osintengine 1.0.3__tar.gz → 1.0.4__tar.gz

{osintengine-1.0.3 → osintengine-1.0.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: osintengine
-Version: 1.0.3
+Version: 1.0.4
 Summary: The OSINT investigation engine — 7-phase pipeline, persistent knowledge graph, community intelligence
 Author-email: Lorenzo Baron <baron.lorenzo99@gmail.com>
 License-Expression: Apache-2.0

{osintengine-1.0.3 → osintengine-1.0.4}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "osintengine"
-version = "1.0.3"
+version = "1.0.4"
 description = "The OSINT investigation engine — 7-phase pipeline, persistent knowledge graph, community intelligence"
 readme = "README.md"
 license = "Apache-2.0"

{osintengine-1.0.3 → osintengine-1.0.4}/src/watson/__init__.py

@@ -7,4 +7,4 @@ must be the truth." — Sherlock Holmes
 Autonomous open-source intelligence with 338 Bellingcat tools.
 Knowledge graph, lead tracking, recursive investigation.
 """
-__version__ = "1.0.3"
+__version__ = "1.0.4"

{osintengine-1.0.3 → osintengine-1.0.4}/src/watson/orchestration/engine.py

@@ -3616,7 +3616,9 @@ Examples:
         except Exception as e:
             logger.warning("resolution_failed: %s", e)
 
-        # Synthesis
+        # Synthesis — filter noise before feeding to LLM
+        findings = self._filter_quality(findings, query=query)
+        
         try:
             from .synthesis import synthesize_brief
             from .llm_config import call_llm

{osintengine-1.0.3 → osintengine-1.0.4}/src/watson/orchestration/synthesis.py

@@ -560,37 +560,45 @@ def _fallback_brief(query: str, findings: list) -> dict:
         "Financial": ["revenue", "profit", "loss", "billion", "million", "funding", "valuation",
                       "stock", "share", "investor", "ipo", "acquisition"],
         "Crime / Investigation": ["criminal", "arrest", "prison", "jail", "fraud", "money laundering",
-                                   "investigation", "probe", "allegation", "misconduct"],
-        "Infrastructure / Technical": ["dns", "ip address", "nameserver", "hosting", "cdn",
-                                        "subdomain", "resolves_to", "mx record"],
+                                   "investigation", "probe", "allegation", "misconduct", "wanted", "fugitive",
+                                   "fbi", "hacker", "cyber"],
     }
     themes = []
     all_text = " ".join(titles + [getattr(f, "description", "") or "" for f in findings]).lower()
     for theme, keywords in theme_keywords.items():
         matches = [kw for kw in keywords if kw in all_text]
         if matches:
-            # Extract a relevant finding snippet
+            # Find the best finding for this theme — prefer ones with real descriptions
+            best_summary = f"Evidence of {matches[0]} found across multiple sources."
+            best_source = ""
             for f in findings:
-                desc = (getattr(f, "description", "") or "").lower()
+                desc = (getattr(f, "description", "") or "")
                 title_l = (getattr(f, "title", "") or "").lower()
                 for kw in matches:
-                    if kw in title_l or kw in desc:
-                        themes.append({
-                            "theme": theme,
-                            "severity": "HIGH" if theme.startswith("Crime") else "MEDIUM",
-                            "summary": getattr(f, "title", "")[:200] or f"Evidence of {kw} found",
-                            "source_titles": [getattr(f, "title", "")[:100]],
-                        })
+                    if kw in title_l or kw in desc.lower():
+                        # Use the description as summary (actual intelligence, not just title)
+                        if desc and len(desc) > 20:
+                            best_summary = desc[:300]
+                            best_source = getattr(f, "source_url", "") or ""
                         break
-                if any(t["theme"] == theme for t in themes):
+                if best_source:
                     break
-            else:
-                themes.append({
-                    "theme": theme,
-                    "severity": "MEDIUM",
-                    "summary": f"Keywords detected: {', '.join(matches[:4])}",
-                    "source_titles": [],
-                })
+            themes.append({
+                "theme": theme,
+                "severity": "HIGH" if theme.startswith("Crime") else "MEDIUM",
+                "summary": best_summary,
+                "source_titles": [best_source] if best_source else [],
+            })
+    
+    # Deduplicate themes — same summary shouldn't appear for multiple themes
+    seen_summaries = set()
+    deduped_themes = []
+    for t in themes:
+        key = t["summary"][:50]
+        if key not in seen_summaries:
+            seen_summaries.add(key)
+            deduped_themes.append(t)
+    themes = deduped_themes[:5]
     
     # Extract timeline events from findings (dates in titles/descriptions)
     timeline_events: list[dict] = []