Address CI review P0/P1 findings for PR #237

P0: Rao-Wu census FPC (f_h >= 1) now checked before m_h, keeping
    original weights for census strata instead of unreachable guard
P1: ContinuousDiD ACRT_glob bootstrap uses survey-weighted dpsi_mean
P1: Document lonely_psu="adjust" rejection in REGISTRY Phase 6 section
P1: Document TROP no-strata pseudo-strata behavior in REGISTRY

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: igerber

diff_diff/bootstrap_utils.py

@@ -605,15 +605,14 @@ def generate_rao_wu_weights(
                     f"({n_h}). FPC must be >= number of PSUs."
                 )
             f_h = n_h / N_h
+            if f_h >= 1.0:
+                # Census stratum — keep original weights (zero variance)
+                rescaled[mask_h] = base_weights[mask_h]
+                continue
             m_h = max(1, round((1.0 - f_h) * (n_h - 1)))
         else:
             m_h = n_h - 1
 
-        if m_h == 0:
-            # Full census — keep original weights
-            rescaled[mask_h] = base_weights[mask_h]
-            continue
-
         # Draw m_h PSUs with replacement
         drawn_indices = rng.choice(n_h, size=m_h, replace=True)
         counts = np.bincount(drawn_indices, minlength=n_h)

diff_diff/continuous_did.py

@@ -1438,7 +1438,11 @@ def _bootstrap_gt_cell(gt, info):
                 mean_dy_treated_pert = (w_treated @ att_glob_score) / n_t
                 att_glob_b = att_glob_gt + mean_dy_treated_pert - mu_0_pert
 
-                dpsi_mean = np.mean(dPsi_treated, axis=0)
+                if sw_treated is not None:
+                    sw_norm = sw_treated / sw_treated.sum()
+                    dpsi_mean = sw_norm @ dPsi_treated
+                else:
+                    dpsi_mean = np.mean(dPsi_treated, axis=0)
                 acrt_glob_b = delta_beta @ dpsi_mean
 
             return att_d_b, acrt_d_b, att_glob_b, acrt_glob_b, info.get("acrt_glob", 0.0)

docs/methodology/REGISTRY.md

@@ -1961,6 +1961,10 @@ ContinuousDiD, EfficientDiD):
   Without FPC: `m_h = n_h - 1`. With FPC: `m_h = max(1, round((1 - f_h) * (n_h - 1)))`.
   Rescaled weight: `w*_i = w_i * (n_h / m_h) * r_hi` where `r_hi` = count of PSU *i* drawn.
 - **Note:** FPC enters through the resample size `m_h`, not as a post-hoc scaling factor.
+  When `f_h >= 1` (census stratum), observations keep original weights (zero variance).
+- **Note:** Bootstrap paths support `lonely_psu="remove"` and `"certainty"` only.
+  `lonely_psu="adjust"` raises `NotImplementedError` for survey-aware bootstrap;
+  use analytical inference for designs requiring `adjust` semantics.
 - **Deviation from R:** R `survey::as.svrepdesign(type="subbootstrap")` uses the same
   formula. Our implementation matches.
 
@@ -1978,6 +1982,10 @@ ContinuousDiD, EfficientDiD):
   creates pseudo-strata as `(survey_stratum x treatment_group)` for Rao-Wu resampling.
   This preserves both survey variance structure and treatment ratio. Survey df computed
   from pseudo-strata structure.
+- **Note:** When `survey_design.strata` is None but PSU/FPC trigger full-design bootstrap,
+  TROP uses treatment group (treated vs control) as pseudo-strata for Rao-Wu resampling
+  to preserve treatment ratio. FPC is applied within these pseudo-strata. This matches
+  TROP's existing treatment-stratified resampling pattern.
 
 ---