| Argentina | Draw | Cabo Verde | Brier | |
|---|---|---|---|---|
| Model | 68% | 22% | 9% | … |
| Market | 86% | 10% | 4% | … |
Live public experiment
Our models vs the markets
We forecast real-world events — football, crypto price moves, politics, weather, esports — and score ourselves against the markets people actually trade on: Polymarket, Betfair, Binance. The model's probabilities are locked before each event, the market's price is frozen at the same instant, and both get scored against reality afterwards. Every call stays on the record — including the ones we lose. Football is track #1; the rest roll out below.
⚽ Football — World Cup 2026 vs Polymarket
| Australia | Draw | Egypt | Brier | |
|---|---|---|---|---|
| Model | 37% | 30% | 34% | … |
| Market | 26% | 33% | 40% | … |
| Spain | Draw | Austria | Brier | |
|---|---|---|---|---|
| Model | 64% | 22% | 14% | 0.201 |
| Market | 76% | 17% | 7% | 0.089 |
All tracks
One protocol — locked calls, frozen benchmarks, public scoring — across every domain we forecast:
- ⚽Football — World Cup 2026 match odds vs Polymarket live now
- 📈Crypto — price rise/fall calls vs Binance prices + Polymarket crypto markets next
- 🗳️Politics — model vs Polymarket on political events planned
- ⛅Weather — model vs Polymarket weather markets planned
- 🎮Esports — match forecasts vs the markets planned
- 🏟️Sports × Betfair — same protocol with exchange prices as the benchmark planned
Methodology, exactly
- Model call = the latest probabilities written strictly before kickoff (UTC timestamps shown per match). The write path refuses post-kickoff entries and never overwrites earlier calls.
- Market benchmark = Polymarket's traded prices from our last snapshot at or before the model's lock time — both sides frozen at the same instant. For negRisk events the three prices are renormalized to sum to 1 (removes the overround), per the pre-committed rule.
- Ground truth = the market's official resolution (exact 0/1), recorded only after UMA settlement.
- Score = 3-outcome Brier, Σ(p−y)² over win/draw/win. 0 is perfect, 2 is maximally wrong, a coin-flipper lands around 0.667. Lower is better.
- Match odds are regulation-time results (a knockout draw goes to extra time — the market resolves "draw").
Don't trust our timestamps — verify them
- Pre-event posts: each call is published on X and Telegram before the event (linked on the cards above). Those timestamps are set by the platforms, not by us.
- Independent archives: this page is snapshotted by the Wayback Machine after every update — browse the archive history and compare any pre-event snapshot with the final scores.
- Raw data: the full call-by-call log is available to journalists and researchers on request.
This is an educational, transparency-first project by NeuPortal — entertainment and research, not betting or financial advice. Probabilities are for understanding events, not acting on them.