By taking the average or mode, we obtain a single, more stable prediction that generally has lower variance and better generalization performance. Presenting a single prediction simplifies the interpretation and use of the model for most applications. It's the main reason why they provide only one output. And there's nothing wrong with evaluating each value produced by the decision tree in the random forests. Such curiosities can help in many cases such as Simpsons paradox.

I searched how to access the individual predictions from each tree in random forests in R Party and I got this, might help you a bit:

library(party)

library(partykit)

# Assuming you have your data 'data' and target 'target'

rf

Random Forest is an ensemble learning method that combines multiple decision trees to make predictions. Each tree in the Random Forest provides an output or prediction for a given input, and the final prediction is typically obtained by averaging or voting over the individual tree predictions. However, it is possible to extract and analyze all the outputs from all trees in a Random Forest. There is no theoretical reason why one couldn't do this, but it may not always be necessary or practical for several reasons:

While there are valid reasons to analyze individual tree outputs, Random Forests' strength lies in their collective decision-making process. The focus is on the overall performance and stability of the ensemble, rather than individual tree outputs. However, depending on the specific use case, there might be situations where analyzing individual tree outputs is beneficial, such as in understanding the uncertainty in predictions or performing model diagnostics. In such cases, techniques like permutation importance or partial dependence plots can be used to interpret the individual tree contributions.