Gameweek 1 · 2026/27

What the model picked, and what I entered instead

Fixed before the deadline of 21 August 2026, 17:30 UTC, from a snapshot taken at 09:45 the same morning. The snapshot’s hash is in the provenance strip below.


The disagreement, in one number

Model’s squad, objective

186.87

My squad, same objective

178.24

Difference

−8.63

That is the model’s own objective function, evaluated on both squads: for each of the five gameweeks in the horizon, the best legal eleven, plus the best starter’s score again for the captain, plus a small share of the bench, all discounted by 0.85 per gameweek away. Run on the model’s own fifteen it reproduces the value the solver itself recorded, 186.86809762841, to within 5 × 10−13.

My squad judgement calls cost 8.63 expected points against the model.

Why not just add up the fifteen players?

Summing all fifteen players’ five-gameweek expected points gives 247.21 against 272.89, a gap of −25.68 — nearly three times the real one. It charges full value for all fifteen when only eleven start, so it penalises a good bench. The optimiser never maximised that quantity.


The model’s fifteen

The strokes on the right are the nine additive components of the gameweek’s expected points, at the same scale for every player on this page.

Model squad — cost £100.0m, bank £0.0m. Roles shown are the model’s own gameweek 1 team sheet.

Player Price GW1 xP 5‑GW xP P(start) ep_next Owned 123456789
B.Fernandes fdr MID · MUN · starts, captain £12.0m 4.82 23.00 0.92 4.0 51.1%
Mbeumo fdr MID · MUN · starts, vice £8.0m 4.52 21.55 0.85 2.8 36.6%
Virgil DEF · LIV · starts £6.5m 3.92 21.86 0.98 3.1 18.6%
Szoboszlai MID · LIV · starts £7.0m 3.91 21.26 0.91 2.5 41.7%
Anderson MID · MCI · starts £6.5m 4.31 21.22 0.94 2.3 8.4%
Rice fdr MID · ARS · starts £7.5m 3.81 19.23 0.90 2.6 18.6%
Raya fdr GKP · ARS · bench 1 £6.0m 3.10 18.04 0.95 4.0 37.0%
Thiaw DEF · NEW · starts £5.0m 3.18 17.90 0.93 2.2 1.7%
Lacroix DEF · CHE · starts £6.0m 3.60 17.59 0.90 2.8 9.5%
Thiago FWD · BRE · starts £8.0m 3.77 17.14 0.73 2.5 17.4%
Lammens fdr GKP · MUN · starts £5.0m 3.76 16.62 0.97 2.6 17.2%
Guéhi DEF · MCI · starts £6.0m 3.33 16.96 0.88 2.8 18.0%
Evanilson FWD · BOU · bench 2 £6.0m 2.99 16.41 0.86 2.0 2.2%
Truffert DEF · BOU · bench 3 £5.5m 2.48 16.32 1.00 2.5 4.6%
Emegha no history FWD · CHE · bench 4 £5.0m 1.55 7.80 0.26 1.4 0.1%

Slot order, left to right — the slot is always drawn, so an empty slot means that component is exactly zero for that player:

  1. 1 appearance
  2. 2 goals
  3. 3 assists
  4. 4 clean sheet
  5. 5 def. contribution
  6. 6 saves
  7. 7 goals conceded
  8. 8 cards
  9. 9 shrink

 above the rule, solid — adds points.  below the rule, dotted — takes them away. Both halves are drawn at the same rate, so equal lengths mean equal magnitudes, and the faint horizontal rules are one expected point apart, so a stroke can be read off and not merely compared. The tallest on the page is Haaland’s goals term, 3.20 xP.


My fifteen, as a diff

Nine of the fifteen are the same player, so here is only what moved. Both goalkeepers survived; the argument about them is which one plays, and that is judgement call four.

Model squad against entered squad, grouped by whether we agreed. Figures are 5‑gameweek expected points.

 Model picked5‑GW xP I entered5‑GW xP
Goalkeepers
agreedRaya £6.0m18.04Raya £6.0m18.04
agreedLammens £5.0m16.62Lammens £5.0m16.62
Defenders
agreedVirgil £6.5m21.86Virgil £6.5m21.86
agreedThiaw £5.0m17.90Thiaw £5.0m17.90
agreedLacroix £6.0m17.59Lacroix £6.0m17.59
swappedGuéhi £6.0m16.96Mitchell £4.5m14.52
swappedTruffert £5.5m16.32Muharemović £5.0m6.96
Midfielders
agreedMbeumo £8.0m21.55Mbeumo £8.0m21.55
agreedSzoboszlai £7.0m21.26Szoboszlai £7.0m21.26
agreedAnderson £6.5m21.22Anderson £6.5m21.22
swappedB.Fernandes £12.0m23.00E.Le Fée £6.0m16.83
swappedRice £7.5m19.23Tzolis £6.5m8.79
Forwards
agreedThiago £8.0m17.14Thiago £8.0m17.14
swappedEvanilson £6.0m16.41Haaland £15.5m22.22
swappedEmegha £5.0m7.80Walle Egeli £4.5m4.72

The fifteen actually entered — cost £100.0m, £0.5m in the bank. Roles are mine, not the model’s.

