feat: make us-cpa questions retrieval-first

skills/us-cpa/tests/test_questions.py

@@ -14,6 +14,12 @@ class QuestionEngineTests(unittest.TestCase):
         corpus = TaxYearCorpus(cache_root=Path(temp_dir))
 
         def fake_fetch(url: str) -> bytes:
+            if "p501" in url:
+                return (
+                    "A qualifying child may be your dependent if the relationship, age, residency, support, and joint return tests are met. "
+                    "Temporary absences due to education count as time lived with you. "
+                    "To meet the support test, the child must not have provided more than half of their own support for the year."
+                ).encode()
             return f"source for {url}".encode()
 
         corpus.download_catalog(2025, bootstrap_irs_catalog(2025), fetcher=fake_fetch)
@@ -93,6 +99,25 @@ class QuestionEngineTests(unittest.TestCase):
             self.assertEqual(analysis["conclusion"]["answer"], "$31,500")
             self.assertIn("Qualifying Surviving Spouse", analysis["conclusion"]["summary"])
 
+    def test_dependency_question_uses_irs_corpus_research_before_primary_law(self) -> None:
+        with tempfile.TemporaryDirectory() as temp_dir:
+            engine = self.build_engine(temp_dir)
+
+            analysis = engine.answer(
+                question=(
+                    "If my daughter went to college in 2025 starting in August, but also worked before that, "
+                    "should she be considered as a dependent?"
+                ),
+                tax_year=2025,
+                case_facts={},
+            )
+
+            self.assertEqual(analysis["issue"], "irs_corpus_research")
+            self.assertFalse(analysis["primaryLawRequired"])
+            self.assertEqual(analysis["authorities"][0]["slug"], "p501")
+            self.assertTrue(any(item["slug"] == "p501" for item in analysis["authorities"]))
+            self.assertTrue(analysis["excerpts"])
+
     def test_complex_question_flags_primary_law_escalation(self) -> None:
         with tempfile.TemporaryDirectory() as temp_dir:
             engine = self.build_engine(temp_dir)