Group D · Matchday 3
United StatesvsTurkey
2026-06-25·19:00 localPredictions finalised
The forecast
Match-outcome probability
- United States win41.3%
- Draw27.4%
- Turkey win31.3%
A clash of identities: United States's balanced approach meets Turkey's pragmatic style in a fixture the model gives to Turkey at 43%.
Why the model says this
Favoring Turkey
- ·Elo advantage of 181 points over United States
- ·Expected goals 1.30 vs 1.27
What the model can't fully price
- ·Squad availability: 2 carrying a fitness doubt across the two squads, 1 of them projected starters. The forecast does not adjust for who is missing: its lineup channel currently contributes zero, so this is context the probabilities do not include.
Form check
United States
SteadyUnited States: 3W-1D-2L in their last 6 internationals.
3W-1D-2L in last 6
Turkey
ImprovingTurkey: 5W-1D-0L in their last 6 internationals.
5W-1D-0L in last 6
Analysis
How it plays out
Neither side has a rigid tactical identity. Both adapt to the opponent, so the first 15 minutes will reveal who imposes their plan first. Turkey's aggressive press (PPDA 23.5) against United States's deeper build-up (PPDA 27.7) creates a clear territory question: can Turkey force errors high up, or will United States play through the press and find space behind it?
What decides it
Turkey adjust shape to the opponent. That flexibility is an asset, but it takes longer to settle into a game. Folarin Balogun carries the marginally higher scoring probability (9.9% vs 6.8%).
Off the pitch
Turkey travel 11,463km, 4x United States's journey. Second-half fatigue is a real factor at that differential.
The angle
Likely the last World Cup for Tim Ream. Tournament experience at this level is hard to quantify but hard to replace.
▸Goals & scorelines
Likeliest score 1–1 (13.4%) · xG 1.3 - 1.3
Expected goals
Mean of the Dixon-Coles joint goal distribution. Same fit that produces the most-likely-scoreline list below.
Most likely scorelines
- 1–113.4%
- 0–19.1%
- 1–08.9%
- 0–08.3%
- 1–28.2%
From the Dixon-Coles joint Poisson with the low-score correction. Scorelines are listed in probability order; this is a description of the model's distribution, not a recommendation.
Most likely half-time scorelines
- 0–028.2%
- 0–117.2%
- 1–016.9%
- 1–112.2%
- 0–25.8%
Same Dixon-Coles fit as the full-time list above, with rates halved to a 45-minute window and the low-score correction applied to that 1st-half block. The 0-0 row sits higher here than at full-time because fewer minutes have elapsed.
Goal totals
- More than 0.5 goals91.7%
- More than 1.5 goals73.7%
- More than 2.5 goals47.8%
- More than 3.5 goals26.1%
- More than 4.5 goals12.0%
- More than 5.5 goals4.8%
- Both teams score53.4%
Each row is the probability the match finishes with more than the listed number of goals. Both-teams-to-score is the probability each side scores at least once. All values are marginals of the Dixon-Coles joint goal grid that produces the scoreline list above — not market lines or any other operator construct.
Event-typed probabilities
- United States clean sheetOpposing team scores zero27.2%
- Turkey clean sheetOpposing team scores zero27.8%
Derived from the same Dixon-Coles joint distribution as the scoreline list. These are descriptive event probabilities — see CLAUDE.md §3/§4 (formerly COMPLIANCE.md §4.2.7) for the framing the project uses.
Win-margin probability
- United States by 4+1.5%
- United States by 3+5.5%
- United States by 2+15.9%
- United States by 1+35.5%
- Draw28.0%
- Turkey by 1+36.5%
- Turkey by 2+16.6%
- Turkey by 3+5.8%
- Turkey by 4+1.6%
Each row is the probability the match ends with the listed margin or larger in that direction. Marginal of the Dixon-Coles joint goal grid; the “by 1+” rows plus the draw row sum to 1.
