Great progress on the AI Beta so far. When it comes to training a model with a fundamental feature as a target and using a predictor off of that model as a node in a ranking system, are the predictions look-ahead biased? Would I need to train a predictor up to time t and evaluate my ranking system for time t + gap? Or is the predictor already trained with all data prior to the prediction date at each timestep?
If the latter is not true, will there eventually be support for using predictors that way within a ranking system?