The Table Is Lying: Six Weeks In, the xPoints Table Says Fulham Are Fine and Newcastle Are Doomed

The story in one line: The official table has Manchester City level with Arsenal on 15 points and Fulham bottom on two. Run every game back through a Poisson model on the underlying xG and City drop to third, Fulham climb seven places, and Newcastle sit dead last.

The true table (xPoints)

Exhibit 1

Manchester City have 15 points from 9.10 expected; Fulham have 2 from 7.70.

Expected points against points won, Premier League 2026/27, after six weeks

	Team	xPts	Actual	Gap
1	Arsenal	12.43	15	+2.57
2	Brighton	10.01	13	+2.99
3	Man City	9.10	15	+5.90
4	Brentford	9.02	10	+0.98
5	Man Utd	8.91	6	-2.91
6	Bournemouth	8.49	3	-5.49
7	Forest	8.39	5	-3.39
8	Sunderland	8.37	4	-4.37
9	Liverpool	8.26	9	+0.74
10	Leeds	7.97	9	+1.03
11	Ipswich	7.94	9	+1.06
12	Spurs	7.77	3	-4.77
13	Fulham	7.70	2	-5.70
14	Everton	7.22	9	+1.78
15	Chelsea	7.10	10	+2.90
16	Villa	6.91	5	-1.91
17	Hull	5.46	8	+2.54
18	Palace	5.04	4	-1.04
19	Coventry	4.64	3	-1.64
20	Newcastle	4.54	8	+3.46

Method: independent Poisson distributions on each side's Opta xG, scorelines 0-10, xPts = 3 x win probability + draw probability. The model assumes finishing is luck; over a season that is one of the safest assumptions in football.

City's fifteen points are a mirage

Manchester City have the largest gap in the league: 15 points banked from 9.10 xPts, an overperformance of 5.90. The wins tell the story. On Sep 13 they beat Manchester United 1-0 from an xG of 1.00 to 1.10 (worth 1.42 xPts). On Sep 20 they beat Sunderland 5-3 while being out-created 2.82 to 3.57 (worth 1.08 xPts). On Sep 5 they edged Coventry 1-0 from 2.12 to 1.37 (worth 1.85 xPts). That is nine points from 4.35 xPts in three games. City are a good team, but the table's co-leaders are the league's luckiest side.

Fulham are not a bottom team

The model has Fulham thirteenth on 7.70 xPts, playing like a mid-table side, with two points to show for it. They lost 3-2 to Crystal Palace from an xG of 3.14 to 1.97 (2.00 xPts, zero points). They lost 1-0 to Sunderland from 2.02 to 0.95 (0.71 xPts, zero points). They lost 3-2 to Chelsea from a defensible 1.33 to 2.24. Fulham are doing the things that create wins and getting the results that create crises. If finishing variance is luck, and it mostly is, the table will come for Fulham before Fulham come for the table.

Bournemouth and Spurs are owed points; Newcastle are spending them

Bournemouth (-5.49) drew 2-2 with Brentford while out-creating them 1.75 to 0.50, and 1-1 with Everton from 2.16 to 1.84. Tottenham have three goals from 6.6 xG in six games, the kind of finishing drought that reads like a curse but regresses like a mean. Sunderland sit on four points from 8.37 xPts.

And then there is Newcastle. Eight points from 4.54 xPts, the worst chance profile in the league, a +3.46 gap that leads the overperformers. Their 2-2 draw with Liverpool on Aug 23 came from an xG of 1.58 to 2.98 (the least likely result of the season so far, 15.9%), and they have not created more than 2.0 xG in a single game. The market will keep pricing them like a top-half team. The model says: fade.

The surprise at the top

Arsenal (12.43 xPts) and Brighton (10.01) are legitimate. The model's best team that the table doesn't know is Manchester United: fifth on 8.91 xPts, fourteenth on six points. The worst team the table doesn't know is Newcastle. Finishing extremes do not persist; the edges above are exactly the kind that get priced in by November. Check back then; we'll re-run the table and see who listened.

---

Where the numbers come from

Every figure is computed from per-match Opta expected goals and results for 56 played Premier League 2026/27 matches (Aug 21 to Oct 10, 2026), queried live from the Dribble API on Oct 14, 2026. Expected points come from independent Poisson distributions on each side's xG. No pundit vibes, no eye test.

Want the underlying data? The same xG feeds behind this model are available as BigQuery tables, API endpoints, and downloadable files. See plans and pricing