S0-R1-C5.5-R6-R5 Discussion Study Interpretation + Evidence-Limited Citation Closure (#93)

- S0-R1-C5.5-R6-R5 Discussion Study Interpretation + Evidence-Limited Citation Closure (d7abdcb1ff5ce630f6264818e5c1f6a41a261f21)

core/generation_runner.py

@@ -2355,6 +2355,20 @@ class GenerationRunner:
             else {"introduction", "theory", "methodology", "discussion"}
         )
 
+    @staticmethod
+    def _quota_driven_citation_recovery_allowed(section_id: str) -> bool:
+        """Discussion evidence is claim-driven, never padded to a word-count target.
+
+        R6-R5 keeps claim-specific planner assignments, but disables reserve/JIT
+        top-ups whose only trigger is a numeric citation target.  After drafting,
+        the existing bounded top-level completion may still attach a directly
+        validated source to an actual external claim.
+        """
+        return str(section_id or '').strip().lower() != 'discussion'
+
+    # Backward-readable name used by R6-R5 replay documentation.
+    _pre_draft_citation_recovery_allowed = _quota_driven_citation_recovery_allowed
+
     @classmethod
     def _adaptive_citation_requirements(cls, tasks, mode: str):
         """Derive evidence demand and semantic-survival reserve from manuscript units."""
@@ -3029,6 +3043,13 @@ class GenerationRunner:
             }
             for display_name, reserve_target in citation_reserve_targets.items():
                 planned_items = citation_plan.setdefault(display_name, [])
+                unit_section_id = str(task_by_name.get(display_name, {}).get("section_id") or "")
+                if not self._quota_driven_citation_recovery_allowed(unit_section_id):
+                    logging.info(
+                        "Discussion pre-draft reserve recovery skipped: unit=%s direct_plan=%s policy=claim_driven",
+                        display_name, len(planned_items),
+                    )
+                    continue
                 shortfall = max(0, reserve_target - len(planned_items))
                 if shortfall <= 0:
                     continue
@@ -3054,6 +3075,12 @@ class GenerationRunner:
                 deficit = max(0, int(required) - int(top_level_counts.get(section_id, 0)))
                 if deficit <= 0:
                     continue
+                if not self._quota_driven_citation_recovery_allowed(section_id):
+                    logging.info(
+                        "Discussion pre-draft JIT quota top-up skipped: section=%s direct_plan=%s required=%s policy=claim_driven",
+                        section_id, int(top_level_counts.get(section_id, 0)), int(required),
+                    )
+                    continue
                 section_units = [
                     name for name, mapped in citation_unit_section_map.items() if mapped == section_id
                 ]
@@ -3643,6 +3670,14 @@ class GenerationRunner:
                     "\nCURRENT-STUDY PROVENANCE RULE: Do not attach external literature citations to this study's own research question, H1-Hn statement, sample/design fact, analytic choice, result, limitation specific to the available study evidence, or author positioning. "
                     "If prior literature provides a rationale, state that external rationale in a separate sentence and cite only that sentence."
                 )
+            if section_id == "discussion":
+                citation_rule += (
+                    "\nDISCUSSION EVIDENCE-LIMITED CONTRACT: Distinguish prior-literature claims from study-owned interpretation. "
+                    "Statements such as 'the present evidence suggests', 'these findings indicate', or 'the present results suggest' are interpretations of this study and must not borrow an external citation. "
+                    "Do not create or broaden an external-literature claim merely to consume a citation or satisfy a citation count. "
+                    "When claim-specific literature is sparse, develop the Discussion from documented study evidence, cautious conceptual reasoning, boundary conditions, and limitations. "
+                    "If comparing with prior literature, put the directly supported prior-literature proposition in its own cited sentence and the study-owned interpretation in a separate uncited sentence."
+                )
             if section_id not in {"results", "findings", "appendix"}:
                 citation_rule += (
                     "\nATOMIC LITERATURE CLAIM CONTRACT: Write one externally verifiable proposition per sentence. "
@@ -4022,7 +4057,11 @@ The previous attempt for this same manuscript unit was rejected by validation. R
 
                 used_after = set(agent.used_citations)
                 survived = used_after - used_before
-                if attempt_assigned_tag_list and not is_quarantined:
+                if (
+                    attempt_assigned_tag_list
+                    and not is_quarantined
+                    and self._quota_driven_citation_recovery_allowed(section_id)
+                ):
                     required_survival = min(
                         len(attempt_assigned_tag_list),
                         max(1, int(math.ceil(len(attempt_assigned_tag_list) * CITATION_SEMANTIC_SURVIVAL_RATIO))),

core/pipeline_agent.py

@@ -3327,6 +3327,63 @@ INPUT:
             return 'STRUCTURAL'
         return 'EXTERNAL_LITERATURE_CLAIM'
 
