## Alfred Korzybski's Insight
"The map is not the territory" — the representation of reality is not reality itself. In betting, your model is the map; the actual match is the territory. The distinction is essential.
## How the Model Fails to Capture the Territory
**1. Model inputs are imperfect:**
xG models are built on historical data. They capture team quality as it was — not as it is after key personnel changes, tactical evolution, or injury absences.
**2. Model structure is simplified:**
A Poisson model assumes each goal is independent. In reality, goals affect match dynamics (a goal changes how both teams play). The model is a simplified map, not the territory.
**3. Unmeasured factors exist:**
Referee personality, weather not captured in data, player personal circumstances, locker room dynamics — these are real factors that the map does not represent.
## The Practical Implication
Model output is your best available estimate — not the truth. Hold it confidently enough to act (bet when the model says there is value) but loosely enough to update (revise when good information not in the model becomes available).
## Over-Fitting: When the Map Is Too Detailed
A model that is too closely fitted to historical data produces a map that is very accurate for the past but less accurate for the future. The model has learned the noise as well as the signal.
Test for overfitting: split historical data into training (80%) and test (20%) sets. If model performance on training data is significantly better than on test data: overfitting. Simplify the model.
## The Epistemic Humility Position
The expert bettor operates with explicit epistemic humility: "My model is a good map. It is not the territory. I will act on it confidently while acknowledging that it is an approximation of an uncertain reality."
Create a free account to track your progress and save bookmarks.
