← all fixtures
Portugal
vs
Croatia
Round of 32 · 7/2/2026, 11:00:00 PM
1-0
alt 1-1 · xG 1.64–0.82
5conf
actual 2-1 ✓ correct
Score-probability matrix
0
1
2
3
4
5
6
0
9
6
1
13
12
2
11
9
3
6
5
4
5
6
↓ Portugal goalsCroatia goals →
Win / draw / loss
W 56%D 26%L 18%
Market
57% / 26% / 18%

Portugal favoured (56% to win); the model's most likely scoreline is 1-0.

Confidence 5/10. Even the favoured outcome (56%) fails about 44% of the time; a single match is high variance.

How we got to 1-0
  1. 1
    Model fit
    Team strengths fit from 15,899 matches (Dixon-Coles, recency-weighted, low-score correction ρ=-0.068). Neutral venue — no home advantage.
  2. 2
    Team strengths (log scale, vs average)
    Portugal: attack +0.91, defence +0.89
    Croatia: attack +0.59, defence +0.65
    Higher attack = scores more; higher defence = concedes fewer.
  3. 3
    Expected goals
    Portugal: exp(base +0.18 + att +0.91 − opp.def +0.65) = 1.55
    Croatia: exp(base +0.18 + att +0.59 − opp.def +0.89) = 0.89
    “base” is the league scoring level the model fitted.
  4. 4
    Agent adjustments
    λ 1.55/0.89 → 1.64/0.82
    • Tactical home defence ×1.08 — Baseline already captures long-run strength and recent form, so adjustments are minimal. Portugal's stingy recent defensive record (0.3 conceded) suggests tactical solidity slightly beyond baseline, warranting a modest defensive bump, while Croatia's leaky form (1.7 conceded) hints their backline may be marginally more exposed than baseline reflects. Attacks left neutral with no availability or stakes information to justify movement at a neutral venue.
    • Tactical away defence ×0.95 — Baseline already captures long-run strength and recent form, so adjustments are minimal. Portugal's stingy recent defensive record (0.3 conceded) suggests tactical solidity slightly beyond baseline, warranting a modest defensive bump, while Croatia's leaky form (1.7 conceded) hints their backline may be marginally more exposed than baseline reflects. Attacks left neutral with no availability or stakes information to justify movement at a neutral venue.
  5. 5
    Scoreline
    A Poisson model on those expected goals (with ρ) gives the score-probability matrix above; its most likely cell is 1-0 at 13%. Win / draw / loss come from summing the matrix.
  6. 6
    Confidence
    5/10 — a sharper peak and a clearer favourite raise it; large agent moves lower it (model agreement 88%).
Portugal vs Croatia — predicted 1-0 · Lambda XI