+    @staticmethod
+    def _citation_discussion_section(section_id='', section_name=''):
+        sid = str(section_id or '').strip().lower()
+        if sid in {'discussion', 'conclusion'}:
+            return True
+        name = re.sub(r'[^a-z]+', ' ', str(section_name or '').lower()).strip()
+        return bool(re.search(r'\b(?:discussion|conclusion|implications?)\b', name))
+
+    @staticmethod
+    def _citation_external_literature_signal(sentence):
+        """High-confidence signal that a clause explicitly attributes a claim to prior literature."""
+        text = re.sub(r'\s+', ' ', str(sentence or '')).casefold().strip()
+        if not text:
+            return False
+        return bool(re.search(
+            r'\b(?:prior|previous|existing|earlier) (?:studies|research|work|literature)|'
+            r'\bthe (?:prior|existing|broader) literature\b|'
+            r'\b(?:studies|researchers|scholars) (?:have )?(?:shown|show|reported|report|argued|argue|suggested|suggest|found|find|documented|document)\b|'
+            r'\bevidence from (?:prior|previous|existing|earlier) (?:studies|research|work)\b',
+            text, flags=re.I,
+        ))
+
+    @classmethod
+    def _citation_study_interpretation_signal(cls, sentence, section_id='', section_name=''):
+        """Identify Discussion/Conclusion interpretations owned by this study, not by external literature.
+
+        R6-R5 distinguishes an interpretation of the frozen study evidence from a
+        literature finding.  The signal is deliberately narrow: it requires both a
+        current-study anchor and an interpretive predicate in a Discussion-like unit.
+        Pure descriptions of prior research remain literature-owned, while mixed
+        comparison sentences are handled by the evidence-ownership splitter.
+        """
+        if not cls._citation_discussion_section(section_id, section_name):
+            return False
+        text = re.sub(r'\s+', ' ', str(sentence or '')).casefold().strip()
+        if not text:
+            return False
+        anchor = bool(re.search(
+            r'\b(?:the present|these|our|the current) (?:evidence|findings|results)|'
+            r'\b(?:the findings|the results|the evidence) (?:of|from|in) (?:this|the present|the current) study|'
+            r'\bthe (?:pattern|association|associations|relationship|relationships) observed here|'
+            r'\bthe observed (?:pattern|association|associations|relationship|relationships)|'
+            r'\btaken together,? (?:the|these|our) (?:findings|results|evidence)|'
+            r'\bthe emergent process model\b|'
+            r'\bthe process model developed here\b',
+            text, flags=re.I,
+        ))
+        if not anchor:
+            return False
+        interpretive = bool(re.search(
+            r'\b(?:suggests?|indicates?|implies?|shows?|reveals?|highlights?|points? to|qualif(?:y|ies|ied)|'
+            r'is consistent with|are consistent with|appears? to|supports? (?:the )?interpretation|'
+            r'warrants?|means? that|by showing)\b',
+            text, flags=re.I,
+        ))
+        return interpretive
+
     @classmethod
     def _citation_study_evidence_signal(cls, sentence, section_name=''):
         text = re.sub(r'\s+', ' ', str(sentence or '')).casefold().strip()
@@ -3400,9 +3457,24 @@ INPUT:
         clean = re.sub(r'\s*\[?CITE-\d+\]?\s*', ' ', raw, flags=re.I)
         clean = re.sub(r'\s+', ' ', clean).strip()
         base = cls._citation_base_provenance(clean, section_name)
+        study_interpretation = cls._citation_study_interpretation_signal(
+            clean, section_id=section_id, section_name=section_name
+        )
 
+        # Discussion interpretations of this study's own evidence override a
+        # superficial RESULT/FINDING token in a subsection title.  If the same
+        # sentence also attributes a proposition to prior literature, ownership
+        # becomes mixed and the sentence must be split before citation QA.
+        if study_interpretation:
+            components = []
+            if cls._citation_external_literature_signal(clean):
+                components.append('EXTERNAL_LITERATURE_CLAIM')
+            components.append('STUDY_INTERPRETATION')
+            provenance = (
+                'MIXED_EVIDENCE_OWNERSHIP' if len(components) > 1 else 'STUDY_INTERPRETATION'
+            )
         # Structural/hypothesis/positioning/result identities stay authoritative.
-        if base in {'STRUCTURAL', 'HYPOTHESIS_STATEMENT', 'STUDY_POSITIONING', 'RESULT_OWNED_FACT'}:
+        elif base in {'STRUCTURAL', 'HYPOTHESIS_STATEMENT', 'STUDY_POSITIONING', 'RESULT_OWNED_FACT'}:
             provenance = base
             components = [base]
         else:
@@ -3439,6 +3511,7 @@ INPUT:
             'METHOD_PRINCIPLE': 'METHOD_AUTHORITY',
             'EXTERNAL_LITERATURE_CLAIM': 'LITERATURE_EVIDENCE',
             'AUTHOR_INFERENCE': 'AUTHOR_REASONING',
+            'STUDY_INTERPRETATION': 'STUDY_EVIDENCE_LEDGER+AUTHOR_REASONING',
             'STUDY_POSITIONING': 'AUTHOR_REASONING',
             'HYPOTHESIS_STATEMENT': 'HYPOTHESIS_CONTRACT',
             'STRUCTURAL': 'NONE',
@@ -3493,9 +3566,9 @@ INPUT:
                     + ', '.join(frame.get('ownership_components') or []) + '.'
                 ),
                 'repair_instruction': (
-                    'Split the sentence by evidence ownership. State the current-study fact from the study evidence without an external citation; '
-                    'state a methodological principle separately and cite only a directly supporting METHOD_AUTHORITY source; keep author inference '
-                    'separate, cautious, and uncited unless a source directly supports that exact inference.'
+                    'Split the sentence by evidence ownership. State the current-study fact or study-owned interpretation from the study evidence without an external citation; '
+                    'state a methodological principle separately and cite only a directly supporting METHOD_AUTHORITY source; state prior-literature claims '
+                    'in their own sentence with direct literature support; keep author inference separate and cautious.'
                 ),
             })
         return rows