Data science / Analytics

Sports intelligence platform

A professional analytics terminal with calibrated match predictions built from historical data.

Client typeSports analysts and media

StatusStage 7 delivered

Live sitestatvector.app

Visit live site

Live demo

A simplified, self-contained version of the core idea. Everything runs in your browser with sample data.

Score model · Poisson + Dixon–ColesInteractive demo · sample data · runs in your browser

Home expected goals (λ)1.60

Away expected goals (μ)1.10

Low-score correction (ρ)-0.10

Home win	47.7%
Draw	27.3%
Away win	25.0%
Over 2.5 goals	50.6%
Most likely score	1–1 (13.0%)
Fair odds (home / draw / away)	2.10 / 3.66 / 4.00

Score probability (rows: home goals, columns: away goals)

01234560864100011013721002995210035531000422100005110000060000000

In the full platform λ and μ come from attack/defence ratings built only from matches before kick-off, and are blended with Elo, Bradley–Terry and regression models.

Loading demo…

The problem

Most sports predictions are opinions. The goal was a transparent engine whose probabilities can be checked against history without look-ahead bias.

Our solution

A Next.js terminal backed by a modelling pipeline: Elo, Bradley–Terry, Poisson, Dixon–Coles, regularised regression and seeded Monte Carlo simulation, combined into one calibrated distribution and checked with walk-forward backtesting.

Look-ahead-bias guard in feature building
Walk-forward backtesting and feature ablation
37-sport directory, five interface languages
Versioned /api/v1 read API and model governance records

Result

Stage 7 delivered: dark-mode terminal, multilingual UI, canonical JSON contracts, a read API, and model evaluation records.