71.6% Win Rate Over 74 Settled Games: Deconstructing Our Open AI Sports Prediction Engine

The sports betting and prediction market industry is saturated with self-proclaimed "experts" and black-box tipsters who tout winning picks on social media while silently deleting their losses.

When building Oracle inside Rawmind, we decided to eliminate the human element completely. Every pick is generated by an autonomous LLM ensemble, locked onto a publicly visible web ledger before kickoff, and settled automatically against official ESPN box scores and DraftKings closing lines.

The 4-Step Autonomous Pipeline

1. Automated Odds Ingestion: Continuous polling daemons ingest DraftKings point spreads, moneylines, and over/under totals alongside ESPN core match metrics.
2. Structured Tabular Features: The model evaluates team efficiency (xG differentials in soccer, offensive rating in NBA, EPA/play in NFL) alongside travel mileage and rest splits.
3. Pre-Game Ledger Lock: The instant a match begins, the pick is immutable on the public scoreboard. No retrospective edits are allowed.
4. Deterministic ESPN Reconciliation: When ESPN marks a game final, the daemon pulls the official box score and automatically stamps the pick as a WIN or LOSS on the site.

Verified Performance Record

Across the first 74 settled matches, Oracle holds a verified 71.6% win rate (53 wins, 21 losses):

• Soccer (EPL, La Liga, UCL): 31-11 (73.8% Win Rate) — Strongest edge on home favorites with superior xG differentials.
• American Football (NFL): 14-6 (70.0% Win Rate) — Driven by situational turnover margins and pass-protection grades.
• Basketball (NBA): 8-4 (66.7% Win Rate) — Exploiting rest advantages against teams on back-to-backs.