EdgeworthArena
Live dataSign in
lab A quant laboratory for sport · fitted on real results · documented, taught, filmed

Sport is a maths problem.

Every match, race and rally reduced to the equations that decide it. Forty models fitted on fifty thousand real results, written up as specifications you can read, taught as courses you can take, set out in five books, and rendered as films you can watch. Licence the models, learn to build them, or just enjoy the mathematics of the games you love.

40
Fitted models
19
Specifications
6
Courses
5
Books
50,000
Real results
Rendered from the model The free kick, integrated Drag, gravity and the Magnus force, step by step, bending the ball round the wall. All the films →
P(x,y) = τ·e−λλx/x!·e−μμy/y!Dixon–Coles P(game) = p4(1+4q+10q2) + 20p3q3·p2/(1−2pq)Tennis Markov chain dv/dt = f − v/τ, dE/dt = σ − f·vKeller pacing m dv/dt = mg − ½ρCdA|v|v + ½ρClA|v|²(ω̂×v̂)Ball flight V(t,d) = maxa Σ P(goal|a,d)·V(t+1,d′)Strategy solver BS = 1/N Σ (pi − oiBrier score P = W/(1+e−(t−t0)/τ) + CPCritical power p̂ = (n·p + k·μ)/(n + k), k = σ²within/σ²betweenEmpirical Bayes win% ≈ PF13.9/(PF13.9 + PA13.9)Pythagorean rSB = 2r/(1+r)Split-half reliability
Read, learn, watch

The mathematics, three ways in

A book to read, a course to take, a model to study. Each one drawn from the same fitted models that run the rest of the site, so nothing here is a toy version of the real thing.

Start with the maths

Where do you want to go?

Learn the models in the Academy, read them in the Library, study their specifications, watch the films, turn the dials in the Lab — and then, if you like, price a match and see how the market disagrees.

Academy

Six courses that teach the maths on this site, with figures drawn from the live models, worked examples, quizzes that explain themselves, and a certificate that verifies.

Start learning →

Library

Five books, the validation report, model cards and league data sheets as PDFs, the certifications, and the model licences. The first chapter of everything is free.

Browse →
P(x,y)=τ·e−λ

Quant Library

Nineteen models written up properly: what goes in, what it does, what comes out, what it was measured at, and where it stops being trustworthy. Free to read, licensed to use.

Read the specs →

Films

Short films of the equations, six of them rendered in 3D from the physics, three studio pieces of boot and ball in slow motion, and flight explorers with wind and altitude on the sliders.

Watch →
∂V/∂t= 0

The equations

Forty equations, each with a picture that moves. Fitted, physics, literature or solved: the evidence grade is on the card.

See them move →

Lab

Sixteen live models with sliders. Move a parameter and watch the optimal strategy move with it.

Open the lab →

Predict

Pick two teams. Every market, the model's fair prices, the confidence band and the words that explain it.

Price a match →

Live

This week's fixtures, the tables, the tennis draws and the NBA slate, each with the model's number beside the market's.

See the week →

Play

Six games where the model is the opponent. Manage a real match, take ten penalties against a keeper who learns, pace an 800 m.

Enter the arena →

Factors

Weather at the ground, altitude, the crowd, the journey, gear and fuelling. Each one graded by its evidence, each one a toggle on the model.

Stack them →
Gear

What the kit is worth

Shoes, spikes, rackets, strings, suits and balls, brand by brand, measured by the papers, modelled by the physics, expressed as a change in the chance of winning.

Compare the kit →

Research

One event modelled every morning as a short paper, and the news desk that scores the week's stories against what the models say.

Read today's →

Paper desk

The model's price beside the market's on every fixture, a paper bankroll to test any rule you like, and a walk-forward that says plainly that not betting beats all of them.

Test a rule →
Maths is the subject

The equations that decide it

Nothing here is a hunch. Each model is written down, fitted on results, and scored against the closing line. These are the six that do the most work.

FootballDixon–Coles

Goals are Poisson arrivals

P(x, y) = τρ · e−λλx/x! · e−μμy/y!
λ = attackhome × defenceaway × home advantage, decayed by time; τ fixes the low-score dependence.

Fitted on every result since 2014, blended with shots on target because shots are a less noisy signal than goals. Taught in Match Pricing →

FootballFitted this week

The chase is real

Second-half scoring by half-time score, after controlling for how good both sides are. A side one goal down scores faster; a side ahead barely slows. The Manager game runs on this.

FootballDecision process

When to press, when to park

V(t, d) = maxa Σ P(goal | a, d) · V(t+1, d′)
Solved backwards from the final whistle for every minute and goal difference. The answer is a policy, not an opinion.

Chase variance when you are behind, kill it when ahead, time the switch by the clock. Try to beat it in the Manager game.

TennisMarkov chain

A 2% edge becomes 25%

P(game) = p4(1 + 4q + 10q2) + 20p3q3 · p2/(1 − 2pq)
Point → game → set → match, nested. Ratings from this season's ATP and WTA results give p for real players.

Morris importance says which points matter. Spend your big serves there. Break Point tests whether you do.

AthleticsOptimal control

Pace is a store, not a feeling

dv/dt = f − v/τ,  dE/dt = σ − f·v
Keller's runner: propulsive force limited, energy store draining. The optimum for 800 m is full force, then hold.

Split Second lets you set your own splits against the optimal ghost and watch the store drain. Taught in The Athlete's Equations →

BasketballBallistics

The rim is bigger from above

m dv/dt = mg − ½ρCdA|v|v + ½ρClA|v|²(ω̂×v̂)
Drag, gravity and the Magnus force integrated step by step. The most forgiving free throw leaves at about 52°.

The same integrator bends the free kick round the wall and dips the kick serve inside the line. See it in 3D on Films. Taught in Physics of the Ball →

Honest numbers

Measured, and reported either way

A model that is never scored is a story. Every one here is tested on results it did not see, and the report says where it wins and where it does not — including where the market forecasts better than the model.

Walk-forward test
27.3k
forecasts across thirteen leagues, every one made before the result was known. Every league's score →
Beyond goals alone
+8.6%
of the gap to the market closed by adding shots on target to the goals model
What is a skill, what is noise
0.55
split-half reliability of a team's foul rate, against 0.18 for corners and 0.05 for goals: fouls are a trait, corners are mostly luck. The booking-market model →
Where the market wins
0.594
mean Brier score across thirteen leagues, still behind the closing line's 0.578 in every one. It says so on every page that could hide it.
Five sports

One method, five arenas

FB

Football

Dixon–Coles with time decay and shots, contextual effects, referee tendencies, state-space form, style clusters, game-state effects, the strategy solver, pitch control, penalties as a game, free-kick flight.

TN

Tennis

Nested Markov chain, Morris point importance, live ratings from this season's tour results, serve flight with topspin.

BK

Basketball

Expected points per shot location, possession value, endgame decision trees, Pythagorean expectation, the forgiving arc.

AT

Athletics & Olympics

Keller optimal pacing, Banister fitness–fatigue and the taper, medal-table allocation by marginal return.

SW

Swimming

Critical power with cubic drag, even-pace optimality, the underwater breakout distance.