CA Platense scores shown first. Bivariate Poisson with team xG from recent form — most-likely score ≠ most-likely winner.
Probabilities derived from a Poisson model on expected goals. Edge % shown only when the book posts both sides of a market.
Joint probability comes from the score grid (the truth). Naïve is what a book gets by multiplying legs — they undershoot correlated positive legs like "home + BTTS" and overshoot anti-correlated ones like "draw + over 2.5". Buy positive correlation; avoid negative.
Model probabilities CA Platense 50.3% · Draw 20.4% · CD Coquimbo Unido 29.3% Why CA Platense is favoured Elo essentially level (1505 vs 1510) — winner depends on context. Recent: CA Platense 3W-1D-2L · CD Coquimbo Unido 3W-1D-2L (last 6). Goal differential per game: CA Platense +0.2, CD Coquimbo Unido +0.3. Form depth CA Platense: avg 1.3 goals scored, 1.2 conceded per game (1/3 at home, 2/3 on the road). CD Coquimbo Unido: avg 1.3 goals scored, 1.0 conceded per game (2/3 at home, 1/3 on the road). Goals expectation Model expects ~2.4 total goals (CA Platense 1.2, CD Coquimbo Unido 1.3) — right around 2.5. But — counter-arguments Elo confidence is low (10%) — ratings are based on a thin sample, so the prediction itself is wider than the headline number suggests. Calibration Raw model: 46.5% home · Platt-calibrated up to 50.3% based on historical reliability across this sport's graded matches. Confidence note Elo data on these teams is thin (10% sample-size confidence); headline number is sharper than the underlying signal — treat it as directional, not absolute.