Validate pscore_trim at fit() to guard against set_params bypass

set_params() can inject invalid pscore_trim values since it bypasses
__init__ validation. Add check at fit() entry to catch 0.0, negative,
and >=0.5 values before they reach IPW/DR weight formulas.

Also update TripleDifference registry fallback note for error mode.

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

Author: igerber

diff_diff/staggered.py

@@ -1082,6 +1082,10 @@ def fit(
         ValueError
             If required columns are missing or data validation fails.
         """
+        # Validate pscore_trim (may have been changed via set_params)
+        if not (0 < self.pscore_trim < 0.5):
+            raise ValueError(f"pscore_trim must be in (0, 0.5), got {self.pscore_trim}")
+
         # Normalize empty covariates list to None
         if covariates is not None and len(covariates) == 0:
             covariates = None

docs/methodology/REGISTRY.md

@@ -1191,7 +1191,8 @@ has no additional effect.
 - Cluster IDs: must not contain NaN (raises `ValueError`)
 - Overlap warning: emitted when >5% of observations are trimmed at pscore bounds (IPW/DR only)
 - Propensity score estimation failure: falls back to unconditional probability P(subgroup=4),
-  sets hessian=None (skipping PS correction in influence function), emits UserWarning
+  sets hessian=None (skipping PS correction in influence function), emits UserWarning.
+  Exception: when `rank_deficient_action="error"`, the error is re-raised instead of falling back.
 - Collinear covariates: detected via pivoted QR in `solve_ols()`, action controlled by
   `rank_deficient_action` ("warn", "error", "silent")
 - Non-finite influence function values (e.g., from extreme propensity scores in IPW/DR

tests/test_staggered.py

@@ -3220,6 +3220,32 @@ def test_set_params_pscore_trim(self):
         cs.set_params(pscore_trim=0.1)
         assert cs.pscore_trim == 0.1
 
+    def test_set_params_invalid_pscore_trim_rejected_at_fit(self):
+        """Invalid pscore_trim via set_params() raises ValueError at fit()."""
+        np.random.seed(42)
+        n_units, n_periods = 50, 6
+        units = np.repeat(np.arange(n_units), n_periods)
+        times = np.tile(np.arange(n_periods), n_units)
+        first_treat = np.zeros(n_units)
+        first_treat[n_units // 2 :] = 3
+        first_treat_expanded = np.repeat(first_treat, n_periods)
+        post = (times >= first_treat_expanded) & (first_treat_expanded > 0)
+        outcomes = 1.0 + 2.0 * post + np.random.randn(len(units)) * 0.5
+        data = pd.DataFrame(
+            {
+                "unit": units,
+                "time": times,
+                "outcome": outcomes,
+                "first_treat": first_treat_expanded.astype(int),
+            }
+        )
+
+        for bad_val in [0.0, -0.1, 0.5]:
+            cs = CallawaySantAnna(estimation_method="ipw")
+            cs.set_params(pscore_trim=bad_val)
+            with pytest.raises(ValueError, match="pscore_trim must be in"):
+                cs.fit(data, outcome="outcome", unit="unit", time="time", first_treat="first_treat")
+
     def test_default_pscore_trim(self):
         """Default pscore_trim is 0.01."""
         cs = CallawaySantAnna()