Reject pscore_trim=0.0 to prevent infinite IPW weights; update docs

- Change pscore_trim validation from [0, 0.5) to (0, 0.5) since zero
  trimming allows exact 1.0 propensity scores through to the weight
  formula pscore/(1-pscore), producing inf/NaN
- Update REGISTRY.md fallback note to document error-mode re-raise
- Add pscore_trim to CallawaySantAnna and CallawaySantAnnaResults docstrings

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

Author: igerber

diff_diff/staggered.py

@@ -164,6 +164,10 @@ class CallawaySantAnna(
         event study aggregation. Requires ``n_bootstrap > 0``.
         When True, results include ``cband_crit_value`` and per-event-time
         ``cband_conf_int`` entries controlling family-wise error rate.
+    pscore_trim : float, default=0.01
+        Trimming bound for propensity scores. Scores are clipped to
+        ``[pscore_trim, 1 - pscore_trim]`` before weight computation
+        in IPW and DR estimation. Must be in ``(0, 0.5)``.
 
     Attributes
     ----------
@@ -265,8 +269,8 @@ def __init__(
             raise ValueError(
                 f"estimation_method must be 'dr', 'ipw', or 'reg', " f"got '{estimation_method}'"
             )
-        if not (0 <= pscore_trim < 0.5):
-            raise ValueError(f"pscore_trim must be in [0, 0.5), got {pscore_trim}")
+        if not (0 < pscore_trim < 0.5):
+            raise ValueError(f"pscore_trim must be in (0, 0.5), got {pscore_trim}")
 
         # Handle bootstrap_weight_type deprecation
         if bootstrap_weight_type is not None:

diff_diff/staggered_results.py

@@ -93,6 +93,8 @@ class CallawaySantAnnaResults:
         Effects aggregated by relative time (event study).
     group_effects : dict, optional
         Effects aggregated by treatment cohort.
+    pscore_trim : float
+        Propensity score trimming bound used during estimation.
     """
 
     group_time_effects: Dict[Tuple[Any, Any], Dict[str, Any]]

docs/methodology/REGISTRY.md

@@ -402,7 +402,8 @@ The multiplier bootstrap uses random weights w_i with E[w]=0 and Var(w)=1:
   - Trimming: Propensity scores clipped to `[pscore_trim, 1-pscore_trim]` (default
     0.01) before weight computation. Warning emitted when scores are trimmed.
   - Fallback: If IRLS fails entirely (LinAlgError/ValueError), falls back to
-    unconditional propensity score with warning
+    unconditional propensity score with warning. Exception: when
+    `rank_deficient_action="error"`, the error is re-raised instead of falling back.
 - Control group with `control_group="not_yet_treated"`:
   - Always excludes cohort g from controls when computing ATT(g,t)
   - This applies to both pre-treatment (t < g) and post-treatment (t >= g) periods

tests/test_staggered.py

@@ -3240,10 +3240,10 @@ def test_pscore_trim_above_half_raises(self):
         with pytest.raises(ValueError, match="pscore_trim must be in"):
             CallawaySantAnna(pscore_trim=0.6)
 
-    def test_pscore_trim_zero_succeeds(self):
-        """pscore_trim=0.0 is valid (no trimming)."""
-        cs = CallawaySantAnna(pscore_trim=0.0)
-        assert cs.pscore_trim == 0.0
+    def test_pscore_trim_zero_raises(self):
+        """pscore_trim=0.0 raises ValueError (would cause division by zero in IPW weights)."""
+        with pytest.raises(ValueError, match="pscore_trim must be in"):
+            CallawaySantAnna(pscore_trim=0.0)
 
     def test_pscore_trim_in_results(self):
         """results.pscore_trim matches the estimator's setting after fit()."""