Mise à jour: routes d'analyse et moteur k2

app/api/routes/analysis.py

@@ -234,6 +234,124 @@ def _normalize_k2_result_for_frontend(result_dict: dict) -> dict:
     Ensures all required fields are present with proper defaults.
     """
     from app.core.logging import logger
+
+    def _first_present(source: dict, keys: list, default=None):
+        for key in keys:
+            value = source.get(key)
+            if value not in (None, "", [], {}):
+                return value
+        return default
+
+    def _as_percent(value, default=80):
+        try:
+            score = float(value)
+            return int(round(score * 100)) if score <= 1 else int(round(score))
+        except (TypeError, ValueError):
+            return default
+
+    def _normalize_contradiction(item):
+        if not isinstance(item, dict):
+            return {"description": str(item), "resolution": "Investigation required", "confidence": 0.8}
+
+        topic = _first_present(item, ["topic", "variable", "title"], "Scientific conflict")
+        conflict = _first_present(item, ["conflict", "description", "explanation", "impact"])
+        statement_a = _first_present(item, ["statement_a", "finding_a"])
+        statement_b = _first_present(item, ["statement_b", "finding_b"])
+
+        if not conflict and (statement_a or statement_b):
+            conflict = " vs. ".join([p for p in [statement_a, statement_b] if p])
+
+        description = conflict or topic or "Scientific contradiction detected"
+        resolution = _first_present(item, ["resolution_path", "resolution", "suggested_resolution"], "Investigation required")
+        confidence = _first_present(item, ["confidence", "confidence_score"], 0.8)
+
+        normalized = dict(item)
+        normalized.setdefault("topic", topic)
+        normalized.setdefault("title", topic)
+        normalized.setdefault("description", description)
+        normalized.setdefault("conflict", description)
+        normalized.setdefault("resolution", resolution)
+        normalized.setdefault("resolution_path", resolution)
+        normalized.setdefault("confidence", confidence)
+        normalized.setdefault("impact_score", _as_percent(confidence))
+        return normalized
+
+    def _normalize_gap(item):
+        if not isinstance(item, dict):
+            return {"description": str(item), "gap_description": str(item), "importance_score": 0.8}
+
+        description = _first_present(
+            item,
+            ["gap_description", "description", "opportunity", "title"],
+            "Research opportunity identified"
+        )
+        suggested = _first_present(
+            item,
+            ["suggested_investigation", "suggested_direction", "recommendation", "next_step"],
+            "Design a targeted follow-up investigation."
+        )
+        score = _first_present(item, ["importance_score", "impact_score", "score"], 0.8)
+
+        normalized = dict(item)
+        normalized.setdefault("gap_description", description)
+        normalized.setdefault("description", description)
+        normalized.setdefault("title", description)
+        normalized.setdefault("suggested_investigation", suggested)
+        normalized.setdefault("suggested_direction", suggested)
+        normalized.setdefault("importance_score", score)
+        normalized.setdefault("impact_score", _as_percent(score))
+        normalized.setdefault("related_variables", [])
+        normalized.setdefault("source_documents", [])
+        return normalized
+
+    def _normalize_divergence(item):
+        if not isinstance(item, dict):
+            return {"description": str(item)}
+
+        topic = _first_present(item, ["topic", "variable", "title"], "Scientific divergence")
+        description = _first_present(
+            item,
+            ["description", "impact", "conflict", "finding_a", "finding_b"],
+            topic
+        )
+
+        normalized = dict(item)
+        normalized.setdefault("topic", topic)
+        normalized.setdefault("title", topic)
+        normalized.setdefault("description", description)
+        return normalized
+
+    def _is_placeholder_summary(value):
+        if not isinstance(value, str):
+            return True
+        return value.strip().lower() in {
+            "",
+            "analysis completed.",
+            "analysis completed",
+            "analysis completed successfully",
+        }
+
+    def _build_summary_from_result(data):
+        comp = data.get("comparative_analysis") or {}
+        parts = []
+
+        findings = comp.get("common_findings") or []
+        if findings:
+            parts.append("Key findings: " + " ".join(str(item) for item in findings[:3]))
+
+        contradictions = comp.get("contradictions") or []
+        if contradictions:
+            descriptions = [
+                item.get("description") if isinstance(item, dict) else str(item)
+                for item in contradictions[:2]
+            ]
+            parts.append("Main conflicts: " + " ".join(d for d in descriptions if d))
+
+        recommendations = data.get("strategic_recommendations") or []
+        if recommendations:
+            parts.append("Recommended strategy: " + " ".join(str(item) for item in recommendations[:2]))
+
+        return "\n\n".join(parts) if parts else None
     
