Estimate heterogeneous treatment effects with causal ML

You hold logged, observational records from a promotion that runs continuously, and your goal is to target users so that incremental revenue is maximized. Lay out a causal machine learning method for estimating heterogeneous treatment effects (HTE) and turning those estimates into a targeting policy. Cover the following:

(a) pick among T-, S-, X- and DR-learners, defending the choice in terms of bias–variance trade-offs and ignorability; (b) which models to use and how cross-fitting limits overfitting and nuisance-estimation bias; (c) how to deal with positivity/overlap violations and with extreme propensities; (d) the evaluation metrics — policy value estimation, uplift/Qini, a PEHE proxy — plus how to run off-policy evaluation with IPW and doubly robust estimators; (e) fairness constraints such as demographic parity of treatment, and what they do to the policy; (f) how you would safely A/B test the learned policy before rolling it out fully.

Overview: This question probes a candidate's grasp of causal machine learning, heterogeneous treatment effect estimation, off-policy evaluation, and fairness-aware policy design built on observational data.