About Degen Predictions
Free ML forecasts across sports, crypto, and stocks — with honest limits.
What this site does
Degen Predictions generates free forecasts across sports, crypto, and stocks. Sports cover six soccer leagues (Premier League, La Liga, Serie A, Bundesliga, Ligue 1, MLS), MLB and NPB baseball, cricket (IPL, BBL, CPL, The Hundred, MLC), and men's ATP tennis — match result / moneyline and related markets from models trained on historical data. Today's Picks and Best Combos surface high-probability multi-market legs (calibrated ~70% bars when available, otherwise a wise ≥65–68% floor). Crypto and stocks are ranked by predicted forward return from a separate ML pipeline. Educational research project — not betting or investment advice.
How fixtures stay current
Upcoming fixtures and predictions refresh automatically every day, and a new soccer round unlocks as soon as every fixture in the previous round has kicked off — so you're always looking at the next set of real, upcoming matches rather than a static list.
Where the data comes from
Soccer: historical results and match statistics come from public football data archives, player-level stats come from Understat (European leagues) and American Soccer Analysis (MLS), and upcoming fixture schedules come from fixturedownload.com.
Baseball (MLB): historical game logs from Retrosheet, live fixtures and probable-pitcher stats from ESPN's public scoreboard API.
Baseball (NPB): schedules and results from spaia.jp's public schedule API.
Cricket: Cricsheet ball-by-ball archives for IPL, BBL, CPL, The Hundred, and MLC; fixtures from Wikipedia (Hundred, CPL) and fixturedownload (IPL/BBL); Open-Meteo weather; provisional last-known XIs for live boards; Elo + ML for winner / runs.
Tennis (ATP men): tour-level match history from TennisMyLife (Sackmann-compatible CSVs; Jeff Sackmann tennis_atp lineage, CC BY-NC-SA — educational use with attribution); live boards from ESPN when available with Wikipedia bracket fallback; overall + surface Elo (hard/clay/grass) and calibrated match-winner ML.
Crypto: price/volume history and market data from CoinGecko's free public API; Bitcoin on-chain metrics (active addresses, hash rate) from blockchain.com. The asset universe is the top ~60 coins by market cap.
Stocks: price history and fundamentals (P/E, margins, growth, debt ratios) from Yahoo Finance via yfinance. The universe is the S&P 500 (US) and NIFTY 50 (India) constituents, sourced from their official index pages.
News sentiment (crypto, stocks, baseball, and cricket live boards): recent headlines from Google News, scored locally with VADER — no third-party sentiment API.
Nothing on this site uses bookmaker odds, brokerage order flow, or any paid data feed as an input — every source above is free and publicly accessible.
Honest limitations
Sports, crypto, and stocks are all high-variance by nature — even a well-built model will miss regularly, and a confident-looking probability or ranking is not a guarantee. Treat every prediction here as an estimate to inform your own thinking, not as a certainty. Full-slate win markets typically land in the mid-50s%. Soccer Safer Result is a different, selective rule: it uses favourite-or-draw at a calibrated 70% bar and only upgrades to an exact favourite win when both calibrated and raw 1X2 scores reach 75%; otherwise it makes no result pick. It is labelled 70%+ only where a later chronological evaluation slice actually cleared that mark with adequate sample size and reported coverage. See the public MLB track record and cricket results pages. For crypto and stocks, rankings pages are upfront about measured overfitting found in backtesting.
Crypto & stocks — honest limits
Market boards rank assets from a separate walk-forward pipeline. They are research leans, not trade signals. Crypto pages often withhold absolute USD return when direction fails holdout — you may see relative lean vs Bitcoin, peer median, or an empirical range instead. Stock cards lead with gap-vs-board / beat-index style reads because absolute window returns are usually positive in long samples. Refresh cadence depends on the ranked CSV (Admin can flag stale files). Always treat outputs as uncertain estimates; this is not financial advice.