WC2026 Model Tracker
Bivariate-Poisson predictions vs actual results Β· a live, honest experiment

πŸ‡ΊπŸ‡Έ USA vs Belgium πŸ‡§πŸ‡ͺ  WINNER MISS

FIFA World Cup 2026, Round of 16 Β· Lumen Field, Seattle (USA host)
Kickoff Jul 07, 2026 Β· 00:00 UTC
Round of 16 Β· who goes through
To advance: USA 50.4%  /  Belgium 49.6%
= P(win in 90β€²) + Β½Β·P(draw); a level game at 90β€² goes to extra time / penalties (treated as a coin flip). 1X2 and Asian handicap (USA +0.00) below are the 90-minute markets.
Final (90β€²) Β· USA 1–4 Belgium
Belgium advanced Β· model to-advance pick USA 50.4% ADVANCE CALL MISS
Goals: 9' Charles De Ketelaere (Belgium); 31' Malik Tillman (USA); 33' Charles De Ketelaere (Belgium); 57' Hans Vanaken (Belgium); 90' Romelu Lukaku (Belgium)
Actual stats: xG 0.63 – 2.01 Β· 8 corners Β· 2 cards (0 red)

Prediction vs result  4/8 markets hit

MarketModel calledActualOutcome
1X2USABelgium (1-4)MISS
Score1-11-4MISS
O/U 2.5Over 2.5 (56.2%)5 goalsHIT
BTTSYes (60.5%)YesHIT
CornersOver 8.5 (66.5%)8 cornersMISS
CardsUnder 3.5 (65.0%)2 cardsHIT
Red cardNo (87.8%)NoHIT
AHUSA +0.00LossMISS

Model vs market

MarketModelBook oddsBook p*EV
USA win37.6%2.3540.2%-11.7%
Draw25.7%3.5027.0%-10.2%
Belgium win36.8%2.8832.8%+5.9%
Over 2.5 goals56.2%1.6756.9%-6.4%
Under 2.5 goals43.8%2.2043.1%-3.6%
Both teams score: yes60.5%1.5360.8%-7.2%
Both teams score: no39.5%2.3839.2%-6.2%
Over 8.5 corners (exp 9.98)66.5%β€”β€”β€”
Under 8.5 corners33.5%β€”β€”β€”
Over 3.5 cards (exp 2.99)35.0%2.1044.2%-26.4%
Under 3.5 cards65.0%1.6755.8%+8.3%
Any red card12.2%β€”β€”β€”
Model-only estimates (no straight market equivalent): expected goals USA 1.49 – 1.48 Belgium Β· fair Asian handicap USA +0.00
In the table above, Model = our probability and Book odds / Book p* = the bookmaker's (p* has the margin removed). EV = value per unit staked at the model's probability (positive = model sees value). β€œβ€”β€ = market not in the captured feed.
Model reasoning (chain of thought)
PREDICTION RATIONALE β€” USA vs Belgium  (FIFA World Cup 2026, Round of 16)
Venue: Lumen Field, Seattle (USA host)

1. DATA. Last-10 internationals (Sofascore interception scraper).
   USA: record WWLLWLWWLW, 2.30 GF / 1.70 GA,
   xG 1.48 / xGA 1.28.
   Belgium: record DWWDWWDDWW, 3.00 GF / 0.80 GA,
   xG 2.09 / xGA 0.92.
   Opponent-strength tiers (configs.MASTER_TIERS) applied per match.

2. RATINGS. USA atk 1.39 def 0.97; Belgium atk 1.43 def 0.80.
   NEUTRAL VENUE: home boost / away penalty set to 1.1491/0.8529 (both teams
   travel to a host-nation site; no host advantage applies β€” a key difference
   from the Korea/Canada/USA host-nation games). lambda USA 1.49, Belgium 1.48
   after 90% model weight + shrink to 1.3 baseline.
   Injuries: {'USA': ['No major absences reported'], 'Belgium': ['No major absences reported']}.
   Context: Round of 16. USA (host, nominal home) at Lumen Field vs Belgium. neutral=False applies the generic home edge to the USA (sea-level venue, appropriate). Belgium needed ET to get past Senegal. Talent prior reinstated (2026-07-06). Knockout: a draw goes to ET/pens.

3. MODEL. Bivariate Poisson (lambda3 0.12). 1X2 USA 37.6% /
   draw 25.7% / Belgium 36.8%. Winner USA; score 1-1; fair AH USA +0.00.
   Over 2.5 56.2%; BTTS 60.5%.

4. EVENTS. Referee TBD: 4.2 Y + 0.13 R/match
   blended (weight 0.20) with WC base 2.6438 -> expected
   cards 2.99, P(over 3.5) 35.0%; P(any red)
   12.2%. Corners expected 10.0 (scale
   0.964) -> P(over 8.5) 66.5%.

5. LIMITATIONS. 10-match samples; hand-set tiers/priors; Poisson corners; lineups
   probable not confirmed; card base rate calibrated on only 4 WC matches.