Ask any model that covers both league and domestic cup football in the same country, and you will get the same answer. Cup matches are not league matches. The patterns that work on a Saturday afternoon stop working on a Tuesday night in January.
Goalsforge covers four domestic cups: the FA Cup, Copa del Rey, DFB-Pokal, and Coupe de France, alongside the league football in those countries. Treating them as a continuation of the league is the most common mistake a model can make.
The premise most models start from
Most prediction models assume league form carries over into cup matches. If a side is on a four-match winning run in the league, the model adds positive weight to their cup forecast.
That assumption is half right. Form does carry over. But cup matches introduce three dynamics that change the math entirely.
Squad rotation. Top-flight sides rest players in early cup rounds against lower-league opposition. The XI that played on Saturday is not the XI that plays on Tuesday.
Single-match stakes. There are no second chances in the FA Cup or DFB-Pokal after extra time. A league side that can absorb a loss in one of thirty-eight matches has zero margin for error in a one-off cup tie.
Lower-league opposition. The gap between a top-flight side and a fourth-tier side looks enormous on paper. In a one-off match at the lower side’s ground, the gap closes. Cup football is where the data’s priors get tested hardest.
What actually happens in the four domestic cups
Each of the four cups Goalsforge covers has its own character. They look similar on paper (knockout, single match, league sides entering in later rounds) but the dynamics differ.
FA Cup. The oldest cup competition in the world. The draw is open, the lower-league sides host the top-flight sides until the later rounds, and the gap between league football and cup football is widest here. Upset rate in the third and fourth rounds is the highest of any competition Goalsforge covers. A Championship side at home to a Premier League side is genuinely competitive.
Copa del Rey. Two-legged ties in the early rounds, single match from the round of 16. The format change at the round of 16 is the moment the model has to switch mode. Two-legged ties reward consistency. Single matches reward peak form. A side that wins 2-1 at home in the first leg can lose 3-0 away in the second and still go through on aggregate. That is a fundamentally different prediction problem.
DFB-Pokal. Single match, top of the table hosts lower sides. Bundesliga sides almost always play their strongest XI in the early rounds. The cup matters culturally in Germany more than in any other major football country. Upset rate is lower than the FA Cup but higher than La Liga or Serie A cups. Third-division sides reach the final regularly enough that the model has to take them seriously.
Coupe de France. The most chaotic of the four. Lower-league sides regularly beat Ligue 1 sides, sometimes several rounds apart. PSG’s first-team squad rarely plays in the early rounds. The competition’s prestige is lower than the league, but its upset rate is the highest of the four domestic cups Goalsforge covers.
What this means for predictions
The model treats cup matches differently in three concrete ways.
Form weight is reduced. A side’s recent league form still matters, but less than for a league match. The XI is uncertain, the opposition is structurally different, and the stakes change the way the match is played.
Starting-XI uncertainty is treated as a feature, not noise. If the manager has hinted at rotation, the model adjusts. If the manager has hinted they will play the cup seriously, the model adjusts the other way.
Home advantage is amplified. Lower-league sides hosting top-flight sides in cup matches get a venue boost larger than the league equivalent. The crowd, the unfamiliar surroundings for the visiting side, and the cultural weight of the occasion all push in the same direction.
The honest limit
Cup football’s upset rate is high because the underlying processes are noisy, not because the model is bad at it. A model that says 60% on an FA Cup third-round tie is doing the right thing. The match is genuinely a 60-40, with a wide range of possible scorelines.
Following cup predictions means accepting that “60%” in an FA Cup third-round tie covers more outcomes than “60%” in a Premier League match the same weekend. The probabilities are calibrated. The range underneath them is wider.
What makes cup football worth following
Cup football is where the model gets tested most. If a model can call league football accurately and cup football reasonably, it is probably well-calibrated. If it can call both, it is probably honest about uncertainty.
Following Goalsforge across both formats in the same country is the best test of whether the model’s confidence figures mean what they say. A 70% league call and a 70% cup call are not the same kind of 70%.
That is the part that makes cup football interesting to follow. The numbers are smaller. The range is wider. The surprises are real.