Gameweek 4 was the column's best round of the season at 7-2 and +3.10pts, and the change that produced it stays: one selection per fixture, and only where the model rates it above the price on offer. The season stands at 20-16 and -2.93pts.
A pattern worth naming before the picks, because it is stark. Five of the ten selections are unders and three more are no-goals-at-both-ends. One is a result, and only Friday night's opener backs goals. That is not a house view, it is where the value sat when every fixture was priced against the model - and it lines up with what we found earlier this week in our test of which leagues are really scoring more: goals are up sharply in Serie A, the Eredivisie and the Championship, and the Premier League is the one division in the sample going the other way. Both teams to score is down from 56.1 percent last season to 52.5, and the over 2.5 rate from 55.0 to 52.5.
The biggest edge of the round is Brighton & Hove Albion against Arsenal at 6.1 points, and it rests on one number: Arsenal have conceded a single goal in four league games.
Fri 20:00
The round opens on Friday night and it is the one fixture where the model wants goals without hesitation. Brentford have had goals at both ends in all five of their last five, and the model makes over 2.5 goals 67 percent at 1/2 and both teams to score 68 percent at 2/5. Chelsea have scored ten in four league games, more than anyone outside the top two, and conceded nine, more than anyone outside the bottom three. Two sides who cannot stop trading.
Against the grain of the rest of this round, which is almost entirely unders - that is what makes it worth taking.
Brentford have seen goals at both ends in all five of their last five matches.
The bet: Over 2.5 goals at 1.50 - full match stats
Sat 12:30
Tottenham Hotspur have not scored in four league games. Aston Villa have scored once in four and have the lowest attacking output in the division at 0.78 expected goals a game. The model makes under 2.5 goals 50 percent here at evens, and it is hard to see where the goals come from: Spurs are creating 0.94 a game, Villa 0.78. Two of the three worst attacks in the league, in the same fixture, at half twelve on a Saturday.
Tottenham Hotspur have scored 0 goals in 4 league games from 3.7 expected. This is finishing, but it has lasted a month.
The bet: Under 2.5 goals at 2.00 - full match stats
Sat 15:00
Ipswich Town have been the most watchable side in the league for the wrong reasons - seven scored, ten conceded in four games - so an unders call here looks contrarian. It is not. Everton have conceded three in four and create 1.24 expected goals a game; the model has this at 48 percent under, 2.7 points clear of its price. The caveat is honest: Ipswich Town have seen goals at both ends in four of their last five, and if this one breaks open the under is dead early.
Ipswich Town have conceded 10 goals in 4 league games, the most in the division.
The bet: Under 2.5 goals at 2.10 - full match stats
Sat 15:00
The biggest edge of the round at 6.1 points. Arsenal have conceded one goal in four league games and 0.75 expected goals a game, the best defensive numbers in England by a distance. Brighton & Hove Albion have scored 13, so this is not a case of two poor attacks - it is a case of the best defence in the division travelling. The model makes under 2.5 goals 51 percent at 11/10 and no goals at both ends 50 percent at 21/20; both are rated well above their price.
Arsenal have conceded once in four league games. No side in the top five leagues has a better defensive record.
The bet: Under 2.5 goals at 2.10 - full match stats
Sat 15:00
The only result call of the round that the model rates above its price, at 62 percent. Hull City sit fourth on eight points but the underlying numbers say tenth: 0.91 expected goals created, 1.44 conceded, and a points-versus-expected-points gap of plus four - the largest overperformance in the division. Newcastle United have their own problems at the back, 2.05 expected goals against per game, but they have won three of five and Hull City's record is built on finishing that will not hold.
Hull City are fourth on 8 points against 4.0 expected points. That +4.0 gap is the biggest in the league.
The bet: Newcastle United to win at 1.60 - full match stats
Sat 17:30
Coventry City have not scored a league goal all season - four games, none - and have conceded ten. Nottingham Forest at 8/13 is barely above its price so the result is not the bet; the clean sheet is. The model makes no goals at both ends 50 percent at 19/20, which is the sensible way to back a side facing the only team in the division yet to find the net.
Coventry City have scored 0 goals in 4 league games and conceded 10. Both are division worsts.
The bet: Both teams to score - No at 1.95 - full match stats
Sun 14:00
The one clear goals call of the weekend. Crystal Palace have conceded 11 goals in four games from 2.46 expected goals against, comfortably the worst defensive numbers in the league, and they have still scored six. Leeds United have had goals at both ends in four of their last five. The model makes both teams to score 59 percent at 7/10, 4.1 points clear, and over 2.5 goals 58 percent.
Crystal Palace have conceded 2.46 expected goals a game, the worst defensive number in the Premier League.
The bet: Both teams to score at 1.70 - full match stats
Sun 14:00
The best price on the card that the model rates, 5.2 points clear at 8/5. AFC Bournemouth have conceded 1.15 expected goals a game, the second-best defensive number in the division behind Arsenal, and Liverpool have drawn three of their last five without finding a rhythm in front of goal. The referee is Michael Oliver, who lets more go than any official this weekend at 6.4 fouls per card - relevant only in that this is unlikely to become a stop-start, set-piece-heavy affair.
AFC Bournemouth concede 1.15 expected goals a game, second only to Arsenal in the Premier League.
The bet: Under 2.5 goals at 2.63 - full match stats
Sun 14:00
Manchester City have won five in a row and conceded two goals in four league games, at 0.68 expected goals against per match - the second-best defence in England. Sunderland have seen goals at both ends in one of their last five. The model makes no goals at both ends 54 percent at 4/5, and under 2.5 goals is 45 percent at 11/8 with a 5.3-point edge if you want the bigger price.
Manchester City have conceded 0.68 expected goals a game. Only Arsenal are better.
The bet: Both teams to score - No at 1.80 - full match stats
Sun 16:30
Manchester United carry the third-best expected goal difference in the league at plus 4.6 and sit thirteenth on four points, which is the gap between how they have played and what they have collected. Fulham are eighteenth on one point with 5.5 expected points, the unluckiest record in the division. Two sides underperforming their numbers meeting is usually a tight game, and the model has under 2.5 goals at 42 percent against an implied 40.
Fulham are 18th on 1 point against 5.5 expected points. That -4.5 gap is the worst in the league.
The bet: Under 2.5 goals at 2.38 - full match stats
One selection per Premier League fixture, settled after every round at 1pt level stakes. Gameweek 4 finished 7-2 and +3.10pts, the column's best round of the season, after Gameweek 3 lost 3-5 and -3.26pts. The season stands at 20-16 and -2.93pts. The change this round: bet only where the model rates a selection above its price.
Every fixture is run through the Statz projections model, which simulates the match from team strength, form and player-level data to produce win, goals and both-teams-to-score probabilities. The writing adds what the model cannot see: team news, motivation, and whether the market has over-reacted to last weekend.
Every gameweek, published ahead of the weekend once the Friday press conferences have landed and the predicted line-ups have firmed up. Prices quoted are the prices at the time of writing and can move before kickoff.
Yes - one selection per fixture, settled after the round and kept in a public running record at one point level stakes, wins and losses alike. Previous gameweeks stay on the page with their results shown.
Player props are markets on an individual player - shots, shots on target, tackles, fouls, cards. Each fixture note carries one prop the model likes, with the player's recent record behind it, taken from the same board that powers the Statz player props page.