fix: FEDERAL_DEFLECTION could not see the participial state-subject form

CHANGELOG.md

@@ -11,6 +11,33 @@ researchers who pin a tag and want to know what they are pinning to.
 
 ## [Unreleased]
 
+### Fixed
+
+- **This moves a published number, in one direction. Re-run
+  `analyse-discourse` on any corpus, then re-read every evasion rate computed
+  from it.** `FEDERAL_DEFLECTION`'s state-subject pattern required a copula:
+  `\b(?:is|are)\s+(?:a|an)\s+[Ss]tate\s+[Ss]ubject\b`. Indian legislative
+  English drops it. "Agriculture **being** a State subject, there is no
+  investment..." matched nothing. The pattern now accepts `being`.
+
+  **Measured by zero-hour over 489,469 answered records (REQ-0072):** 34,632
+  answers mention "State subject", 5,542 match the shipped copula pattern, and
+  3,080 use the participle. Of those, 2,982 match no copula, and **518 carry no
+  label from any tier at all**.
+
+  **Two kinds of record move, and only the first is pure recall.** The 518
+  unlabelled records gain `FEDERAL_DEFLECTION`, which raises both the evasive
+  count and the classified denominator. The other 2,464 already carry a label
+  from some other pattern, and `FEDERAL_DEFLECTION`'s 0.92 confidence outranks
+  it. Where that old label was substantive, the evasion rate rises a second
+  time. Where it was evasive, only the label distribution changes.
+
+  **That precedence is not new and is not introduced here.** The copula form
+  already outranks `ACCEPTED` and `REJECTED` today at the same confidence. The
+  alternation makes the participle behave exactly as the copula does. Whether
+  a state-subject clause should outrank a substantive disclosure is a separate
+  question, and it applies to the shipped copula pattern first.
+
 ## [2.7.0] — 2026-08-20
 
 ### Changed

commoner_analyse/discourse.py

@@ -122,8 +122,13 @@ def _compile(patterns: Iterable[str]) -> tuple[re.Pattern, ...]:
     "Uses federalism/State Subjects to dodge Central responsibility for national standards.",
     _compile([
         r"library\s+is\s+a\s+state\s+subject",
-        # Broader state-subject dodge pattern (mined from Azad corpus UNCLASSIFIED)
-        r"\b(?:is|are)\s+(?:a|an)\s+[Ss]tate\s+[Ss]ubject\b",
+        # Broader state-subject dodge pattern (mined from Azad corpus UNCLASSIFIED).
+        # `being` is not optional decoration. Indian legislative English drops the
+        # copula in this exact construction, and the participial form carries the
+        # same federal dodge: "Agriculture being a State subject, ...". Measured
+        # over 489,469 answered records, 2,982 use the participle and match no
+        # copula, and 518 of those carry no label from any tier (REQ-0072).
+        r"\b(?:is|are|being)\s+(?:a|an)\s+[Ss]tate\s+[Ss]ubject\b",
         r"\b(?:Concurrent|[Cc]oncurrent)\s+[Ll]ist\b",
         r"\bunder\s+entry\s+\d+\s+of\s+the\s+[Ss]tate\s+List\b",
         r"\bSeventh\s+Schedule\b",

tests/test_discourse.py

@@ -407,6 +407,54 @@ def test_state_subject_fires_federal_deflection(self):
         )
         self.assertEqual(result.label, "FEDERAL_DEFLECTION")
 
+    def test_participial_state_subject_fires_federal_deflection(self):
+        """REQ-0072. Indian legislative English drops the copula in this exact
+        construction, and the pattern required it. Verbatim from the request,
+        source key ``LS|S|734|2000-05-17``, Minister of Agriculture. zero-hour
+        measured 2,982 answers using this form and invisible to the pattern,
+        of which 518 carried no label from any tier at all.
+        """
+        from commoner_analyse.discourse import CHANNEL_QA
+        result = classify_response(
+            "Agriculture being a State subject, there is no investment in "
+            "agriculture directly by the Central Government in the States.",
+            channel=CHANNEL_QA,
+        )
+        self.assertEqual(result.label, "FEDERAL_DEFLECTION")
+
+    def test_the_participle_and_the_copula_agree(self):
+        """The alternation adds recall. It does not create a second behaviour.
+
+        The two forms make the same claim, so they must reach the same label at
+        the same confidence. A drift between them is the defect returning.
+        """
+        from commoner_analyse.discourse import CHANNEL_QA
+        pairs = [
+            ("Agriculture is a State subject, so the Centre does not intervene.",
+             "Agriculture being a State subject, the Centre does not intervene."),
+            ("Health is a State subject and States decide.",
+             "Health being a State subject, States decide."),
+        ]
+        for copula, participle in pairs:
+            with self.subTest(participle=participle):
+                first = classify_response(copula, channel=CHANNEL_QA)
+                second = classify_response(participle, channel=CHANNEL_QA)
+                self.assertEqual(first.label, second.label)
+                self.assertEqual(first.confidence, second.confidence)
+
+    def test_the_copula_form_still_matches(self):
+        """Acceptance 3. The alternation must not move a copula match."""
+        from commoner_analyse.discourse import CHANNEL_QA
+        for text in (
+            "Land is a state subject under the Seventh Schedule.",
+            "These are a State subject and rest with the States.",
+        ):
+            with self.subTest(text=text):
+                self.assertEqual(
+                    classify_response(text, channel=CHANNEL_QA).label,
+                    "FEDERAL_DEFLECTION",
+                )
+
     def test_concurrent_list_fires_federal_deflection(self):
         from commoner_analyse.discourse import CHANNEL_QA
         result = classify_response(