The storyline arrives in two colours
The Manchester derby at Old Trafford on 13 September finds City atop the Premier League after three matches: nine points, three wins, seven goals scored and two conceded. United sit 11th on four points from one win, one draw and one defeat, with seven scored and six conceded. That gap is the headline tension. City's start is immaculate; United's is jagged. Yet the match is at Old Trafford, where United have won their only listed home league fixture 5-2. Nutmegly's AI read still leans away, giving City 43.9%, United 31.1% and the draw 25.0%. The question is whether United's home scoring punch can make that lean look too cautious.
What the stored form actually shows
United's stored run is a 4-0 win over Sabah FA, a 2-2 draw at Everton, a 5-2 home win over Ipswich, a 0-2 defeat at Hull City and an older 3-0 win at Brighton on 24 May. That is not a smooth current-season arc; the dates span the available sample, so it should be read as mixed form rather than a clean sequence. City's stored run is more coherent: 2-0 at FC Porto, 1-0 at home to Coventry, 4-1 at Crystal Palace, 2-1 at home to Bournemouth, then an older 1-2 home defeat to Aston Villa on 24 May. Four wins in the stored run, and a recent-points-per-match signal of 2.38 against United's 2.12, explain why the model's default is away.
The case for the away lean
City's table line is the strongest evidence: three wins from three, top of the pile, seven goals for and only two against. Their one listed away fixture produced four goals and one conceded, so the away sample, while tiny, points the same way. United's away record is the weak spot: no win, one draw, one defeat, two scored and four conceded. The AI read also gives over 2.5 goals a 64.9% probability, which fits two teams with recent scorelines like 5-2, 2-2, 4-1 and 2-1 in the stored record. If the match becomes open, City's finishing and United's defensive numbers are a plausible combination for the away side to exploit.
Where the read gets uncomfortable
The counterevidence is real. A separate Premier League-only sample in the data has both clubs at 2.00 points per match, which undercuts the idea that City's early table lead is a chasm. The three-game table is a thin basis for a firm conclusion, and United's home signal is awkward for the model: the data's home-away matchup gives United 2.80 home goals per match and City 1.00 away goals conceded per match. The draw sits at 25.0% in the model, not a negligible tail, and home win at 31.1% is close enough that one Old Trafford goal could flip the mood. An external forecast in the data also points a different way from Nutmegly's AI read, which is a warning against treating any model lean as consensus.
There is also an availability caveat. No lineups are confirmed, and the supplied injury feed reports no absences; that is not proof of a fully fit squad, only an information gap. The model's own counterpoints note that both teams have scoring routes and that a single defensive error can amplify the result. The rest-window signal in the data points home, but the same notes caution that gaps in records should not be treated as guaranteed rest or advantage. A derby with unsettled teamsheets can move away from a pre-match probability quickly.
The watchpoint: first goal, home punch
The cleanest way to test the away lean is to watch the first goal and whether United's home scoring signal survives contact with City's away defensive record. If United score first at Old Trafford, their 2.80 home-goals-per-match signal and City's 1.00 away-goals-conceded signal suggest the 43.9% away probability could come under pressure. If City score first, their three-win start, seven-goal output and 2.38 recent points-per-match signal would align with the model's lean. The match might not need a dramatic tactical story; the numbers already frame a simple conditional: can United's home punch turn into points, or does City's efficiency make the away read look justified?