▸How the match unfolds
Over 2.5 goals 47.8% · BTTS 53.4%
Game state through the match
- United States ahead36.3%
- Level26.5%
- Turkey ahead37.2%
Probability of each game state at minutes 0, 15, 30, 45, 60, 75, 90 — derived from two independent thinned-Poisson processes with the Dixon-Coles per-team rates. The three lines always sum to 1 at each minute. The right column shows the state at the match's closing minute.
When the first goal arrives
- 0–1535.0%
- 15–3022.8%
- 30–4514.8%
- 45–609.6%
- 60–756.3%
- 75–904.1%
- No goal7.5%
Probability the match's first goal arrives in each 15-minute window. Homogeneous Poisson with combined rate λ = λh + λa from the Dixon-Coles fit; the seven rows (six windows + no-goal tail) sum to 1.
Half-time / full-time grid
| HT ↓ / FT → | HUnited States win | DDraw | ATurkey win |
|---|---|---|---|
| HUnited States ahead | 22.0% | 5.0% | 1.8% |
| DLevel | 12.3% | 16.7% | 12.6% |
| ATurkey ahead | 1.8% | 5.0% | 22.7% |
Each cell is P(half-time result, full-time result). All nine cells sum to 1. Derived from a halved-λ Dixon-Coles fit for the first half plus an independent-Poisson second-half convolution.
Comeback probability
- United States trail at HT, avoid defeat at FT6.8%
- Turkey trail at HT, avoid defeat at FT6.9%
Joint probability — P(side trailing at half-time AND avoiding defeat at full-time). NOT conditional on trailing at HT. Derived from the same half-time / full-time decomposition that produces the HT/FT grid above; a tied first half is neither a home nor an away comeback opportunity.
Cards
- Expected yellow cardsMean of the Poisson on total yellow cards.3.45
- Total yellows over 2.567.0%
- Total yellows over 3.545.3%
- Total yellows over 4.526.5%
- Any red cardP(at least one red card in the match).9.5%
Referee not yet assigned. Using the 2026 pool-mean per-match rate as a placeholder; the model picks up the referee's personal rate once the assignment is published. Total yellow cards modelled as a Poisson with mean equal to two team baselines plus the referee's deviation from the pool mean. Reds are modelled the same way, independently. See /docs/methodology/.
▸Teams & players
Top scorer: Balogun (9.9%)
Match detail
United States
Model-rated key players: Folarin Balogun (FW) — P(scores) 9.9%; Diego Luna (FW) — P(scores) 3.8%; Haji Wright (FW) — P(scores) 2.8%.
United States under Mauricio Pochettino play a balanced game with 50% possession. Their likely shape is a 4-3-3. They sit deeper and pick their moments to press (PPDA 27.7).
United States will need to leverage their strengths while managing the physical demands of a tournament spread across three host countries. Managing minutes for Tim Ream across what could be seven matches will test the coaching staff's rotation planning.
Turkey
Model-rated key players: Hakan Çalhanoğlu (MF) — P(scores) 6.8%; Kenan Yıldız (FW) — P(scores) 2.7%; Kerem Aktürkoğlu (FW) — P(scores) 2.7%.
Turkey under Vincenzo Montella play a pragmatic game with 50% possession. They apply moderate pressing intensity (PPDA 23.5). They generate a high volume of shots (13.2 per 90) and rely heavily on set pieces (20% of their xG).
Turkey play a pragmatic, results-oriented game that adapts shape to the opposition. Tactical flexibility is their strength. The risk is inconsistency — without a default identity, a poor result can cascade if the team struggles to find a Plan B.
United States historically converts 5.2% of xG from set-pieces, contributing 0.07 expected set-piece goals in this fixture. Turkey converts 20.5% from set-pieces (0.27 expected). Combined, the model expects 0.33 set-piece goals across the 90 minutes.
- P(United States scores set-piece goal) 6.4%
- P(Turkey scores set-piece goal) 23.4%
- P(set-piece goal in match) 28.3%
United States: Timothy Tillman on corners (42 corners), Gianluca Busio on free kicks (per fbref 2021 22) · Turkey: Hakan Çalhanoğlu on corners (28 corners) (per fbref 2022 23)
If a penalty is awarded to United States, the model gives 71.4% conversion, 73.3% for Turkey.
