Team Value · tables

Four tables. One league.

Every participant valued both sides of each match before it was played. Read those values forward and each one implies a league table — who it thinks would have beaten whom, with nothing about the results in it. Here is that table beside the real one.

Premier League, by Arena shots model

through round 9 · 60 matches priced · closest 25% of its signals called draws

#	Team	P	W	D	L	Pts	For	Against	Diff/game	Actual	Δ
1	Nottingham Forest	6	6	0	0	18	116.0	53.0	+10.50	17 (4 pts)	▲16
2	Man Utd	6	6	0	0	18	99.0	71.0	+4.67	19 (3 pts)	▲17
3	Brighton	6	6	0	0	18	93.0	76.0	+2.83	1 (15 pts)	▼2
4	Everton	6	5	1	0	16	78.0	91.0	−2.17	13 (8 pts)	▲9
5	Brentford	6	4	1	1	13	66.0	100.0	−5.67	9 (9 pts)	▲4
6	Fulham	6	3	2	1	11	79.0	88.0	−1.50	10 (9 pts)	▲4
7	Tottenham	6	3	1	2	10	111.0	60.0	+8.50	4 (11 pts)	▼3
8	Bournemouth	6	2	2	2	8	88.0	60.0	+4.67	7 (10 pts)	▼1
9	Man City	6	2	2	2	8	92.0	69.0	+3.83	3 (12 pts)	▼6
10	Arsenal	6	2	2	2	8	74.0	79.0	−0.83	2 (15 pts)	▼8
11	Leeds	6	2	1	3	7	72.0	71.0	+0.17	14 (6 pts)	▲3
12	Hull	6	2	1	3	7	71.0	86.0	−2.50	16 (4 pts)	▲4
13	Sunderland	6	1	3	2	6	76.0	92.0	−2.67	20 (1 pts)	▲7
14	Liverpool	6	0	5	1	5	90.0	71.0	+3.17	8 (10 pts)	▼6
15	Newcastle	6	1	2	3	5	81.0	94.0	−2.17	15 (6 pts)	—
16	Chelsea	6	1	1	4	4	78.0	69.0	+1.50	6 (10 pts)	▼10
17	Ipswich	6	1	0	5	3	66.0	90.0	−4.00	12 (8 pts)	▼5
18	Aston Villa	6	0	1	5	1	81.0	101.0	−3.33	5 (11 pts)	▼13
19	Coventry	6	0	1	5	1	70.0	110.0	−6.67	11 (9 pts)	▼8
20	Crystal Palace	6	0	0	6	0	60.0	110.0	−8.33	18 (4 pts)	▼2

Points here were never earned: each fixture is called from this participant’s standing signal before kick-off, and 3–1–0 accumulates. The table covers only the matches the board scores for it, and Actual is the real table over those same matches — so both columns describe the same football, and early in a season both are noisy.

How to read it

An opinion about the league, and nothing here is scored

These tables carry no agreement figure. The obvious one — how closely a table matches the standings — rewards reconstruction rather than forecasting: measured over two full seasons, a table built from goal difference alone loses a sixth of that agreement as soon as it is judged on matches it did not see, because its input and that target are the same goals. The comparison worth publishing is whose table calls the football that follows it, and it needs half a competition behind it. It opens here at each league’s halfway point.

Until then, Actual and Δ mark where this participant disagrees with what has happened — an argument worth having, not a rating. Forecasting is judged on the board, fixture by fixture, with an interval that says whether a difference is real.

Four attributed tables, never averaged into one: the moment the Arena blended them it would have stopped measuring and started competing with other people’s numbers. To see what any single row is built from, open a team and its matches from the team index.