Gameweek 9 scored: Spurs defenders did what Haaland did not

The locked file was written at 06:38 UTC on Saturday, 4.4 hours before the deadline. Morning deadline again, so the backstop job got there before the two-hour one, as it is supposed to.

All ten games are played and the bonus is in. FPL has still not marked the week as final, and the average manager score has already moved from 42 to 44 since Monday, so a point here or there could still change. Nothing below depends on it.

The list beat the baseline by the most so far

The twenty players averaged 4.80 points. The 175 players the model expects to play an hour or more averaged 3.39. That is 1.41 points a player above the baseline. The previous best was +0.74 last week, so this is almost double that again.

Against the 207 players who actually played 60 minutes, who averaged 3.72, it is still more than a point ahead. Same caveat as always: nobody knew on Saturday morning who would play 60.

The other number I have been watching is the gap between what the twenty were projected and what they got. It was 1.0, 1.2 and 1.45 in the bad weeks, then 0.31 last week. This week the twenty were projected 4.55 and got 4.80. That is the first time the list has scored more than the model said it would.

The captain: Haaland, 2

Top of the locked list was Haaland at 6.07, more than a point clear of Tavernier. City lost 1-0 at home to Brighton. He had 0.47 xG, no return, no bonus, and got 2. He was the most captained player in the game again, so it is a draw again: 4 against 4.

The rest of the captain list would have done better. Tavernier 3, Hill 6, Van Hecke 15 and Evanilson 9. I am not going to pretend the model saw Van Hecke as a captain. He was 4.71, the same as Hill. But he was on the list, at 10% owned, and he was the best player in the game this week.

Across the five scored weeks, before doubling, the top pick has 27 points and the crowd's captain has 38. The three weeks it picked Haaland are level. The gap is still all from the two weeks it picked someone else.

Every pick, scored

Nine of the twenty got six points or more. Eight got two or fewer. Last week it was seven and eleven. So it was not one huge haul carrying the average, it was a lot of sixes and nines. Ten of the twenty played for a team that kept a clean sheet, which is the main reason.

Only seven players in the game got 10 or more this week, and one of them was on the list. It was the biggest one, Van Hecke's 15. Last week it was three of fifteen.

What went right

Spurs at home to Palace. Three Spurs defenders on the list, all under £5.0m, projected 13.6 points between them. Spurs won 2-0 and they got 29. Van Hecke scored, kept a clean sheet and took all three bonus points. Robertson had 0.77 xA without getting the assist and still got 8 with two bonus. Van de Ven got the clean sheet for 6. Robertson was also this week's against-the-crowd pick, at 2% owned, instead of Calafiori at 48%. Calafiori got 6, Robertson got 8. That is two weeks in a row the against-the-crowd pick has won, after losing the two before.

Bournemouth at home to Leeds. Three Bournemouth picks, 18 points. Evanilson scored and took the bonus for 9. Hill kept a clean sheet with nine defensive actions for 6. Tavernier, second on the list, only got the clean sheet and finished on 3.

Everton at Newcastle. Barry scored the only goal with 1.29 xG and got 9. Tarkowski kept a clean sheet for 6. Barry came into the week with three goals from about five xG, and the model kept backing the xG. This week one went in.

Rogers. Scored in the 2-0 win over United, clean sheet, eight defensive actions, 9 points. Half the game owns him so this did not win anyone a mini-league, but the model had him right.

What went wrong

Haaland and City. Covered above. Cherki also came off after 35 minutes and got 1. City picks were projected 10.6 and got 3.

Hull at home to Ipswich. The model had three Hull players, the goalkeeper, a defender and a midfielder, projected 13.5 between them. It finished 2-2 and they got 7. Belloumi had an assist for 5. Tzolakis and Egan got 1 each, and Egan only lasted 45 minutes.

Both sides of the same game. This one is on me for not spotting it before the deadline. The list had three Bournemouth players and also two Leeds players, Muharemović and Trafford, from the same match. A Leeds defender and the Leeds goalkeeper are not a natural pairing with Bournemouth's striker. Bournemouth won 2-0. The Bournemouth side got 18, the Leeds side got 5. Muharemović made 12 defensive actions and still only got 3 after the two goals. The defensive contribution floor is real, but it does not cover conceding twice.

Mbeumo, Isak and Suzuki. Mbeumo got 1 with a yellow in the defeat at Chelsea. Isak got 2 when Liverpool lost 1-0 at home to Arsenal. Suzuki made four saves, got an own goal against him in the 1-1 with Fulham, and ended on 1.

The model's own team

Last week I set up a team in the official game called Deadline FPL. The wildcard planner on this site picked all fifteen, no edits from me, and the rule is that it plays the eleven with the highest projection and captains the top one.

First week: 43 points, against an average of 44. That is a captaincy on 2 doubled to 4, and 10 points on the bench, including Vuskovic's clean sheet. Van Hecke, the top scorer in the game, was on the list but not in the team. So the list had its best week and the team built from the same model was slightly below average. Those two things are measuring different things, and I want both on the record. I will post the team score after every week alongside this.

The record so far

Across the five scored weeks the list is ahead by about 0.51 points a player, four weeks out of five. "Likely starters" is still filtered on what the model expects now, not what it expected before the week, so the baseline is a bit soft. It is not soft enough to explain a point and a half.

The honest summary

Two good weeks in a row, and this one did not hang on a single fixture the way Arsenal carried Gameweek 8. Spurs, Bournemouth, Everton and Chelsea all came good. Hull and City did not. That spread is closer to what a model like this should look like when it works.

I said that if the gap between projected and actual was still above a point after Gameweek 10, the model needs fixing. It was 0.31 last week and the list beat its own projection this week. One more week to go before I call that settled.

The thing I will be looking at before Friday is the Leeds and Bournemouth problem. The model scores players one at a time, so nothing stops it backing both ends of the same game. For Gameweek 10 it has four Arsenal defenders in the top five, at home to Hull, which is the same kind of pile-in that went wrong with City in Gameweek 7 and right with Arsenal in Gameweek 8.

Gameweek 10's deadline is Friday 6 November at 18:30 UK time. The numbers lock before that and get scored the same way next week.