The record
How the model actually did.
Every gameweek, the projections are written down two hours before the deadline and left alone. After the final whistle they are scored against what really happened, including the weeks the model was wrong, which so far is most of them in some way.
Gameweek reviews
Nothing here is written after the fact. The projections in each review are the ones that were on the site before anyone kicked a ball. You can read them yourself on the scorecard page, exactly as they were stored.
Gameweek 8 scored: Arsenal carried the list to its best week
The locked twenty averaged 4.30 against 3.56 for likely starters, 0.74 points a player ahead and the best week so far. Five Arsenal picks got 38 between them, Gabriel 17. Four Chelsea picks got 3. Haaland got 8, level with the crowd.
Gameweek 8: do not chase the 5-0
The deadline is tonight at 18:30. Haaland is 0.92 clear as captain. Groß and De Cuyper are among the most bought players, and the model has them 29th and 96th at Anfield. Plus Bruno for the third week, and the flags.
Gameweek 7 scored: the captain finally landed, and the list lost anyway
The locked twenty averaged 3.40 against 3.66 for likely starters, the first week below the baseline. Haaland was top of the list and got 13, level with the crowd. Brighton came through; the four City picks behind Haaland got 2 between them.
Gameweek 6 scored: right about the flags, wrong about the captain again
The locked twenty averaged 3.65 against 3.35 for likely starters, the same small edge as Gameweek 4. The captain pick lost to Haaland again. Waiting on João Pedro was right; the cheap defenders were not.
Gameweek 6: the £4.0m defenders the model would start
Four defenders at £4.1m or less are in the model's top 25 this week, all above Gvardiol at Anfield and De Cuyper at Sunderland. Three of those numbers I would act on. The fourth is Mendy at 5.17, and the model has his minutes wrong.
Gameweek 6, midweek: what the break bought
No flag has moved since Monday. Three of the four most bought players are from Brighton, Arsenal have five flagged players at home to Leeds, and the sixth best projection of the week is in 3.7% of teams.
Gameweek 6: the captaincy is closer than the ownership says
Saka, Haaland and Bruno are separated by 0.16 projected points. Haaland is in 74% of teams, Bruno is being sold, and Everton at Hull is the game the model likes most. Plus the five flags that decide the week.
Five gameweeks in: who is running hot
Five gameweeks in, the points table and the chances table disagree a lot. Groß, João Pedro and Gvardiol are living well above what they have created. Thiago, Barry and Le Fée are living well below. Here is how much of that my model believes.
Three weeks off: what actually changes
Nothing has happened on the pitch, so the projections have barely moved. Prices have moved by a tenth at most. What has changed is a stack of 75% flags, and those are hiding more than they look like, including in my own model.
The fixture run from GW6 to GW10
Fulham have the kindest five gameweeks by any measure. Leeds are tied for the hardest. The model still projects four Leeds players above anyone at Fulham, and the reason is a rule most fixture tickers ignore.
Why one big score ruins a projection
A season total cannot tell 2, 23, 2 apart from 9, 9, 9. The averages are identical and the players are not. This is how the model counts evidence instead, and which players it currently trusts least.
The week we lost the record
Gameweek 5 was played and its data is intact. What is missing is what the model said beforehand, and building that now would be easy, which is exactly why it is not being done.
Gameweek 4 scored: the model beat the field and lost the captaincy
Twenty players written down before the deadline. They averaged 4.20 points against 3.84 for a likely starter. A real edge, and a small one. The captain pick was a different story.