The model doesn't care about your club's name
Five matchweeks in, the Premier League table tells you Manchester City won all five and Arsenal are close behind. The xG model agrees with that — then it disagrees with almost everything else.
Brighton are the league's most dangerous attack by xG. 11.48 expected goals from five games — a full goal clear of City (10.50). But they've conceded 8.67, so their 3-1-1 record is fine… except they've scored 16, a full 4.52 goals more than their xG says they should have. That's the league's biggest overperformance. Enjoy the goals; expect regression.
Newcastle are the mirage of the season. 9 goals scored from 5.32 xG (+3.68), sitting 2-2-1 with a +0 actual goal difference despite a -3.73 xG difference. They've banked results their chances never promised.
And then there's Sunderland. 9.77 xG, 6 goals. The third-best attack by xG in the league, sitting on a 1-1-3 record and -3.77 goals vs xG. Nobody in the league is more "due" than the Black Cats. Coventry (1 goal from 4.87 xG, -3.87) are due too — but unlike Sunderland, Coventry create nothing (xG difference -3.76). Finishing regresses; chance creation doesn't.
The defensive story is simpler: Arsenal are a wall. 4.04 xGA in five games (0.54 defensive rating), while Crystal Palace have shipped 10.34 xG — the league's worst defensive xG, and it shows in their 1-1-3 record.
And Tottenham? Two goals in five games. That's not a team, that's a rumor.
Matchweek 6 — the model probabilities (Oct 10–12)
| Fixture | Home | Draw | Away |
|---|---|---|---|
| Arsenal vs Leeds United | 55% | 27% | 18% |
| Aston Villa vs Brentford | 13% | 19% | 68% |
| Chelsea vs Bournemouth | 37% | 26% | 37% |
| Ipswich Town vs Fulham | 45% | 22% | 33% |
| Sunderland vs Brighton | 40% | 19% | 40% |
| Man United vs Tottenham | 67% | 20% | 13% |
| Crystal Palace vs Nottingham Forest | 11% | 23% | 66% |
| Hull City vs Everton | 19% | 28% | 53% |
| Liverpool vs Man City | 41% | 24% | 35% |
| Coventry City vs Newcastle United | 39% | 29% | 32% |
The three calls that matter
1. Liverpool vs Man City is closer than the form says (41/24/35). City's perfect record is real — but the xG model rates this as a coin flip tilted to Anfield, not a City procession. Liverpool's defensive rating (0.82) plus City's away scoring baseline of 1.30 gives a 2.29 combined expected total. If the title is decided by margins, the model says this weekend adds almost nothing.
2. Brentford are the weekend's away-day special (68%). Unbeaten (2-3-0), third-best attack by xG (9.90), traveling to a Villa side creating 4.53 xG across five games. The model expects 2.10 Brentford goals. Nobody is talking about Brentford. The model doesn't care.
3. Sunderland vs Brighton is a pure 40/40. The league's luckiest finisher (Brighton, +4.52) against its unluckiest (Sunderland, -3.77). Expected scoreline implies roughly 2.28–2.29 xG each way. If regression is real, this is the weekend it shows up.
Honorable mention: Chelsea vs Bournemouth at 37/26/37 — the model can't split them, which is the model politely telling you Chelsea's +2.71 overperformance and Bournemouth's draw-heavy grind (3 draws in 5) have collided.
Where the numbers come from
Every figure above is computed from per-match xG data — 58.3M match events deep in our full database, distilled here from the first 50 played Premier League matches of 2026/27. No pundit vibes, no "eye test", no bookmaker prices baked in. When the model and the table agree, it's because the data said so.
Want the underlying data? The same xG, zone, and event feeds behind this model are available as BigQuery tables, API endpoints, and downloadable files. See plans and pricing →