High values of est.score_ for LinearDRLearner

Hi and thanks a lot for your great package! I am analysing an AB test and my main goal is to explore treatment heterogeneity. I had a question about how to interpret the score values for a model. My outcome has a mean of 2.2 and sd of 5.1 - my first stage propensity model has a nuisance score of 0.53 (T comes from an experiment), and the regression nuisance for weighted lasso is 0.02. Finally, est.score _ends up being 105, which seems really high given the scale of my outcome variable. Is this just because the model is fitting poorly, or am I missing something? I already supplied W=None in the model since treatment is randomised, but should I also plug in 0.5 as the propensity instead of estimating it?

Could you also direct me to the part of the code base based on which the score is computed (the code for drlearner somehow doesn't specify where the nuisances are calculated). Thank you!

Start with the LinearDRLearner implementation and trace where est.score_ and its nuisance models are computed; the issue does not name specific files or tests. Compare that path with the documented score behavior and identify the code or documentation needed to explain the reported scale and randomized-treatment configuration.