Wednesday 2 Dec, 19:30 · engine xGB Core v1 · doesn't know lineups or injuries

Chelsea

57Index63Attack50Defense
LWLWW
58%CHE WIN
2–1
Likeliest CHE win · 9%

Crystal Palace

34Index48Attack22Defense
LLLLW
CHELSEA 58%DRAW 21%CRYSTAL PALACE 21%
Over 2.5 goals
70%
Both teams score
67%
Expected goals
2.3 · 1.3
Why
Head to head · 24 league meetings since 2014/15
CHE 183 drawn3 CRY
2025/26CRY1–3CHE2026-01-25
2025/26CHE0–0CRY2025-08-17
2024/25CRY1–1CHE2025-01-04
2024/25CHE1–1CRY2024-09-01
2023/24CRY1–3CHE2024-02-12
Chelsea — projected scorers
Cole Palmer34%43% ast88'
Morgan Rogers31%25% ast89'
João Pedro18%6% ast33'
Estêvão11%8% ast23'
Maxence Lacroix7%9% ast72'
Pep Chavarría1%14% ast39'
Crystal Palace — projected scorers
Adam Wharton1%24% ast90'
Daichi Kamada11%17% ast89'
Tyrick Mitchell11%11% ast90'
Jean-Philippe Mateta19%1% ast20'
Anan Khalaili11%6% ast78'
Jørgen Strand Larsen8%2% ast47'

Projections assume recent minutes patterns — the model doesn’t know lineups.

What the grid implies
Goals
Over 1.5 goals88%
Over 2.5 goals70%
Over 3.5 goals48%
Both teams score67%
Double chance
CHE or draw79%
Draw or CRY42%
CHE or CRYno draw79%
Draw no bet
CHEstake returned on a draw73%
CRY27%
Clean sheets
CHE clean sheet26%
CRY clean sheet10%

Model probabilities, not odds and not advice. Our own leaderboard shows no model here beats the closing market once the bookmaker’s margin is paid.

Every scoreline, priced — tap a cell
CHE↓ CRY→012345
0
1
2
3
4
5
Darker green = likelier. Rows are CHE goals, columns CRY. Scores of six or more each way carry the remaining probability.

Price held steady since first assessment (2026-11-18) — reassessed after every played match, frozen at kickoff.

What this is built from
The engine weighs
  • Chance quality created and conceded, both sides, all season
  • Opponent strength and venue
  • Recent play, weighted toward now
It cannot see
  • Confirmed lineups
  • Injuries and suspensions
  • Cup rotation plans

Engine internals are proprietary. The complete performance record — including every miss — is public on the record.