Three different season engines leave Arsenal and Manchester City separated by only half a percentage point. That uncertainty is not a failure of the forecast; it is the most important thing the forecast knows.

A league table is the end of a season, not an honest description of one before it begins. Turning uncertain squads, new managers, injuries, European workload and 380 unplayed matches into a single expected-points column creates precision the evidence cannot support.

This analysis takes a different position. It builds a transparent player and club evidence layer, sends that evidence through three deliberately different season engines, and preserves the disagreement between them. Data science is most useful here when it measures several plausible routes through the season instead of pretending August already knows May.

Building the evidence layer

The forecast begins with people rather than club badges. The player score combines EA FC overall ratings, position-adjusted Fantasy Premier League prices and five-season FPL production. It is imperfect—game ratings and fantasy prices are opinions encoded as numbers—but it gives current squad quality a separate voice from last season's table.

Figure 1. The Premier League 100 by club, separating merit places from additions required by the three-player club floor.

The distribution is deliberately top-heavy. Arsenal, Manchester City and Liverpool place many players on merit because the source measures see more elite quality in those squads. That concentration is information: title contenders generally need both exceptional peaks and enough depth to survive rotation, injury and fixture congestion.

The amber club-floor additions are equally important because they prevent the model from mistaking incomplete coverage for an absence of talent. Every club contributes at least three players before remaining places are awarded league-wide. The safeguard does not make squads equal; it ensures that promoted and less fashionable clubs enter the evidence layer rather than disappearing from it.

