The primary evaluation metric for this competition is the **Root Mean Squared Logarithmic Error (RMSLE)**.

**Root Mean Squared Logarithmic Error (RMSLE)** measures the square root of the average of the squared differences between the logarithm of the predicted values and the logarithm of the actual values. It's particularly useful when you want to penalize under-prediction more than over-prediction, and when the target variable has a wide range or skewed distribution, as it treats errors in the same proportion regardless of the magnitude of the actual values.

The formula for Root Mean Squared Logarithmic Error is:

$$\text{RMSLE} = \sqrt{\frac{1}{n} \sum_{i=1}^{n} (\log(Y_i + 1) - \log(\hat{Y}_i + 1))^2}$$

Where:
* n is the number of data points
* Y_i is the actual (true) value for the i-th data point
* $$\hat{Y}_i$$ is the predicted value for the i-th data point
* $$\log$$ is the natural logarithm. The '+1' is added to both the true and predicted values to handle cases where Y_i or $$\hat{Y}_i$$ might be zero, preventing $$\log(0)$$ which is undefined.

**Your goal is to minimize the RMSLE.** A lower RMSLE indicates a better fit of your model, especially reflecting proportional errors.

Your submitted predictions will be scored against the true `Policy Cost` values in the hidden test set. As per the rules, your public leaderboard score will be based on 50% of the hidden test data, providing immediate feedback. However, the **final ranking will be determined by your model's performance on the remaining 50% of the hidden test data (the private leaderboard)**, which remains unseen until the competition concludes. This ensures that models generalize well to completely new data.