Inspiration
Public football match data is abundant, but coaches and student analysts still need a trustworthy way to turn it into match preparation decisions. A prediction without methodology, uncertainty or backtested error can mislead. ScoutBoard makes the full reasoning inspectable.
What it does
ScoutBoard is a working interactive Premier League analytics application. Analysts can choose any home and away team and get estimated home-win, draw and away-win probabilities, predicted goals and an actionable scouting briefing. A scenario slider illustrates sensitivity to home attacking output. Results-based standings show team form and goal difference. An upcoming-fixture queue prioritizes uncertain matchups. The evidence panel compares chronological model forecasts against a simple league-history baseline, reports a three-class Brier score, displays calibration and exports per-match forecasts as CSV.
How we built it
The dashboard is built with semantic HTML, responsive CSS and vanilla JavaScript. Its forecasting engine uses venue-specific attack and defense scoring rates, smoothed toward league priors with a six-match equivalent regularization weight, then constructs an independent Poisson distribution of scorelines to estimate 1X2 probabilities. A rolling chronological replay predicts each date's matches before incorporating any results from the same date to avoid temporal leakage. The model is evaluated against a rolling smoothed unconditional-outcome baseline. OpenFootball provides publicly available Premier League fixture and result JSON for the 2023/24 through 2026/27 seasons. Unscored matches are treated as future or unreported fixtures, not invented results. Offline mode is clearly labeled as a limited, verified 100-fixture preview rather than pretending to represent the complete league.
Challenges
Mixed scheduled/completed match feeds require strict result validation. Same-day match leakage was prevented through grouped chronological replay. Sparse venue-specific history required shrinkage toward priors. The interface was designed to surface data limitations and forecast error instead of presenting probabilities as guarantees.
Accomplishments
A working, publicly deployed browser application with explainable probabilities, fixture prioritization, standings, backtesting, calibration and export. The JavaScript engine has 10 passing deterministic model tests. No backend, login, API key or installation is necessary.
What we learned
Simple statistical models become more useful when predictions include uncertainty, honest comparisons and an auditable data trail. Historical scoring patterns alone cannot account for lineup availability, injuries or tactical context; the application is for exploratory analysis and coach/scout decision support, not betting.
What's next
Bootstrap confidence intervals, stronger benchmarks, rolling season weighting, independent source quality audits, richer team form features, and prospective validation on newly played matches.
Technologies
HTML5, CSS3, JavaScript, Node.js built-in test runner, Python Playwright, OpenFootball open match data, independent Poisson goal model, Vercel. AI assisted with coding, UI, writing and QA.
Links
- Live app: https://scoutboard-aqx-2026.vercel.app/
- Public source (MIT): https://github.com/samertall/tools/tree/main/hackathons/scoutboard-aqx-2026
- Source data: https://github.com/openfootball/football.json
This project was developed for AQX Data Bowl 4.0 with AI assistance. It does not claim the third-party sports records as original data.
Built With
- ai-assisted-development
- css3
- html5
- javascript
- node.js
- openfootball
- poisson-model
- vercel
Log in or sign up for Devpost to join the conversation.