United States primary PK: Folarin Balogun (2/2 in 2022-23, per fbref 2021 22) · Turkey primary PK: Hakan Çalhanoğlu (3/3 in 2021-22, per fbref 2022 23).
Derived from the model's per-fixture forecast joint and supporting reference data (predicted squads, set-piece xG share, PK posteriors, club minutes). See /docs/methodology/ for the full methodology.
Tactical forecast
- PPDA
- 27.7
- Possession
- 50%
- Directness (yds/pass)
- 6.4
- Long balls/90
- 38
- Set-piece xG
- 5%
- PPDA
- 23.5
- Possession
- 50%
- Directness (yds/pass)
- 5.5
- Long balls/90
- 33
- Set-piece xG
- 21%
Style profile per side from StatsBomb open-data aggregation across recent international tournaments (Euro 2020/2024, Copa America 2024, AFCON 2023, World Cup 2018/2022). The tactical-fingerprint badge maps each team’s observed style vector into one of eight canonical archetypes via a rule-based classifier; teams with fewer than three matches of qualifying coverage carry an “insufficient-data” label rather than being forced into a default. Sides outside the StatsBomb-open corpus use FotMob team match stats from recent qualifiers and friendlies instead (possession and shot volume only), marked as partial coverage. PPDA = passes the side allows per defensive action (lower = more intense press). Formation distributions are not yet produced — that head of the §2.7 classifier is pending its own data pull. See /docs/methodology/.
Squad depth
Most irreplaceable starters
United States
- Christian PulisicWingerCover: Alejandro Zendejas · 0.570.27gap
- Tyler AdamsDefensive midfieldNo natural backup0.26gap
- Antonee RobinsonFull-backCover: Joe Scally · 0.770.22gap
Turkey
- Orkun KökçüCentral midfieldNo natural backup0.59gap
- Ferdi KadıoğluFull-backCover: Eren Elmalı · 0.620.30gap
- Zeki ÇelikFull-backCover: Eren Elmalı · 0.620.15gap
Gap = how far a side's rating at the position falls from the starter to his likely in-squad replacement (named under each name). Larger = harder to replace. Descriptive metric, does not feed the published probabilities. Methodology →
Match conditions
- AltitudeNear sea level26 m
- Avg temperatureFive-year mean over the tournament window20.8 °C
- Avg humidity70%
- Heat stressShade WBGT ~22.5 °CLow heat stress
- Pitch surfacetemporary natural grass over artificial turf
Indoor artificial-turf stadium; natural grass is grown on a drainage-tray system over the turf under the translucent roof.
Heat stress is a shade Wet-Bulb Globe Temperature proxy from the venue's climatology mean temperature and humidity; FIFA mandates cooling breaks at WBGT 32 °C. Evening kickoff (local time). These are long-window averages, not a match-day forecast, and they are not inputs to the forecast.
Top scorers · P(scores in this match)
- Folarin BalogunPKFW9.9%
- Diego LunaFW3.8%
- Haji WrightFW2.8%
- Hakan ÇalhanoğluPKMF6.8%
- Kenan YıldızFW2.7%
- Kerem AktürkoğluFW2.7%
Per-player scoring rate from Model #5 (`p_score_per_match`). Reflects each player's npxG/90, expected minutes, team xG share, and the average opposing-team defence. See /docs/methodology/.
Recent match form
Last match player ratings
United States
vs Bosnia and Herzegovina · avg 5.5
Worked well: Their offensive movement and ability to create chances, particularly from wide areas and set pieces, proved effective. They maintained their attacking threat even after a player was dismissed.
Struggled: The team struggled with offside calls, indicating issues with timing runs. A red card also highlighted a lapse in discipline.
Turkey
vs Paraguay · avg 6.0
Worked well: Their early goal provided a crucial advantage, and their defense, coupled with strong goalkeeping, was highly effective in holding off Turkey's attacks, even with a numerical disadvantage.