Rank Player Club Pos EA FPL £m 5y pts/90 Score Raw rank Route
1 Gabriel dos Santos Magalhães Arsenal DEF 88 8.0 5.01 97.9 1 Merit
2 Bukayo Saka Arsenal MID 88 9.5 6.31 97.7 2 Merit
3 Erling Haaland Man City FWD 90 15.5 7.25 97.7 3 Merit
4 David Raya Martín Arsenal GKP 87 6.0 4.15 97.2 4 Merit
5 Cole Palmer Chelsea MID 87 9.5 6.40 97.0 5 Merit
6 Bruno Borges Fernandes Man Utd MID 87 12.0 5.47 96.0 6 Merit
7 Gianluigi Donnarumma Man City GKP 89 5.5 3.97 95.9 7 Merit
8 Alexander Isak Liverpool FWD 88 9.0 6.53 95.4 8 Merit
9 Virgil van Dijk Liverpool DEF 90 6.5 4.16 94.7 9 Merit
10 Phil Foden Man City MID 85 7.0 6.21 93.6 10 Merit
11 Florian Wirtz Liverpool MID 89 7.5 4.74 93.5 11 Merit
12 Bryan Mbeumo Man Utd MID 85 8.0 5.33 92.8 12 Merit
13 William Saliba Arsenal DEF 87 6.0 4.31 92.4 13 Merit
14 Martin Ødegaard Arsenal MID 87 6.5 5.09 91.8 14 Merit
15 James Maddison Spurs MID 84 6.5 5.74 90.9 15 Merit
16 Cody Gakpo Liverpool MID 84 7.0 5.16 90.2 16 Merit
17 Matheus Santos Carneiro da Cunha Man Utd MID 83 8.0 5.33 89.8 17 Merit
18 Declan Rice Arsenal MID 87 7.5 4.28 89.1 18 Merit
19 Jordan Pickford Everton GKP 84 5.5 3.75 88.7 19 Merit
20 Rodrigo 'Rodri' Hernandez Cascante Man City MID 90 6.5 4.15 88.6 20 Merit
21 Tijjani Reijnders Man City MID 86 6.0 5.10 88.6 21 Merit
22 Eberechi Eze Arsenal MID 83 6.5 5.42 88.0 22 Merit
23 Jeremie Frimpong Liverpool DEF 83 5.5 5.49 88.0 23 Merit
24 Omar Marmoush Man City FWD 84 7.0 6.30 88.0 24 Merit
25 Jurriën Timber Arsenal DEF 82 6.5 4.87 87.6 25 Merit
26 Ollie Watkins Aston Villa FWD 84 8.0 5.62 87.2 26 Merit
27 Joško Gvardiol Man City DEF 84 5.5 4.59 87.2 27 Merit
28 Rúben dos Santos Gato Alves Dias Man City DEF 86 5.5 3.99 86.7 28 Merit
29 Hugo Ekitiké Liverpool FWD 83 7.5 6.26 86.5 29 Merit
30 Mikel Merino Zazón Arsenal MID 83 6.0 5.53 86.3 30 Merit
31 Viktor Gyökeres Arsenal FWD 87 7.5 5.20 86.2 31 Merit
32 Dominik Szoboszlai Liverpool MID 83 7.0 4.63 86.1 32 Merit
33 Benjamin White Arsenal DEF 83 5.5 4.69 85.7 33 Merit
34 Rayan Cherki Man City MID 81 7.5 6.86 85.7 34 Merit
35 Morgan Gibbs-White Nott'm Forest MID 82 8.0 4.77 84.7 35 Merit
36 Piero Hincapié Arsenal DEF 83 5.5 4.38 84.5 36 Merit
37 Sávio Moreira de Oliveira Man City MID 82 6.5 5.18 84.4 37 Merit
38 Dejan Kulusevski Spurs MID 83 6.5 4.52 84.1 38 Merit
39 Morgan Rogers Chelsea MID 82 7.5 4.66 83.9 39 Merit
40 Enzo Fernández Chelsea MID 84 7.0 3.94 83.5 40 Merit
41 Emiliano Martínez Romero Aston Villa GKP 85 5.0 3.54 83.1 41 Merit
42 Alysson Edward Franco da Rocha dos Santos Aston Villa MID 89 5.0 5.74 82.7 42 Merit
43 Gabriel Martinelli Silva Arsenal MID 81 6.5 5.35 82.0 43 Merit
44 Marc Guéhi Man City DEF 82 6.0 3.75 81.5 44 Merit
45 Xavi Simons Spurs MID 84 6.0 4.12 81.1 45 Merit
46 Reece James Chelsea DEF 81 5.5 4.88 81.1 46 Merit
47 Jérémy Doku Man City MID 80 7.5 5.70 80.7 47 Merit
48 Ronald Araujo Liverpool DEF 83 5.5 — 80.5 48 Merit
49 Giorgi Mamardashvili Liverpool GKP 84 5.0 3.53 80.3 49 Merit
50 Jacob Murphy Newcastle MID 81 6.0 5.42 80.0 50 Merit
51 Youri Tielemans Man Utd MID 85 6.0 3.64 79.9 51 Merit
