Fix CI review R4: persist target_label on HonestDiDResults, render dynamically

- Add target_label field to HonestDiDResults (default: equal-weight avg)
- HonestDiD.fit() detects common l_vec patterns and sets human-readable label
  (on-impact, equal-weight, or custom with vector)
- Summary renders hd.target_label instead of hard-coded string
- Add test_honest_did_custom_l_vec_summary_label: attaches custom-target
  results and asserts summary shows "on-impact" not "Equal-weight"

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

Author: igerber

diff_diff/chaisemartin_dhaultfoeuille_results.py

@@ -987,7 +987,7 @@ def _render_honest_did_section(
                 "HonestDiD Sensitivity (Rambachan-Roth 2023)".center(width),
                 thin,
                 f"{'Method:':<35} {method_label} (M={_fmt_float(m_val)})",
-                f"{'Target:':<35} {'Equal-weight avg over post horizons'}",
+                f"{'Target:':<35} {hd.target_label}",
                 f"{'Original estimate:':<35} {_fmt_float(hd.original_estimate):>10}",
                 f"{'Identified set:':<35} "
                 f"[{_fmt_float(hd.lb)}, {_fmt_float(hd.ub)}]",

diff_diff/honest_did.py

@@ -191,6 +191,7 @@ class HonestDiDResults:
     original_se: float
     alpha: float = 0.05
     ci_method: str = "FLCI"
+    target_label: str = "Equal-weight avg over post horizons"
     original_results: Optional[Any] = field(default=None, repr=False)
     # Event study bounds (optional)
     event_study_bounds: Optional[Dict[Any, Dict[str, float]]] = field(default=None, repr=False)
@@ -2252,13 +2253,23 @@ def fit(
                 "coefficient to compute bounds."
             )
 
-        # Set up weighting vector
+        # Set up weighting vector and target label
         if self.l_vec is None:
             l_vec = np.ones(num_post) / num_post  # Uniform weights
+            target_label = "Equal-weight avg over post horizons"
         else:
             l_vec = np.asarray(self.l_vec)
             if len(l_vec) != num_post:
                 raise ValueError(f"l_vec must have length {num_post}, got {len(l_vec)}")
+            # Detect common patterns for a human-readable label
+            basis = np.zeros(num_post)
+            basis[0] = 1.0
+            if np.allclose(l_vec, basis):
+                target_label = "First post-treatment effect (on-impact)"
+            elif np.allclose(l_vec, np.ones(num_post) / num_post):
+                target_label = "Equal-weight avg over post horizons"
+            else:
+                target_label = f"Custom l_vec ({l_vec.tolist()})"
 
         # Compute original estimate and SE
         original_estimate = np.dot(l_vec, beta_post)
@@ -2318,6 +2329,7 @@ def fit(
             original_se=original_se,
             alpha=self.alpha,
             ci_method=ci_method,
+            target_label=target_label,
             original_results=results,
             survey_metadata=survey_metadata,
             df_survey=df_survey,

tests/test_chaisemartin_dhaultfoeuille.py

@@ -3404,6 +3404,25 @@ def test_honest_did_custom_l_vec_on_impact(self):
             rtol=1e-10,
         )
 
+    def test_honest_did_custom_l_vec_summary_label(self):
+        """summary() renders custom target label when l_vec is overridden."""
+        from diff_diff.honest_did import compute_honest_did
+
+        df = self._make_data()
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            r = ChaisemartinDHaultfoeuille(seed=1).fit(
+                df, "outcome", "group", "period", "treatment",
+                L_max=2,
+            )
+        # Attach custom-target HonestDiD to results
+        r.honest_did_results = compute_honest_did(
+            r, l_vec=np.array([1.0, 0.0])
+        )
+        text = r.summary()
+        assert "on-impact" in text.lower()
+        assert "Equal-weight" not in text
+
     def test_honest_did_with_trends_nonparam(self):
         """End-to-end trends_nonparam + honest_did=True."""
         rng = np.random.RandomState(42)