disclosed either way — wins and losses, with the closing price we were measured against.

09 Trust

Checked, and checked again

Outlier filter
Palpable-error prices are dropped before they can create a fake edge.
Market shrink
Data-driven markets are pulled toward the market price, capping how far any signal can diverge.
No score without a measured error bar
A model that has never had its error measured cannot carry a score at all — the database refuses the write, not just the display. That is what stops an uncalibrated model quietly looking like a confident one.
Integrity check
Every engine cycle re-checks that all probabilities, edges and scores are internally consistent, and alerts on any violation.
10 Honesty

What we can’t do

No model is perfect. Team-strength and goals inputs update periodically and don’t yet capture late lineup changes, weather or motivation. The corners and cards models lean on patchy international stats, which is exactly why they’re anchored to the market. Treat every MaxEdge number as one input among many.

And the score does not predict the payout. Across the settled fixtures we hold a closing line for, the correlation between a pick’s score and its realised closing-line value is 0.035 — indistinguishable from zero. The edge in the book is real; it is just as real on a 70 as on a 45. The score tells you how unusual a disagreement is, and we present it as conviction, which is a stronger claim than the measurement currently supports. It is on this list until that changes.

Please bet responsibly

18+. MaxEdge is model output and market data for information only — not financial advice and not a guarantee. Tracking only: we never place a bet. Never stake more than you can afford to lose · BeGambleAware