Fix summary() crash when outcome model is fit without adjust()

_outcome_estimates_summary() unconditionally called aipw(), which raises
when the frame has not been adjusted (aipw() needs adjust()-calibrated
responder weights on the target population scale). The changelog documents
the no-adjust() outcome-model workflow, so summary() on such a frame
crashed with a ValueError instead of reporting. Omit the mu_DR/AIPW block
when not adjusted while still showing mu_IPW and mu_OM, and add a
regression test.

Author: rootkiller6788

balance/balance_frame.py

@@ -3752,8 +3752,6 @@ def _outcome_estimates_summary(self) -> str | None:
             for col in model["outcome_columns"]
             if f"{col}_hat" in target_predictions.columns
         }
-        dr_estimates = {str(col): float(v) for col, v in self.aipw().items()}
-
         def _fmt_points(estimates: dict[str, float]) -> str:
             return "\n".join(
                 f"    {col}    {val:.3f}" for col, val in estimates.items()
@@ -3770,9 +3768,20 @@ def _fmt_points(estimates: dict[str, float]) -> str:
             "Outcome model / g-computation (mu_OM), point estimate "
             "(CI not shown; not cached):",
             _fmt_points(om_estimates),
-            "Doubly robust / AIPW (mu_DR), point estimate:",
-            _fmt_points(dr_estimates),
         ]
+
+        # μ̂_DR requires adjust()-calibrated responder weights on the same
+        # population scale as the target (``aipw()`` enforces this). A frame
+        # with a fitted outcome model but no adjust() is a supported workflow
+        # (see the changelog), so omit the AIPW block instead of crashing.
+        if self.is_adjusted:
+            dr_estimates = {str(col): float(v) for col, v in self.aipw().items()}
+            blocks.extend(
+                [
+                    "Doubly robust / AIPW (mu_DR), point estimate:",
+                    _fmt_points(dr_estimates),
+                ]
+            )
         return "Outcome estimates:\n" + "\n".join(blocks)
 
     def summary(self) -> str:

tests/test_aipw.py

@@ -315,6 +315,24 @@ def test_summary_shows_ipw_om_dr_when_model_fit(self) -> None:
         # the rich section replaces the plain "Outcome weighted means" block
         self.assertNotIn("Outcome weighted means", summary)
 
+    def test_summary_omits_dr_when_model_fit_but_unadjusted(self) -> None:
+        """summary() with a fitted outcome model but NO adjust() must not crash.
+
+        The μ̂_DR/AIPW block requires adjust()-calibrated weights, so it is
+        omitted (μ̂_IPW and μ̂_OM are still shown) rather than raising.
+        """
+        sample_df, target_df = _make_aipw_fixture()
+        s = Sample.from_frame(
+            sample_df, id_column="id", weight_column="weight", outcome_columns=["y"]
+        )
+        t = Sample.from_frame(target_df, id_column="id", weight_column="weight")
+        st = s.set_target(t).fit_outcome_model(model=LinearRegression())
+        summary = st.summary()  # previously raised ValueError from aipw()
+        self.assertIn("Outcome estimates:", summary)
+        self.assertIn("mu_IPW", summary)
+        self.assertIn("mu_OM", summary)
+        self.assertNotIn("mu_DR", summary)
+
     def test_summary_unchanged_without_outcome_model(self) -> None:
         sample_df, target_df = _make_aipw_fixture()
         s = Sample.from_frame(