52 Marcus Rashford Man Utd MID 80 7.0 5.45 79.7 52 Merit
53 Nikola Milenković Nott'm Forest DEF 83 5.5 3.51 79.7 53 Merit
54 Ryan Gravenberch Liverpool MID 85 6.0 3.54 79.5 54 Merit
55 Antoine Semenyo Man City MID 80 8.5 5.11 79.4 55 Merit
56 Alexis Mac Allister Liverpool MID 87 5.5 3.97 79.2 56 Merit
57 Pedro Porro Sauceda Spurs DEF 82 5.5 3.78 78.6 57 Merit
58 Daniel Muñoz Mejía Crystal Palace DEF 81 5.5 4.34 78.5 58 Merit
59 Noni Madueke Arsenal MID 80 6.5 5.42 78.3 59 Merit
60 Federico Chiesa Liverpool MID 81 5.5 9.57 78.0 60 Merit
61 Kai Havertz Arsenal FWD 82 7.5 5.07 77.8 61 Merit
62 Matz Sels Nott'm Forest GKP 83 5.0 3.51 77.3 62 Merit
63 Jean-Philippe Mateta Crystal Palace FWD 82 6.5 5.25 77.1 63 Merit
64 Harvey Barnes Newcastle MID 80 6.0 5.59 76.4 64 Merit
65 Granit Xhaka Sunderland MID 85 5.5 3.84 76.4 65 Merit
66 Anthony Elanga Newcastle MID 81 6.0 4.59 76.3 66 Merit
67 Yoane Wissa Newcastle FWD 82 6.0 5.48 76.3 67 Merit
68 Nick Pope Newcastle GKP 81 5.0 3.78 76.1 68 Merit
69 Rayan Aït-Nouri Man City DEF 81 5.5 3.81 75.8 69 Merit
70 Murillo Costa dos Santos Nott'm Forest DEF 83 5.5 3.11 75.2 70 Merit
71 Chris Wood Nott'm Forest FWD 82 6.0 5.40 75.0 71 Merit
72 Dean Henderson Crystal Palace GKP 81 5.0 3.53 73.5 78 Merit
73 James Tarkowski Everton DEF 80 6.0 3.47 72.9 81 Merit
74 Nordi Mukiele Sunderland DEF 79 5.5 4.88 70.7 86 Merit
75 Anton Stach Leeds MID 79 6.0 5.20 69.0 91 Merit
76 Justin Kluivert Bournemouth MID 79 6.0 5.17 68.9 93 Merit
77 Caoimhín Kelleher Brentford GKP 79 5.0 3.89 67.3 99 Merit
78 Mikkel Damsgaard Brentford MID 80 5.5 4.39 66.9 101 Club floor
79 Pascal Groß Brighton MID 80 5.5 4.33 66.7 103 Club floor
80 Brennan Johnson Everton MID 79 6.0 4.51 65.6 107 Club floor
81 Alex Iwobi Fulham MID 80 5.5 3.98 64.5 112 Club floor
82 Francisco Evanilson de Lima Barbosa Bournemouth FWD 80 6.0 4.27 62.5 121 Club floor
83 Kevin Schade Brentford MID 78 6.0 4.92 62.5 122 Club floor
84 Daizen Maeda Ipswich Town MID 79 5.5 — 60.5 130 Club floor
85 Lucas Estella Perri Leeds GKP 81 4.5 2.69 60.4 131 Club floor
86 Rayan Vitor Simplício Rocha Bournemouth MID — 6.5 5.43 59.7 136 Club floor
87 Matt O'Riley Brighton MID 78 5.5 5.32 59.3 137 Club floor
88 Antonee Robinson Fulham DEF 82 4.5 3.09 58.7 140 Club floor
89 Carlos Baleba Brighton MID 81 5.0 2.87 57.6 146 Club floor
90 Bernd Leno Fulham GKP 80 4.5 3.33 57.0 151 Club floor
91 Harry Wilson Leeds MID 76 6.5 5.66 56.8 153 Club floor
92 Florentino Ibrain Morris Luís Ipswich Town MID 80 5.0 3.60 56.1 159 Club floor
93 Omar Alderete Sunderland DEF 78 5.0 4.02 54.2 170 Club floor
94 Hidemasa Morita Hull City MID 79 5.0 — 54.0 173 Club floor
95 Konstantinos Tzolakis Hull City GKP 79 4.5 — 53.9 175 Club floor
96 Gustavo Hamer Coventry City MID 77 5.5 3.34 44.3 226 Club floor
97 Abdul Fatawu Ipswich Town MID 76 5.5 4.06 42.6 235 Club floor
98 Taiwo Awoniyi Coventry City FWD 75 5.5 6.72 41.8 241 Club floor
99 Jack Butland Hull City GKP 75 4.5 3.53 33.8 289 Club floor
100 Matt Grimes Coventry City MID 74 5.0 — 30.1 315 Club floor

