Diego Garcia

live · review room

VAR

Every edge goes to review.

KairoX Labs

Three football models that check themselves against the bookmakers.

Tempo
1X2
Blaze
BTTS
Storm
O/U
Kick-off

What it is

Three independent models, one per market, each trained separately for every league.

VAR puts probabilities on football matches in three markets: match outcome, both teams to score, and total goals. Each market has its own model, and each league its own champion per model.

The bookmakers' closing odds never enter the features. With the margin stripped out, they become the benchmark: every model is scored against the de-vigged closing line on the same matches.

There is no neural network anywhere, only gradient-boosted trees and Poisson rating models.

It exists to find out whether the models know anything the market doesn't, so most of the work goes into the checking.

The name: variance, and the video assistant referee.

Three angles

Architecture

Pick a camera, pick a league. The diagram redraws from that league's champion, and one example fixture runs through all three.

Champions board · 14 leagues · Tempo

Tempo: features selected, then the fitted Poisson pool weight w. Left is all XGBoost, right is all Poisson.

championLa Liga · N 1,900 · ensemble · F 32 · w 0.53

Branch A: 12 XGBoost members

Features

X: (1,900, 32)

raw (N_raw, 101) · home-minus-away deltas, shift(1)-lagged per team

raw · 101
candidates · 91
selected · 32
  • 37 rolling
  • 13 running-table
  • 6 head-to-head
  • 4 attack/defence ELO
  • 25 style/matchup
  • 6 others

4 regimes × 3 seeds

(B, F) → (B, 3)

multi:softprob, num_class=3

slow deep: lr 0.018 · 764 rounds · depth 6 · gamma base

width = rounds, height = depth, one outline per seed (42, 1042, 2042)

Uniform mean

(B, 3)

stack (12, B, 3)

2,257 rounds × 3 seeds × 3 classes = 20,313 trees

Branch B: Poisson member

Team ids

(B, 2)

Rates λ

(B, 2)

λ_home = exp(μ + home_adv + att[h] − def[a])

λ_away = exp(μ + att[a] − def[h])

L-BFGS-B, time decay ξ = 0.002/day, no L2

2 + 2T + 1 scalars · T = 26

example fixture · drag

Score grid

(B, 13, 13)

away →home →0044881212

1 : 1 · 13.0% · likeliest score

0 to 12 goals. Brightness is probability.

Outlined: the 4 low-score cells with the Dixon-Coles correction. Then renormalise.

Dixon-Coles ρ fitted second-stage, bounded to [−0.3, 0.3]

Read-out

(B, 3)

triangle sums, for the example fixture:

homedrawaway

Log-opinion pool

(B, 3)

p ∝ p_xgb^(1−w) · p_poisson^w

log-opinion pool, row-normalised. w is one scalar fitted on out-of-fold rows.

w = 0.53

0 · XGBoostPoisson · 1

Calibration

(B, 3) = P(home), P(draw), P(away)

softmax(log p / T)

temperature scaling, one scalar, bounds [0.25, 4.0]

Drawn: what T does to one probability, at four temperatures.

Under review

Training and promotion

A challenger has to win the replay, then clear every check.

Walk-forward CV · expanding by season · 3 folds

21-2222-2323-2424-2525-26
fold 0train · 2 seasonsvalidate
fold 1train · 3 seasonsvalidate
fold 2train · 4 seasonsvalidate
finalrefit on all N rows. nothing held out.

Minimum two training seasons. A partial season is dropped before any of this starts.

Registry · step 1 · rank

  • Paired bootstrap on per-match out-of-fold log loss, against the champion.
  • 10,000 resamples.
  • The challenger must win in ≥ 95% of them.
95%

10,000 resamples

≥ 95% to pass

one dot per resample. the challenger needs the yellow ones.

Registry · step 2 · gates

check complete

Checking: Tempo

Checking: Blaze

Checking: Storm

vs climatology

  • beats climatology
  • beats climatology
  • beats climatology at 2.5
  • no line worse than climatology + 0.005

vs the market

  • market gap not worse than the champion's
  • market gap not worse than the champion's

not asked

calibration

  • ECE ≤ 0.05
  • ECE ≤ 0.05
  • ECE ≤ 0.05 on all lines

returns

  • return CI upper bound > 0
  • return CI upper bound > 0

not asked

fold stability

  • fold std < 0.02, worst fold ≤ mean + 0.05
  • fold std < 0.015, worst fold ≤ mean + 0.03
  • worst fold ≤ 0.03

own market

  • draw recall > 0
  • AUC CI lower bound > 0.50
  • BTTS rate in [0.40, 0.70]
  • monotonicity = 100%
  • ratings sanity

Checking: Tempo

  • beats climatology
  • market gap not worse than the champion's
  • ECE ≤ 0.05
  • return CI upper bound > 0
  • draw recall > 0
  • fold std < 0.02, worst fold ≤ mean + 0.05

Checking: Blaze

  • beats climatology
  • market gap not worse than the champion's
  • AUC CI lower bound > 0.50
  • ECE ≤ 0.05
  • return CI upper bound > 0
  • BTTS rate in [0.40, 0.70]
  • fold std < 0.015, worst fold ≤ mean + 0.03

Checking: Storm

  • beats climatology at 2.5
  • no line worse than climatology + 0.005
  • monotonicity = 100%
  • ECE ≤ 0.05 on all lines
  • worst fold ≤ 0.03
  • ratings sanity

A league's first run is seeded as champion regardless of gates.

Offside lines

Shared vs separate

Three lanes. A bar is something the models have in common. A cut is something that never crosses between them.

TTempo1X2

BBlazeBTTS

SStormO/U

  • One raw CSV per league feeds all three models.

  • One partial-season rule and one walk-forward CV scheme. Duplicated code, same logic.

  • The same Poisson rate formula in all three.

  • The same four regimes and three seeds in Tempo and Blaze.

  • The same registry rule.

  • No fitted object crosses models. Each league fits three separate Poisson models.

  • Tempo's and Blaze's Poisson fits use identical inputs and settings, so their ratings are equal. Each still computes its own.

  • Feature caches are separate. Blaze recomputes its three Tempo-style quality features itself.

  • No cross-package import. All three carry their own copy of the training logic.

(B, 3) · (B,) · (B, 5)

Slate / dashboard

The three outputs join in one place only: the slate and dashboard layer, per fixture.

Laws of the game

Principles

Why it is built this way.

  1. 1

    Walk-forward only

    Folds go by season, in order, and are never shuffled. Every headline metric is pooled out-of-fold predictions.

  2. 2

    The closing line is the benchmark

    Closing odds are deliberately kept out of the features. De-vigged, they are what every model is scored against, on the same matches.

  3. 3

    Rank, then gate

    Candidates are ranked on log loss alone, and every other metric is a pass-or-fail gate. An earlier composite score counted log loss and Brier as if they measured two different things.

  4. 4

    Pre-registered rules, tracked separately

    Rules written down in advance are kept apart from rules the grid search found.

A champion lasts until a challenger beats it. The checks stay the same.

Replays

Related writing

Results, experiments and failures live in Briefs and Logs.