Two teams, seven points, different trajectories
Real Sociedad host Atletico Madrid at Anoeta on 13 September in La Liga's fifth round, and the table offers an immediate puzzle: both clubs have seven points, but they sit ninth and seventh respectively. Atletico have played one fewer match, which gives their identical haul a different weight. Real Sociedad's current line is 2W-1D-2L with six goals for and eight against; Atletico are 2W-1D-1L with seven scored and six conceded. At home, Real Sociedad have taken four points from two games (1W-1D, 2 goals for, 1 against). Away, Atletico have three points from two (1W-1L, 3 for, 4 against). The headline storyline is the gap between the model's preference and the visitors' recent results: Nutmegly's AI read leans away, even though Atletico's two most recent stored results are defeats.
What the form lines actually say
Real Sociedad's five most recent stored results, newest first, read: a 3-2 win at Elche, a 0-0 home draw with Celta Vigo, a 2-1 home win over Espanyol, a 1-4 loss at Real Madrid and a 0-1 loss at Real Betis. That is two wins and a draw from the last three stored matches. Atletico's five most recent stored results read: a 1-2 loss at Liverpool, a 0-3 loss at Athletic Club, a 3-1 win at Sevilla, a 2-2 home draw with Villarreal and a 2-0 home win over Malaga. Their last two stored results are defeats, but the three before that produced two wins and a draw. These stored results may span competitions and seasons, so they are a historical sample rather than a clean current-season form line.
The longer sample is where the model finds its away lean. Over eight stored matches each, Real Sociedad have averaged 1.12 points per game and Atletico 1.62. That is a substantial gap, and it is the most direct form-based argument for the visitors. But the visible five for Atletico end with two losses, and the visible five for Real Sociedad end with a win and a draw. Sample length changes the story, which is why the model's 41.0% away probability should be read as a lean rather than a conclusion.
The home and away numbers behind the read
The same-venue splits add another layer. In a five-match stored sample at the venue, Real Sociedad have averaged 1.40 goals per home game, while Atletico have conceded 2.40 per away game. That combination suggests a potential home threat, even though the aggregated signal is flagged as leaning away. The same-competition sample is more balanced: both teams average 1.40 points per La Liga game in the stored data, a signal flagged as draw-leaning. The rest windows are close, with the stored gap listed as five days for Real Sociedad and four for Atletico, a marginal home edge.
Nutmegly's AI read, generated on 10 September, gives Real Sociedad a 31.8% win probability, the draw 27.2% and Atletico 41.0%. The model also puts the over-2.5 line at 49.15%. So the away lean is real but not commanding: the combined home-win and draw probability is 58.96%, and the goals line is essentially a coin flip. The model's case rests on Atletico's better longer-sample points-per-game rate, their higher current rank and the fact that they have the same seven points from one fewer match. It is a reasonable read, but it is not an overwhelming one.
Counterevidence and the lineup caveat
The strongest supplied counterevidence is Atletico's two most recent stored results: a 1-2 loss at Liverpool and a 0-3 loss at Athletic Club. A two-match losing run sits directly against an away lean. The same-venue numbers also cut the other way: Atletico's 2.40 goals conceded per away game in the stored sample is a vulnerability that Real Sociedad's 1.40 home scoring rate could exploit. And the model's own caveats include a disagreement with an external forecast, which suggests the away read is not universally shared.
Lineups are not confirmed, and the underlying caveats flag that starting XIs and in-game shapes could still change the picture. No injury absences are reported in the available data, but the absence of reports is not proof of a fully healthy squad. The data quality notes also remind us that sample sizes can include older seasons, so the 1.62 and 1.12 points-per-game figures are not current-season totals, and the 1.40 and 2.40 venue averages come from a five-match stored sample rather than a complete season.
The watchpoint: Atletico's away defence
The match may turn on whether Atletico's away defensive record holds. In the stored same-venue sample, they have conceded 2.40 per away game, while Real Sociedad have scored 1.40 per home game. If that pattern repeats, the home side could find the openings that the model's away lean underplays. If instead Atletico reproduce the efficiency of their longer sample — 1.62 points per game across eight stored matches — the 41.0% away probability will look justified.
The specific question to watch: can Atletico's longer-sample control outweigh their two most recent defeats, or will Anoeta expose the away-day fragility that the same-venue numbers hint at? The answer will not be found in the pre-match probabilities alone, but in how the first half treats Atletico's defensive record on the road.