The list should be read as a model input, not an eternal verdict on the league's best hundred footballers. Its value is consistency and coverage. The raw rank remains visible so readers can see exactly where the club-floor rule changes the published list.

Its shape also matters more than a debate over one borderline selection. The table gives the simulation a consistent measure of elite concentration and squad depth while keeping the adjustments visible enough to challenge or replace later.

Separating recent history from current squad strength

Figure 2. Five-season history against current squad strength; bubble size shows top-100 representation and colour shows contextual uncertainty.

History and squad quality agree at the very top, but the distance from the diagonal is where the forecast becomes interesting. A club above its historical position may have recruited faster than results have caught up. A club below it may still carry a strong recent record while its current squad, manager or availability picture has weakened.

The five-season window also prevents one extraordinary or disastrous campaign from becoming the whole prior. Recent seasons receive more weight, but older evidence keeps the model from declaring a permanent new order after a single year. Current players can then move that baseline rather than merely repeat it.

Establishing a structural baseline

Figure 3. Model 1's complete structural forecast, including expected points, 10–90% intervals and title probability.

Model 1 is the base case: a mixture of weighted five-season history and current squad quality, translated into match strength and simulated over a full schedule. The mean points tell us where the structural evidence centres each club; the interval is the more honest output because it shows how many very different seasons remain compatible with the same inputs.

This model favours Arsenal and Manchester City because both the historical and player layers place them at the front. It is intentionally conservative. It assumes strength persists unless the simulation's season-level randomiser, K, or match outcomes provide enough evidence to move it.

Turning certainty into an explicit choice

Figure 4. Title probability under low, base and high K, showing how season-level uncertainty changes apparent confidence.

K is not a secret adjustment used to produce a preferred champion. It is a sensitivity control for everything a structural model cannot know in August: whether finishing runs hot or cold, whether a tactical change works immediately, and whether a run of close games breaks in one direction. Increasing K widens the range of possible seasons without rewriting the underlying evidence.

The chart shows why a title percentage should never appear without an uncertainty philosophy. Low K converts a small quality edge into strong confidence. Higher K allows more challengers and more reversals. The base setting is therefore a declared judgement about how volatile one Premier League season can be, not a discovered law of football.

Letting context and feedback disagree

Model 2 decomposes uncertainty into club-specific mechanisms: stadium strength, a zero-centred new-coach effect, transfer churn, promotion uncertainty, European workload and match-level availability. Its premise is that disruption is uneven. A new manager can improve one club and delay another; European competition creates fatigue in particular weeks rather than subtracting an arbitrary number of points in August.

Model 3 uses a regularised gradient-boosted classifier trained on five chronological seasons of pre-match Elo, rolling points, goal difference, venue form, rest and season progress. Elo and form update after each simulated fixture, while heavy-tailed regime shocks allow rare breakout or collapse seasons. Its 2025/26 holdout log loss of 1.058 beats the unconditional 1.082 baseline modestly—useful evidence, but nowhere near a licence to call the algorithm an oracle.

Figure 5. Each club's three expected-points estimates, connected by their full spread; the diamond is the weighted consensus.

The connected ranges reveal whether apparent certainty comes from shared evidence or shared assumptions. Arsenal and Manchester City remain strong in every engine, but several clubs—particularly those facing managerial change, squad turnover or promotion—move materially depending on whether structure, context or feedback receives the greater voice.

That variation is precisely why the models are kept separate. Averaging three near-identical formulas would only disguise one opinion as an ensemble. Here Model 1 can be wrong because history persisted less than expected, Model 2 because contextual effects were mis-signed, and Model 3 because learned relationships failed to transfer. Different failure modes make the comparison useful.

Combining the season distributions

Figure 6. Title probabilities from the structural, contextual and machine-learning engines alongside the weighted consensus.

The leading conclusion is not that Arsenal are certain champions. It is that Arsenal and Manchester City occupy effectively the same title tier: 37.2% and 36.7% in the weighted view, with Liverpool retaining a meaningful 17.9% route. A half-point difference is ranking convenience, not substantive separation.

Below the leaders, the probabilities communicate asymmetry better than expected rank alone. Some clubs have similar central points but different chances of reaching the top four or falling into relegation because their simulated distributions have different widths and tails. The distribution contains the forecast; the mean is only its centre.