Struggled: The red card for Aldarete put them under immense pressure for a significant portion of the match, forcing them to defend deep and limiting their attacking outlets.
Player scores from official highlight analysis of each team's most recent match. Observational, not a model input. Methodology →
Video analysis: player performance
Per-player ratings and event breakdowns from official highlights analysis. Tap a player to see their full match timeline.
8Folarin Balogun16'–35'Scored the crucial opening goal and showed good attacking instincts, despite an earlier disallowed goal.
1goals▼
Scored the crucial opening goal and showed good attacking instincts, despite an earlier disallowed goal.
Match timeline
8Timothy WeahScored an early goal from a scramble, demonstrating quick reactions and opportunism.
Scored an early goal from a scramble, demonstrating quick reactions and opportunism.
7Christian Pulisic27'–27'Showed flashes of brilliance with a powerful shot on target and a creative backheel pass that initiated an attack.
1shots1on target▼
Showed flashes of brilliance with a powerful shot on target and a creative backheel pass that initiated an attack.
Match timeline
7Sergiño DestShowed attacking intent from his defensive position with a header that almost scored, contributing to offensive pressure.
1headers▼
Showed attacking intent from his defensive position with a header that almost scored, contributing to offensive pressure.
Match timeline
6Weston McKennie109'–109'Had a goal disallowed for offside, indicating attacking presence and willingness to get into scoring positions.
▼
Had a goal disallowed for offside, indicating attacking presence and willingness to get into scoring positions.
Match timeline
6Sebastian BerhalterDemonstrated a determined attacking effort, indicating a willingness to get into scoring positions.
1goals▼
Demonstrated a determined attacking effort, indicating a willingness to get into scoring positions.
Match timeline
7BIH GoalkeeperMade multiple crucial saves throughout the match, demonstrating excellent shot-stopping ability under continuous pressure.
Made multiple crucial saves throughout the match, demonstrating excellent shot-stopping ability under continuous pressure.
6MimicRegistered one of his team's few attempts on goal, showing some offensive initiative for a struggling side.
Registered one of his team's few attempts on goal, showing some offensive initiative for a struggling side.
6BIH Player #8Showed good attacking initiative by scoring after a run, though the goal was not reflected in the final match score.
Showed good attacking initiative by scoring after a run, though the goal was not reflected in the final match score.
3BIH Player #18Received a red card, severely hindering his team's ability to compete for the remainder of the match.
Received a red card, severely hindering his team's ability to compete for the remainder of the match.
Match observations
- The match was a high-scoring affair with both teams demonstrating attacking intent from the outset.
- USA took an early lead, but Turkey responded quickly to equalize, setting the tone for a competitive encounter.
- The game saw multiple lead changes, with Turkey ultimately securing a narrow victory in the closing stages.
▸Under the hood
Model-by-model comparison
United States vs Turkey
| Model | Weight | Home | Draw | Away |
|---|---|---|---|---|
| EloRating-based strength estimate | 32% | 23.3% | 22.0% | 54.7% |
| Dixon-ColesGoal-process model with low-score correction | 63% | 34.4% | 27.7% | 38.0% |
| Hierarchical PoissonBayesian model with confederation pooling | 6% | 33.3% | 27.1% | 39.6% |
| Bayesian stackingLearned-weight combination | — | 30.8% | 26.6% | 42.6% |
| Ensemble (published)Uniform average + isotonic calibration | — | 32.5% | 24.4% | 43.1% |
How each model works
- Elo
- Each team carries a single strength rating updated after every match by a margin-aware K-factor. Match probabilities come from the logistic function of the rating gap. Elo is fast-adapting but coarse — it sees only who won and by how much, not how the goals were scored.
- Dixon-Coles
- A Poisson regression on team-level attack and defence parameters, fitted via maximum likelihood with an exponential time-decay weighting. The Dixon-Coles correction adjusts the four low-score cells (0-0, 1-0, 0-1, 1-1) where independent Poisson underestimates dependence. Produces full scoreline distributions, not just H/D/A.