Player Price GW1 xP 5‑GW xP P(start) ep_next Owned 123456789
Raya fdr GKP · ARS · starts £6.0m 3.10 18.04 0.95 4.0 37.0%
Lammens fdr GKP · MUN · bench, GK £5.0m 3.76 16.62 0.97 2.6 17.2%
Virgil DEF · LIV · starts £6.5m 3.92 21.86 0.98 3.1 18.6%
Lacroix DEF · CHE · starts £6.0m 3.60 17.59 0.90 2.8 9.5%
Mitchell DEF · CRY · bench 3 £4.5m 2.87 14.52 0.90 1.7 6.5%
Thiaw DEF · NEW · starts £5.0m 3.18 17.90 0.93 2.2 1.7%
Muharemović no history DEF · LEE · bench 2 £5.0m 1.43 6.96 0.43 2.2 3.3%
Mbeumo fdr MID · MUN · starts, vice £8.0m 4.52 21.55 0.85 2.8 36.6%
Tzolis no history fdr MID · ARS · starts, captain £6.5m 1.75 8.79 0.43 2.3 22.2%
Anderson MID · MCI · starts £6.5m 4.31 21.22 0.94 2.3 8.4%
E.Le Fée MID · SUN · starts £6.0m 3.87 16.83 0.89 2.1 10.9%
Szoboszlai MID · LIV · starts £7.0m 3.91 21.26 0.91 2.5 41.7%
Haaland FWD · MCI · starts £15.5m 4.71 22.22 0.79 4.0 69.3%
Thiago FWD · BRE · starts £8.0m 3.77 17.14 0.73 2.5 17.4%
Walle Egeli no history FWD · IPS · bench 4 £4.5m 0.99 4.72 0.15 1.0 2.2%

Where the gap opens up

The objective is a weighted sum over five gameweeks. Its weight falls by 0.85 each week out, so gameweek 1 counts for nearly twice as much as gameweek 5 — a squad that is wrong now is punished harder than a squad that will be wrong later. Both columns are scored the same way, re-picking the best legal eleven and the best captain each week, so this compares the two squads rather than my team-sheet decisions.

Best legal eleven per gameweek, before and after the decay weight.

GWModel, rawMine, raw WeightWeighted difference
1 49.08 47.98 1.0000 -1.10
2 53.97 49.34 0.8500 -3.94
3 49.81 47.71 0.7225 -1.52
4 50.10 48.41 0.6141 -1.04
5 48.20 46.21 0.5220 -1.04
Total251.16 239.64 -8.63

The seven judgement calls

Where I overruled the model, or fed it a number I chose myself. Listed separately so each can be graded on its own when the week is scored.

#The callWhat it costs if I am wrong
1Belgian → Premier League attack factor set to 0.58Tzolis becomes the squad’s best player at 23.64 xP. At a factor of 0.30 he falls to 17.1 and Rice is the better pick.
2Bruno Fernandes fitness factor set to 0.85He drops out, and the £12.0m he frees funds Haaland. At 0.95 he keeps his place and Haaland is unaffordable.
3Diop P(start) set to 0.85Corrects a stale Fulham bench role. He still did not make the squad, so this one costs nothing either way.
4Raya started over Lammens, against the modelThe model prefers Lammens on a fixture priced by the FDR fallback rather than by expected goals.
5Le Fée over AmpaduThe model prefers Ampadu by 1.22 xP. Le Fée’s 16 bonus points last season are invisible to a formula with no bonus term.
6Egeli over Thomas-AsanteBought £0.5m of headroom against price rises. The model prefers Thomas-Asante.
7Tzolis captained, decided on the dayThe same bet as call 1, doubled. If the 0.58 factor holds he is a 4.72 xP captain and the armband is worth 4.585; if it does not he is 1.75 and it is worth 2.874. Nothing in between. He is in every published predicted Arsenal XI, at home to a promoted side.

Outside the snapshot

George Hirst

Our first choice at £5.0m, and reported to be mid-medical at Stoke City on a £10m permanent deal. FPL still lists him as a clean, available Ipswich forward. Nothing in the snapshot, the archive or the Wikipedia collection would have carried it.

Outside the snapshot

Coventry’s striker

Coventry were reported to be close to a £17m signing, which would push both of their existing forwards down the order. Unconfirmed at the deadline and worth re-checking; it did not change this squad, but it is the kind of thing that silently invalidates a price-sorted shortlist.


The rest of the figures

Where this squad sits against the crowd and against FPL’s own projection, position by position; the 4 candidates the model cannot price because they have no Premier League history; and every number on this page unrounded, with all nine components for each of the 38 players named.

All figures for gameweek 1


Provenance

The hash below is of the data snapshot the solve consumed. The same snapshot and the same formula version return the same fifteen names.

Snapshot hash
b7f5cf2940cb5b25e30df1fec0216b7debb45f59cc9b68316c0656a7a1eebef4
Snapshot taken
2026-08-21T09:45:01Z
Fixtures hash
18745575078b4bbf1c3a1c85862945e8d5ff15825130c68b4d11124e2c7bbc54
Deadline
2026-08-21T17:30:00Z
Formula version
v1.0-phase0, rules 2026-27, fidelity A
Solver
highs — proven optimal, 13,986 columns × 8,491 rows over 561 candidates, solved in 1,867 ms
Objective
186.86809762841
What optimal means
Proven optimal means optimal given the formula — which is the thing actually on trial. The solver found the best squad the scoring function can describe; whether the scoring function describes football is a separate question, and the one the track record is for.
Team strength
chain[xgls>fdr] — of the 100 club-gameweeks in the horizon (20 clubs over 5 weeks), 82 were priced by expected goals and 18 by the difficulty-rating fallback. In gameweek 1 the fallback prices ARS, COV, HUL and MUN; COV and HUL fall back in every week of the horizon.
Coverage
455 of 561 selectable players matched to history (81.1%); 89.4% at clubs the archive knows
Overrides applied
departed 38, doubtful 21, out-indefinite 44, out-with-date 7, suspended 3