The headline: a 0.2-point lean at a ground that should matter
Union Berlin sit 17th with six points from seven games; 1. FC Köln sit 12th with eight. That two-point gap is the table's obvious frame, but the more interesting tension is at An der Alten Försterei on Sunday: Nutmegly's AI read leans away, giving Köln a 36.6% chance to Union's 36.4%, with the draw at 27.0%. That is a plurality of 0.2 percentage points, not a majority. The model is not saying Köln are clear favorites; it is saying the three-way distribution is almost level. The question is whether that tiny away lean survives the venue split.
The venue split is the first place to test it. Union's current home record is 1W2D1L, with five goals scored and six conceded in four games. Köln's current away record is 0W0D3L, with two goals scored and seven conceded in three. Union's overall goal difference is -12, but their home goal difference is -1; Köln's overall goal difference is -3, but their away goal difference is -5. The table says Köln are better so far; the splits say this match is not played on neutral ground. That is why the away lean feels fragile rather than forceful.
Form: Union's newer scorelines are tighter; Köln's points are at home
Union's stored recent run is partial and may span seasons, so it is a sample rather than a current-season form table. The newest three entries are a 1-1 draw at Augsburg, a 0-0 home draw with Borussia Dortmund and a 1-0 home win over SV Elversberg: two goals scored, one conceded. The two older entries are a 0-7 defeat at Bayern München and a 1-3 home loss to Schalke 04. Those heavy defeats explain much of the overall goal difference, but they are older entries in the sample. The newer scorelines point to a tighter, lower-scoring profile, without proving a tactical shift.
Köln's stored run tells a more divided story. The newest five are a 0-0 home draw with Schalke, a 0-1 away loss at Eintracht Frankfurt, a 2-1 home win over Borussia Mönchengladbach, a 1-2 away loss at Hamburger SV and a 1-1 home draw with Werder Bremen. The away entries are defeats; the home entries are unbeaten. That matches the current table: Köln's eight points are built on a home split of 2W2D0L, while their away split is 0W0D3L. If the model is leaning away, it is not leaning on Köln's away performances in the supplied evidence.
The AI read: a plurality, not a conviction
Nutmegly's AI read is still worth stating plainly: away lean, 36.6% Köln, 36.4% Union, 27.0% draw, with over-2.5 goals at 47.63%. The model's own counterpoints are unusually candid. They note that the three-way probabilities are not clearly separated and that the draw band should be retained. They note that an external forecast disagrees with the model's direction and that the two samples may not be measured the same way. They note that lineups are not confirmed and that 11 potential absentees are reported across the match. Those are not decorations; they are reasons to treat the away lean as a soft estimate.
The strongest supplied case for the away lean is the table. Köln have two more points, two wins to Union's one, and a goal difference of -3 against Union's -12. They have also conceded 11 goals to Union's 18. If the model is weighting overall season performance, that is the evidence it would use. But the same table shows where those points came from. Köln's two wins are home wins; their away split has no wins and no draws. The case for the away lean is therefore real but narrow, and it runs into the venue evidence rather than through it.
The countercase: home split and the draw band
The strongest countercase is the home/away split plus the draw band. Union have lost once at home; Köln have lost all three away. Union have scored five at home; Köln have scored two away. If the match is decided by venue-specific performance, the home side has the clearer route to the points. The draw at 27.0% is not negligible, and the model's over-2.5 probability of 47.63% means the under is marginally more likely than the over. That fits a game where both attacks have been modest: Union have six goals in seven games, Köln eight in seven. A tight, low-event match is a plausible path, not a contradiction of the model.
The counterevidence to a home win is just as important. Union's home scoring is five goals in four games, which is serviceable but not prolific. Köln's away concession is seven in three, which is poor, but Union still have to convert that into goals. Union's older 0-7 and 1-3 defeats are a reminder that their defensive numbers are not uniformly tight, even if the newer scorelines are better. And with 11 potential absentees and no confirmed lineups, the match could tilt on selections the data cannot show us. That does not tell us who will miss out or how either side will line up; it does mean the single-match variance is higher.
What to watch: the first goal and the half-time score
The watchpoint is specific and conditional. Watch whether Union can turn home territory into the first goal. If they score first, the away lean looks weakest, because Köln's current away split shows no points and two goals in three games. If the game reaches half-time level, the 27.0% draw band and the sub-50% over-2.5 read point to a tight second half. If Köln score first, the model's away lean gets its best support, but the supplied data does not tell us how either side would manage the game from there.
The question to watch is whether Union can turn their home split into the goal that makes the away lean irrelevant, or whether the match drifts into the draw band the model refuses to dismiss.