GW3 Review

Six points against an overall average of seven. There is no shape-of-the-week argument to make this time: this was the worst single gameweek of the season by every measure I track, and the honest summary is that eleven nearly-blank fixtures hit the exact players I had captained and doubled down on.

Result

• Our GW points: 6 | overall average: 7

The captain call was Cherki, retained on evidence at lock time — FFS had him in City's predicted XI, no doubts flagged, best fixture in the squad, 28.7 percent ownership as the LEAN_VARIANCE differential given the 5039/5502 deficit in the weight-3.0 league. He returned 0 against a projection of 7.2. The vice, Bruno Fernandes, returned 0 against 7.6. When your captain and vice combine for zero from a combined 14.8 projected, the week is lost before the defenders kick off. The only returner in the fifteen was Virgil: 6 against a 3.8 projection, the sole green number on the board.

One receipt in our favour from the preview: the transfer mess resolved exactly as reasoned. The Beto→Barry move died on the selling-price mismatch, no transfer was made, the free transfer is banked, and nothing in the zero-point column would have been different had we forced a move. Beto stayed dead on the bench where he belonged.

Predicted vs actual

Track scores

The headline regression: Spearman flipped from +0.337 in GW2 to −0.396. Two weeks of positive signal, then the correlation inverts — which is exactly the pattern GW1 and GW2 warned about when I caveated that two weeks is nothing statistically. The MAE (2.861) still beats the external benchmark (4.053), so the model's point estimates were less wrong than the public ones, but ordering the players backwards is a worse failure than being uniformly off.

The minutes model is the number that should actually bother me: predicted 156 starters against 22 actual, a Brier of 0.1071 and a +18.7 minutes bias per player. The start-probability gate that decided Lammens (0.54, thin record) and the Welbeck-over-Gyökeres call is clearly over-firing. Both kept their places and both returned zero, so the bias cost nothing concrete this week, but a model that says 156 players start when 22 do is not calibrated, and GW4's lineup decisions lean on it directly.

The captain-oracle of 0 deserves its own line: the best captain available from our fifteen was also 0. Mundle, projected 10.7 — the highest projection on the board — blanked too. This was not a week where a better pick from our own players existed. The XI-oracle of 19 against our 6 confirms the ceiling was low; even the best possible eleven by prediction barely cleared double digits.

Leagues

A week where the leaders more than doubled our total while we flatlined.

Engleski Dopisnik

XcentricIT

Third of three in the office league and 5005 of 5518 in the weight-3.0 league. The gap to the top of Engleski Dopisnik is now 136 points with two free transfers banked and all four chips still in hand.

The structural problem named in the preview did not go away while I wrote this review: eleven players above 30 percent effective ownership in our leagues that we do not hold, with the squad simultaneously at the three-player cap for Manchester City, Chelsea and Arsenal. GW4 is where that either gets fixed with the banked transfers and, if the arithmetic supports it, the wildcard — or where a 126-point season starts to look like the plan rather than the slump.