Add front-door validation to lpbwselect_mse_dpi (P1)

CI AI review P1: the port's lpbwselect_mse_dpi is advertised as the
advanced-use entry point for callers outside the HAD Phase 1b
wrapper surface. It accepted x, y, cluster without validating
shapes, and the default vce="nn" branch reindexed y/cluster by
argsort(x) -- so a longer y would be silently truncated to match
x's length, producing a bandwidth on misaligned data with no
warning.

Added front-door validation at the top of lpbwselect_mse_dpi:
- x and y ravel()ed and required to have the same shape
- x must be non-empty
- x, y, and eval_point must be finite
- cluster (if supplied) must match x.shape

Seven new tests in TestLpbwselectMseDpiValidation:
- test_mismatched_shapes_raise
- test_longer_y_silent_truncation_rejected (regression for the
  specific nn-reindex bug)
- test_cluster_wrong_length_rejected
- test_empty_direct_port_input_rejected
- test_non_finite_x_rejected
- test_non_finite_y_rejected
- test_non_finite_eval_point_rejected

185 tests pass (up from 178).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: igerber

diff_diff/_nprobust_port.py

@@ -650,10 +650,31 @@ def lpbwselect_mse_dpi(
     if vce not in _VALID_VCE:
         raise ValueError(f"Unknown vce {vce!r}. Expected one of {_VALID_VCE}.")
 
-    x = np.asarray(x, dtype=np.float64)
-    y = np.asarray(y, dtype=np.float64)
+    x = np.asarray(x, dtype=np.float64).ravel()
+    y = np.asarray(y, dtype=np.float64).ravel()
+    if x.shape != y.shape:
+        raise ValueError(
+            f"x and y must have the same 1-D shape; got "
+            f"{x.shape} and {y.shape}"
+        )
+    if x.size == 0:
+        raise ValueError(
+            "x and y must be non-empty; lpbwselect_mse_dpi cannot "
+            "estimate a bandwidth from zero observations."
+        )
+    if not np.all(np.isfinite(x)):
+        raise ValueError("x contains non-finite values (NaN or Inf)")
+    if not np.all(np.isfinite(y)):
+        raise ValueError("y contains non-finite values (NaN or Inf)")
+    if not np.isfinite(eval_point):
+        raise ValueError(f"eval_point must be finite; got {eval_point}")
     if cluster is not None:
-        cluster = np.asarray(cluster)
+        cluster = np.asarray(cluster).ravel()
+        if cluster.shape != x.shape:
+            raise ValueError(
+                f"cluster must have the same shape as x; got "
+                f"{cluster.shape} and {x.shape}"
+            )
     if q is None:
         q = p + 1
 

tests/test_nprobust_port.py

@@ -241,3 +241,78 @@ def test_R_is_zero_when_scale_zero(self):
         )
         # With scale=0, BWreg is never computed -> R stays 0.
         assert C_d1.R == 0.0
+
+
+# =============================================================================
+# lpbwselect_mse_dpi: input validation on the advanced-use entry point
+# =============================================================================
+
+
+class TestLpbwselectMseDpiValidation:
+    """The public wrapper is restricted to the HAD surface; the port is
+    the advertised advanced-use entry point. It must enforce its own
+    shape / emptiness / finiteness contract -- silently truncating a
+    longer y or cluster through sort-index reindexing would be a real
+    bug."""
+
+    def test_mismatched_shapes_raise(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        x = np.array([0.1, 0.2, 0.3])
+        y = np.array([1.0, 2.0, 3.0, 4.0])  # length 4 != 3
+        with pytest.raises(ValueError, match="same 1-D shape"):
+            lpbwselect_mse_dpi(y, x, eval_point=0.0)
+
+    def test_longer_y_silent_truncation_rejected(self):
+        """Regression: len(y) > len(x) previously got truncated via
+        the sort-indexer under vce='nn'. Must now raise."""
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        rng = np.random.default_rng(0)
+        x = rng.uniform(0.0, 1.0, size=100)
+        y = rng.normal(size=200)  # twice the length
+        with pytest.raises(ValueError, match="same 1-D shape"):
+            lpbwselect_mse_dpi(y, x, eval_point=0.0, vce="nn")
+
+    def test_cluster_wrong_length_rejected(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        rng = np.random.default_rng(0)
+        x = rng.uniform(0.0, 1.0, size=100)
+        y = rng.normal(size=100)
+        cluster = np.arange(50)  # wrong length
+        with pytest.raises(ValueError, match="cluster must have"):
+            lpbwselect_mse_dpi(y, x, cluster=cluster, eval_point=0.0)
+
+    def test_empty_direct_port_input_rejected(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        x = np.array([], dtype=np.float64)
+        y = np.array([], dtype=np.float64)
+        with pytest.raises(ValueError, match="non-empty"):
+            lpbwselect_mse_dpi(y, x, eval_point=0.0)
+
+    def test_non_finite_x_rejected(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        x = np.array([0.1, np.nan, 0.3])
+        y = np.array([1.0, 2.0, 3.0])
+        with pytest.raises(ValueError, match="x contains non-finite"):
+            lpbwselect_mse_dpi(y, x, eval_point=0.0)
+
+    def test_non_finite_y_rejected(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        x = np.array([0.1, 0.2, 0.3])
+        y = np.array([1.0, np.inf, 3.0])
+        with pytest.raises(ValueError, match="y contains non-finite"):
+            lpbwselect_mse_dpi(y, x, eval_point=0.0)
+
+    def test_non_finite_eval_point_rejected(self):
+        from diff_diff._nprobust_port import lpbwselect_mse_dpi
+
+        rng = np.random.default_rng(0)
+        x = rng.uniform(0.0, 1.0, size=100)
+        y = rng.normal(size=100)
+        with pytest.raises(ValueError, match="eval_point"):
+            lpbwselect_mse_dpi(y, x, eval_point=np.nan)