Fix bootstrap RCS cohort-mass weighting, reset stale event-study VCV

- Bootstrap overall/event-study reaggregation now uses agg_weight (fixed
  cohort mass) for panel=False, matching the analytical aggregation path
- Reset self._event_study_vcov = None at start of fit() to prevent stale
  VCV from prior fit leaking into reused estimator objects

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

Author: igerber

diff_diff/staggered.py

@@ -1320,6 +1320,9 @@ def fit(
         if not (0 < self.pscore_trim < 0.5):
             raise ValueError(f"pscore_trim must be in (0, 0.5), got {self.pscore_trim}")
 
+        # Reset stale state from prior fit (prevents leaking event-study VCV)
+        self._event_study_vcov = None
+
         # Normalize empty covariates list to None
         if covariates is not None and len(covariates) == 0:
             covariates = None

diff_diff/staggered_bootstrap.py

@@ -237,8 +237,14 @@ def _run_multiplier_bootstrap(
                     _cohort_mass_cache[g] = float(np.sum(survey_w[unit_cohorts == g]))
             all_n_treated = np.array([_cohort_mass_cache[gt[0]] for gt in gt_pairs], dtype=float)
         else:
+            # Use agg_weight if available (RCS: fixed cohort mass);
+            # fall back to n_treated for panel data
             all_n_treated = np.array(
-                [group_time_effects[gt]["n_treated"] for gt in gt_pairs], dtype=float
+                [
+                    group_time_effects[gt].get("agg_weight", group_time_effects[gt]["n_treated"])
+                    for gt in gt_pairs
+                ],
+                dtype=float,
             )
         post_n_treated = all_n_treated[post_treatment_mask]
 
@@ -572,7 +578,10 @@ def _agg_weight(g: Any, t: Any) -> float:
                 if g not in _cohort_mass:
                     _cohort_mass[g] = float(np.sum(survey_w[unit_cohorts == g]))
                 return _cohort_mass[g]
-            return group_time_effects[(g, t)]["n_treated"]
+            # Use agg_weight if available (RCS: fixed cohort mass)
+            return group_time_effects[(g, t)].get(
+                "agg_weight", group_time_effects[(g, t)]["n_treated"]
+            )
 
         # Organize by relative time
         effects_by_e: Dict[int, List[Tuple[int, float, float]]] = {}