Fix VCV index alignment, add stationarity warning for panel=False

VCV index: Build event_study_vcov_index from event times that actually
contributed psi vectors (skipping NaN-only periods), not from all
sorted_periods. Fixes misalignment when interior event times drop out.

Stationarity: Document "stationary repeated cross-sections" in panel
parameter docstring. Emit UserWarning on panel=False noting the
stationarity assumption is not data-checkable.

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

Author: igerber

diff_diff/staggered.py

@@ -183,9 +183,11 @@ class CallawaySantAnna(
         in IPW and DR estimation. Must be in ``(0, 0.5)``.
     panel : bool, default=True
         Whether the data is a balanced/unbalanced panel (units observed
-        across multiple time periods). Set to ``False`` for repeated
-        cross-sections where each observation has a unique unit ID and
-        units do not repeat across periods. Uses cross-sectional DRDID
+        across multiple time periods). Set to ``False`` for stationary
+        repeated cross-sections where each observation has a unique unit
+        ID and units do not repeat across periods. Requires that the
+        cross-sectional samples are drawn from the same population in
+        each period (stationarity). Uses cross-sectional DRDID
         (Sant'Anna & Zhao 2020, Section 4) with per-observation influence
         functions.
 
@@ -1323,6 +1325,17 @@ def fit(
         # Reset stale state from prior fit (prevents leaking event-study VCV)
         self._event_study_vcov = None
 
+        if not self.panel:
+            warnings.warn(
+                "panel=False uses repeated cross-section DRDID estimators "
+                "(Sant'Anna & Zhao 2020, Section 4) which assume stationary "
+                "cross-sectional sampling: the population distribution of "
+                "(Y, X, G) must be stable across periods. This assumption "
+                "is not data-checkable.",
+                UserWarning,
+                stacklevel=2,
+            )
+
         # Normalize empty covariates list to None
         if covariates is not None and len(covariates) == 0:
             covariates = None

diff_diff/staggered_aggregation.py

@@ -627,6 +627,7 @@ def _aggregate_event_study(
         agg_ses_list = []
         agg_n_groups = []
         _psi_vectors = []  # Per-event-time combined IF vectors for VCV
+        _psi_event_times = []  # Event times that contributed a psi column
         for e, effect_list in sorted_periods:
             gt_pairs = [x[0] for x in effect_list]
             effs = np.array([x[1] for x in effect_list])
@@ -666,6 +667,7 @@ def _aggregate_event_study(
             agg_ses_list.append(agg_se)
             agg_n_groups.append(len(effect_list))
             _psi_vectors.append(psi_e)
+            _psi_event_times.append(e)
 
         # Batch inference for all relative periods
         if not agg_effects_list:
@@ -753,9 +755,7 @@ def _aggregate_event_study(
 
         # Store the event-time index that matches VCV columns (for subsetting
         # in HonestDiD when some event times are filtered out)
-        self._event_study_vcov_index = (
-            [e for e, _ in sorted_periods] if event_study_vcov is not None else None
-        )
+        self._event_study_vcov_index = _psi_event_times if event_study_vcov is not None else None
 
         # Attach VCV to self for CallawaySantAnna to pick up
         self._event_study_vcov = event_study_vcov