StatLab — Statistical Simulation & Football Analytics Desktop App

A production desktop application with two fully independent modules behind one premium dark-glassmorphism interface:

1. Crazy Time Statistical Simulator — a Monte Carlo laboratory that reproduces the Evolution Crazy Time wheel (54 segments in the physical layout, Top Slot, all four bonus games, calibrated to the published per-spot RTPs). Play it hands-on at the Live Table — chips, a running balance, single spins on the physical wheel and fully playable bonus rounds — or compare 23 betting strategies across millions of independent random spins, including your exact chip layout. It does not predict outcomes — it quantifies the risk profile of staking plans under a fixed house edge.
2. Football AI Prediction Engine — calibrated match probabilities from a seven-model ensemble (Poisson, Dixon-Coles, Elo, SPI-like ratings, gradient boosting, random forest, MLP neural network), with 10,000-run match simulation, walk-forward backtesting (accuracy, log loss, Brier, calibration, ROI, CLV), daily fixture predictions with imported bookmaker odds, an EV-ranked accumulator builder, and a Live Center: every match of the day auto-listed, live scores, statistics, lineups, substitutions, event timelines, momentum charts and in-play Poisson probability updates.

Download (Windows)

From the Releases page:

• StatLab-Setup.exe — recommended. A proper Windows installer (Start-menu & desktop shortcuts, uninstaller). Once installed, the app updates itself from inside — open Updates and click Download & install; no manual re-download.
• StatLab-windows.zip — portable build; extract anywhere and run StatLab.exe.

No Python required. Your data (match database, reports, live settings) is stored in a data/ folder next to the app.

Staying up to date

StatLab checks GitHub for new releases on startup and flags the Updates page. On the installed build, one click downloads the new installer and applies it in place.

Run from source

pip install -r requirements.txt

# native desktop window (Edge WebView2)
python run_desktop.py

# — or — serve to your default browser
python run_server.py

Live football data — no API key needed

The Live Center's default auto provider cross-references multiple open sources with no registration: ESPN public JSON feeds (22 competitions: live scores, statistics, lineups, events) and OpenLigaDB (German leagues), merging duplicate fixtures by team-name matching and labelling every match with its sources. The full source registry lives in app/football/live/sources.json — add an adapter and an entry to plug in more.

Matches are grouped by category (international national-team games, continental club cups, domestic leagues) then by country and competition, with a category filter — so a World Cup night no longer mixes with club fixtures. Any match in a supported league shows a 🎯 Predict button that runs the full ensemble prediction on the fly (feed team names are matched to the historical database and the league is downloaded on demand).

National teams & World Cup, with player markets

National-team matches (World Cup, Euro, any nation vs nation) get their own model built from the open international results dataset (~49k matches since 1872): a neutral-venue-aware, tournament-weighted Poisson + Elo + Dixon-Coles engine. When the starting XIs are published, the 🎯 Predict + scorers button also produces player markets — anytime / first / 2+ goalscorer and penalty — computed from real international goal data (each player's share of his team's goals scales the model-expected goals into a personal Poisson rate), plus per-player analysis. Card markets are shown as an honest positional estimate (no per-player international card log is available without a paid feed). Club player markets require the optional API-Football key.

Honesty note: diretta.it / Flashscore / SofaScore / FotMob are deliberately not scraped — their terms prohibit automated access and they run anti-bot protection; they are listed as disabled in the registry with the reason documented.

Optionally, a free api-football.com key (100 requests/day) enriches the auto feed further; the app tracks the daily quota and caches responses so the free tier is never burned by auto-refresh.

The app runs entirely on your machine at http://127.0.0.1:8765 (interactive API docs at /docs). First use of the football module: open Data Manager → Refresh all leagues to download the historical match database (football-data.co.uk, ~45k matches across 12 leagues, cached on disk).

Tests

python -m pytest tests -q

20 tests cover wheel calibration against published RTPs, bonus-game mechanics, every betting strategy, Monte Carlo aggregation, all statistical models on a synthetic league with known strengths, feature leakage, ensemble weighting, market simulation and the slip builder.

Layout

See docs/ARCHITECTURE.md for the full design: module boundaries, data flows, database schema, API surface, caching, concurrency and extension points (new strategies, new competitions, wheel updates — all config-driven, no code changes).

app/core       shared config, SQLite, background jobs, PDF/Excel/CSV/JSON exports
app/crazytime  wheel model, bonus engines, Monte Carlo engine, 22 strategies, metrics
app/football   data pipeline, features, models/, ensemble, simulation, backtest, slips
app/static     SPA frontend (vanilla JS + vendored Chart.js, offline-capable)
data/          runtime database, download cache, generated reports
tests/         engine test suites

Responsible use

• The Crazy Time module is a simulator. Every bet spot has negative expected value (95.9–96.1% RTP numbers, 94.3–95.8% bonuses); no staking plan changes that. The UI displays expected loss alongside every result.
• The football module outputs calibrated probabilities with risk indicators and model-disagreement diagnostics — estimates, never certainties, never financial advice.