     logger.info(f"NORMALIZE: Input result keys: {list(result_dict.keys())}")
     logger.info(f"NORMALIZE: Status is: {result_dict.get('status')}")
@@ -277,6 +395,19 @@ def _normalize_k2_result_for_frontend(result_dict: dict) -> dict:
             
         comp_analysis.setdefault('research_gaps', result_dict.get('research_gaps', []))
         comp_analysis.setdefault('confidence_score', result_dict.get('confidence_overall', 0.8))
+
+        comp_analysis['contradictions'] = [
+            _normalize_contradiction(item) for item in comp_analysis.get('contradictions', [])
+        ]
+        comp_analysis['divergences'] = [
+            _normalize_divergence(item) for item in comp_analysis.get('divergences', [])
+        ]
+        comp_analysis['research_gaps'] = [
+            _normalize_gap(item) for item in comp_analysis.get('research_gaps', [])
+        ]
+        result_dict['research_gaps'] = [
+            _normalize_gap(item) for item in (result_dict.get('research_gaps') or comp_analysis.get('research_gaps', []))
+        ]
     
     # Ensure other required fields exist
     result_dict.setdefault('research_gaps', [])
@@ -296,6 +427,11 @@ def _normalize_k2_result_for_frontend(result_dict: dict) -> dict:
         result_dict['comparative_analysis'].setdefault('confidence_score', result_dict['confidence_overall'])
         if result_dict['comparative_analysis'].get('confidence_score') == 0:
             result_dict['comparative_analysis']['confidence_score'] = result_dict['confidence_overall']
+
+    if _is_placeholder_summary(result_dict.get('reasoning_summary')):
+        rebuilt_summary = _build_summary_from_result(result_dict)
+        if rebuilt_summary:
+            result_dict['reasoning_summary'] = rebuilt_summary
     
     # Ensure arrays are properly formatted
     for field in ['research_gaps', 'counter_hypotheses', 'strategic_recommendations', 'reasoning_trace']:

app/services/k2_think_engine.py

@@ -207,12 +207,12 @@ class K2ThinkEngine:
 {{
   "comparative_analysis": {{
     "document_ids": ["doc_id1", "doc_id2"],
-    "divergences": [{{"topic": "description"}}],
-    "contradictions": [{{"topic": "...", "conflict": "...", "resolution_path": "...", "confidence": 0.8}}],
+    "divergences": [{{"topic": "...", "description": "...", "impact": "..."}}],
+    "contradictions": [{{"topic": "...", "description": "...", "conflict": "...", "resolution": "...", "resolution_path": "...", "confidence": 0.8, "impact_score": 80}}],
     "common_findings": ["finding1"],
     "confidence_score": 0.9
   }},
-  "research_gaps": [{{"gap_description": "...", "importance_score": 0.8, "related_variables": [], "suggested_investigation": "...", "source_documents": ["doc_id1"]}}],
+  "research_gaps": [{{"title": "...", "description": "...", "gap_description": "...", "importance_score": 0.8, "impact_score": 80, "related_variables": [], "suggested_investigation": "...", "suggested_direction": "...", "source_documents": ["doc_id1"]}}],
   "counter_hypotheses": [{{"hypothesis": "...", "rationale": "...", "potential_bias": "...", "validation_experiment": "...", "confidence_against": 0.8}}],
   "proposed_protocol": {{
     "title": "...", "objective": "...", "steps": [{{"step_number": 1, "description": "...", "materials": [], "critical_parameters": []}}],
@@ -227,6 +227,7 @@ class K2ThinkEngine:
 - NO preamble, NO explanations before or after JSON.
 - KEEP REASONING CONCISE: Focus on direct analysis to stay within processing time limits.
 - Use valid citations e.g. (Author, Year).
+- Avoid placeholder text. Every contradiction and research gap must include a display-ready description.
 """
 
             # 3. Appel au modèle via ChatOpenAI (manual JSON parsing)