- Hierarchical Poisson
- A Bayesian Poisson model fitted via MCMC (PyMC) with hierarchical priors that pool attack and defence parameters within confederations. Shrinks small-sample teams toward their confederation mean — helpful for nations with few recent competitive fixtures. Slower to fit but better-calibrated on the tails.
- Bayesian stacking
- Optimises simplex weights (w_elo, w_dc, w_hp) to maximise the leave-one-out log-score across a walk-forward backtest (Yao et al. 2018). The result is a weighted average of the three component models' probabilities, then isotonic-calibrated. Adds no extra features — just learns which component to trust more from historical accuracy.
- Ensemble (published)
- Equal-weight average of all three component models, followed by per-class isotonic regression calibration fitted on 24 months of walk-forward out-of-fold predictions. This is the probability published on the site. The uniform mean is deliberately simple — it avoids overfitting to the stacking weights' training window.
Three independent component models feed two combination strategies. The uniform ensemble is the published probability; Bayesian stacking uses learned weights. Amber bars flag >5pp divergence from the published number. Full methodology
Probability decomposition (transparency surface)
- Baseline ensemble — P(United States win)30.8%
- + Lineup contribution0.0pp
- + Style-matchup contribution0.0pp
- Published P(United States win)30.8%
Decomposition of the published P(United States win) into the calibrated- baseline plus contributions from the §2.3 expected-XI lineup delta and the §2.7 style-matchup interaction. The §2.7 roadmap is explicit that style effects are second-order to team strength — single-digit-percentage P(win) shifts on extreme style matchups, near-zero on balanced ones. We surface the decomposition for transparency even when the contributions are small; the baseline carries the prediction. Methodology: /docs/methodology.
For this fixture both contributions round to under 0.05pp — the fitted style-matchup pair effect is in the small-magnitude regime the model expects to dominate.
Head-to-head history
| Date | Competition | Venue | Score | Result | xG |
|---|---|---|---|---|---|
| 7 Jun 2025 | Friendly | HEast Hartford | 1–2 | L | — |
| 1 Jun 2014 | Friendly | HHarrison | 2–1 | W | — |
| 29 May 2010 | Friendly | HPhiladelphia | 2–1 | W | — |
| 19 Jun 2003 | Confederations Cup | NSaint-Étienne | 1–2 | L | — |
| 4 Sep 1991 | Friendly | AIstanbul | 1–1 | D | — |
United States vs Turkey, every senior international meeting in the martj42 results dataset (score from United States's perspective; H/A/N = home/away/neutral).
Latest news & match context
No recent headlines for United States or Turkey.
- Stage:
- Group D · Matchday 3
- Date:
- 25 Jun
Ranked by likely importance. None of these feed the forecast: the probabilities rest on team strength, venue conditions and the style matchup.
- 1.Squad availability: 1 carrying a fitness doubt across the two squads, 1 of them projected starters. The forecast does not adjust for who is missing: its lineup channel currently contributes zero, so this is context the probabilities do not include.
United States
United States: 1 carrying a fitness doubt.
- DoubtChristian Pulisic, the first-choice forward, is recovering from Calf injury and is a fitness watch item; if unavailable the projected XI shifts.
Turkey
Turkey come in at close to full strength.
Availability runs in Turkey's favour here: United States are managing a fitness concern over Christian Pulisic, while Turkey's projected XI looks intact.
Availability from the predicted squads and injury feed; forecast adjustments from the model's own decomposition. See /docs/methodology/.
Standard Pass
This match is a free preview
You're seeing the model's full forecast for this fixture for free. Unlock the same depth: probabilities, expected goals, scoreline distributions, and per-player scoring, for all 104 matches with a Standard Pass, valid through the tournament.
Every forecast graded against the real result, scored on 987 matches since 2014. See the scorecard.
24h money-back, no questions asked·No subscription, no auto-renewal·Access through 31 Dec 2026. See refund policy.