Gameweek 8 scored: Arsenal carried the list to its best week

The locked file was written at 06:38 UTC on Friday, 10.9 hours before the deadline. That is the longest margin so far: the deadline moved up to Friday evening for Ipswich against Forest, and the morning job caught it first. A longer margin is a harder test, because less team news was known.

All ten games are finished and the bonus points are in. FPL had not marked the week as final when I pulled the numbers on Monday, so a point could still move. Nothing below depends on one point.

The list beat the baseline by the most so far

The twenty players averaged 4.30 points. The 189 players the model expects to play an hour or more averaged 3.56. That is 0.74 points a player above the baseline, double the previous best (+0.36 in Gameweek 4).

Against the 210 players who actually played 60 minutes, who averaged 4.01, it is still ahead. That has not happened before. Same caveat as always: nobody knew on Friday who would play 60.

The number I said to watch last week went the right way. The twenty were projected 4.61 and got 4.30, a gap of 0.31 points a player. It had been 1.0, 1.2 and 1.45. One week does not undo three, but it is the first time the top of the list has scored roughly what the model said it would.

The captain: a draw again

Top of the locked list was Haaland at 5.90, 0.92 clear of Thomas. City lost 2-1 at Villa. He scored City's goal, took two bonus points and got 8. He was also the most captained player in the game, so it is 16 against 16.

The rest of the captain list did not help: Thomas 1, João Pedro 0, Rogers 1 and Mbeumo 2. Across the four scored weeks, before doubling, the top pick has 25 points and the crowd's captain has 36. The gap is all from the two weeks the model picked someone else.

Every pick, scored

Seven of the twenty got six points or more, the most since Gameweek 4. Eleven got two or fewer, so this was not a week where everything worked. It was a week where the hits were big.

Gabriel's 17 was the top score in the game. Fifteen players got 10 or more this week, and three of them were on the list: Gabriel 17, White 10 and Tavernier 10. Isak and Muharemović were one short on 9. Last week it was one of five.

What went right

Arsenal at home to Everton. The model had five Arsenal players, the same kind of pile-in that went wrong with City last week. This time the fixture came good. Arsenal won 2-0, and the five got 38 points. Gabriel scored, kept a clean sheet, made 14 defensive actions and took all three bonus points. White had an assist and a clean sheet for 10. Raya kept a clean sheet for 7. The one miss was Calafiori, who went off after 22 minutes and is now listed with a hip injury.

Last week I said that when the model likes a fixture this much, one bad afternoon takes a quarter of the list with it. That works both ways. It is the same risk, and this week it paid.

Bournemouth at Old Trafford. Tavernier was the most bought player of the week, and the model had him 10th. Bournemouth won 3-0, he scored, kept a clean sheet and got 10.

Muharemović, defensive contributions and a clean sheet. Leeds drew 0-0 at Sunderland. He made 13 defensive actions and got 9 without touching the ball in either box.

Isak. Scored in the 2-0 win over Brighton, took the bonus and got 9. He was in 13% of teams.

What went wrong

Chelsea at home to Spurs. The preview called this "the rare week where the crowd and the model agree". Both were wrong. Spurs won 2-0. The four Chelsea picks were projected 18.6 points between them and got 3. João Pedro did not play and is now listed with an unspecified knock. Palmer was sent off and finished on −1. Rogers played 90 and lost a point to a yellow card. Neto got 3, all of it from 12 defensive actions.

Thomas. Second on the whole list at £4.0m. He went off after 59 minutes, one short of the hour, in a 1-0 home defeat to Fulham, and got 1. He is now flagged at 75%. I wrote that I had been wrong about cheap defenders before. Make that twice.

Manchester United. Mbeumo played 90 in the 3-0 home defeat and got 2. Bruno, who I stopped arguing for this week, did not play at all.

Forest at Ipswich. Murillo and Gibbs-White got 3 between them in a 1-1 draw. Gibbs-White had 0.76 xG and nothing to show for it.

Brighton, as promised

The preview said not to buy Brighton for this week. Groß and De Cuyper had about 460,000 transfers in between them after the 5-0. At Anfield, Brighton lost 2-0. Groß scored an own goal and got 0. De Cuyper got 1.

That also settles the against-the-crowd test for once. The locked file left out Groß, in 46% of teams, and backed Murillo, in 2%. Murillo got 1, Groß got 0. It is the first week the against-the-crowd pick has won, by one point. After losing by 12 over the previous two, I am not going to make much of it.

The record so far

Across the four scored weeks the list is ahead by about 0.29 points a player, three weeks out of four. That is back to roughly where it was before Gameweek 7. "Likely starters" is still filtered on what the model expects now, not what it expected before the week.

The honest summary

A good week, mostly because one fixture went right instead of wrong. The list put five picks into Arsenal and four into Chelsea. If those two results had gone the other way round, this would be another losing week. That is the thing to take from Gameweeks 7 and 8 together: the model piles into the fixtures it likes, and the weekly result is decided by whether those one or two games go to plan.

The gap between projected and actual dropped to 0.31. I said that if it was still above a point after Gameweek 10, the model needs fixing. Two weeks to go, and this week it is well under.

Gameweek 9's deadline is Saturday 31 October at 11:00 UK time, now that the clocks have gone back. The numbers lock before that, and get scored the same way next week.