Rank Club M1 pts M1 title M2 pts M2 title M3 pts M3 title Final pts Final title Top four Relegated
1 Arsenal 78.2 37.9% 78.1 36.0% 76.3 38.0% 77.7 37.2% 91.2% 0.0%
2 Man City 78.2 37.5% 78.6 39.0% 75.2 31.9% 77.6 36.7% 90.8% 0.0%
3 Liverpool 72.8 17.2% 73.9 19.3% 70.1 16.8% 72.6 17.9% 77.8% 0.0%
4 Man Utd 60.4 1.7% 61.1 1.6% 62.3 5.4% 61.2 2.6% 31.7% 0.7%
5 Chelsea 60.6 1.9% 60.5 1.5% 54.3 1.0% 59.0 1.5% 24.5% 1.6%
6 Aston Villa 58.5 1.2% 59.7 1.1% 57.5 2.6% 58.8 1.5% 22.8% 1.3%
7 Newcastle 58.2 1.2% 60.5 1.2% 54.2 0.8% 58.1 1.1% 21.1% 1.7%
8 Spurs 55.6 0.7% 54.6 0.2% 47.9 0.2% 53.3 0.4% 10.7% 5.1%
9 Nott'm Forest 51.0 0.2% 49.8 0.0% 52.4 0.8% 50.9 0.3% 6.6% 5.9%
10 Bournemouth 48.1 0.1% 46.4 0.0% 53.8 0.8% 48.8 0.2% 5.5% 9.3%
11 Brighton 47.5 0.1% 46.9 0.0% 51.1 0.4% 48.2 0.2% 4.0% 9.8%
12 Brentford 46.1 0.1% 46.8 0.0% 47.9 0.3% 46.9 0.1% 3.0% 12.1%
13 Crystal Palace 47.4 0.1% 45.2 0.0% 47.2 0.1% 46.5 0.1% 2.5% 13.2%
14 Everton 45.4 0.1% 44.1 0.0% 47.2 0.2% 45.3 0.1% 2.0% 15.0%
15 Fulham 44.9 0.0% 45.2 0.0% 45.9 0.1% 45.3 0.0% 1.9% 15.9%
16 Leeds 41.6 0.0% 44.1 0.0% 47.5 0.2% 44.1 0.1% 1.9% 19.4%
17 Sunderland 40.9 0.0% 40.2 0.0% 41.8 0.0% 40.9 0.0% 0.7% 29.6%
18 Ipswich Town 39.7 0.0% 39.3 0.0% 40.7 0.1% 39.8 0.0% 0.6% 34.4%
19 Coventry City 34.1 0.0% 35.7 0.0% 43.4 0.1% 37.1 0.0% 0.7% 47.2%
20 Hull City 29.6 0.0% 28.2 0.0% 32.4 0.0% 29.8 0.0% 0.0% 77.7%

The central ranks are close enough that a small change in form or availability can reorder several clubs without changing the broad structure of the forecast. Expected position should therefore be read as a midpoint, not a fixed destination.

The tail probabilities add the information the rank removes. They distinguish a stable mid-table profile from a similarly ranked club whose simulations include both European contention and a meaningful relegation risk.

Seeing the whole argument at readable scale

Figure 7. A large-format evidence board connecting player representation, squad and historical strength, structural uncertainty, model disagreement and the final title distribution.

The vertical board is designed to keep each chart legible at normal article width. Read from top to bottom, it shows the chain of inference rather than presenting a collage: who supplies the squad signal, how that signal compares with history, what the structural model does with it, where the engines disagree and how the final probabilities are formed.

The advantage of the combined view is diagnostic. If the conclusion feels too bullish for a club, the reader can identify the source of that confidence. Arsenal and City's advantage enters through both history and squad strength, survives the K sensitivity test and appears in all three engines. The model is confident that they belong to the leading tier, but not confident about which of them finishes first.

What I would predict

My narrow call is Arsenal, but the defensible prediction is a two-club title tier rather than a single ordained winner. Arsenal's weighted 37.2% is only 0.5 percentage points above Manchester City's 36.7%; that gap is far smaller than the uncertainty in squad availability, tactical adaptation or a few close matches. Liverpool are the clear third route and the club most capable of turning a two-team expectation into a three-team race.

The deeper finding is that the models agree more on hierarchy than on exact outcomes. They consistently identify the same leading group, yet differ on the size of the gaps and on several volatile clubs underneath it. That is where new coaches, European schedules and feedback from early results matter most. Tottenham, Chelsea, Manchester United and the promoted teams should be treated as distributions with unusually consequential tails, not as fixed positions.

This also explains why an apparently modest ensemble is preferable to a spectacularly certain algorithm. Football seasons contain regime changes that historical data cannot observe in advance. A model should reward strong evidence while leaving room for those breaks. The purpose of the three-engine design is not to eliminate judgement, but to put the judgement where readers can inspect it.

The best preseason forecast is not the one that produces the neatest table. It is the one that shows which assumptions create the table, how often other tables occur, and what new evidence would make us change our minds. On that standard, Arsenal are the smallest possible favourite, Manchester City are essentially level, Liverpool remain live, and the uncertainty is the result.

Methods and original sources

Sources: EA SPORTS FC 26, Fantasy Premier League, football-data.co.uk, Premier League manager guide, and 2026/27 qualification summary.