Overview | Standard approach: classify then convert | WB approach: combine in density-ratio space | Clipping in the WB pathway | Custom DR-returning wrappers (required) | When to prefer WB over NNLS | Full example | References

Overview | 1. Density ratios: is the intervention supportable? | weight_diagnostics() | policy_change_summary | Picking a trim | 2. Q calibration: are the outcome models trustworthy? | branch_cal_summary() | Alternative: pure Q-view with qreg() | Tuning the targeting library (itmle() only) | 3. Final estimation | 4. Post-estimation sanity checks | diagnostics$recursion_diag | diagnostics$target_cal (iTMLE only) | diagnostics$sl_summary | ic_df | Workflow summary

Overview | Data structure | Simulated example data | Step 1 — Define a policy | Step 2 — Density ratios | Weights, intervention windows, and trim | Using dr_sl = TRUE (Wu-Benkeser metalearner) | Step 3 — SDR estimator | Step 4 — Infinite-dimensional TMLE (iTMLE) | Step 5 — Q-recursion sensitivity check (qreg) | Diagnostics | SuperLearner weights — sl_summary | Per-fold summaries — fold_diag | Branch calibration — diagnostics$branch_cal (SDR / qreg) | Targeting diagnostics — target_diag, targeting_gap (iTMLE only) | Density-ratio ESS | Parallelism | References