fina p3

frontend/src/App.tsx

@@ -15,7 +15,7 @@ function App() {
   const logContainerRef = useRef<HTMLDivElement>(null);
   
   // Processing Config State
-  const [modelProvider, setModelProvider] = useState('nrp');
+  const[modelProvider, setModelProvider] = useState('nrp');
   const[apiKey, setApiKey] = useState('');
   const[baseUrl, setBaseUrl] = useState('https://ellm.nrp-nautilus.io/v1'); // NRP Default
   const[modelName, setModelName] = useState('qwen3'); // Default
@@ -23,32 +23,32 @@ function App() {
   const [location, setLocation] = useState('us-central1');
   const[includeComments, setIncludeComments] = useState(false);
   const[reasoningMethod, setReasoningMethod] = useState('cot');
-  const [promptTemplate, setPromptTemplate] = useState('standard');
+  const[promptTemplate, setPromptTemplate] = useState('standard');
   const[customQuery, setCustomQuery] = useState('');
   const [maxRetries, setMaxRetries] = useState(1);
-  const [availablePrompts, setAvailablePrompts] = useState<any[]>([]);
+  const[availablePrompts, setAvailablePrompts] = useState<any[]>([]);
   
-  const [useSearch, setUseSearch] = useState(false);
+  const[useSearch, setUseSearch] = useState(false);
   const[useCode, setUseCode] = useState(false);
 
   // Data States
   const[queueList, setQueueList] = useState<any[]>([]);
   const [selectedQueueItems, setSelectedQueueItems] = useState<Set<string>>(new Set());
-  const [expandedQueueItems, setExpandedQueueItems] = useState<Set<string>>(new Set());
+  const[expandedQueueItems, setExpandedQueueItems] = useState<Set<string>>(new Set());
   const[lastQueueIndex, setLastQueueIndex] = useState<number | null>(null);
   
-  const [singleLinkInput, setSingleLinkInput] = useState(''); 
+  const[singleLinkInput, setSingleLinkInput] = useState(''); 
   const [profileList, setProfileList] = useState<any[]>([]);
   const[selectedProfile, setSelectedProfile] = useState<any>(null);
   const [profilePosts, setProfilePosts] = useState<any[]>([]);
-  const [integrityBoard, setIntegrityBoard] = useState<any[]>([]);
+  const[integrityBoard, setIntegrityBoard] = useState<any[]>([]);
   
   const[datasetList, setDatasetList] = useState<any[]>([]);
   const[selectedItems, setSelectedItems] = useState<Set<string>>(new Set());
   const[lastDatasetIndex, setLastDatasetIndex] = useState<number | null>(null);
 
   const [benchmarks, setBenchmarks] = useState<any>(null);
-  const [leaderboard, setLeaderboard] = useState<any[]>([]); 
+  const[leaderboard, setLeaderboard] = useState<any[]>([]); 
   const[refreshTrigger, setRefreshTrigger] = useState(0);
 
   // Tags
@@ -57,13 +57,13 @@ function App() {
   // Manual Labeling State
   const[manualLink, setManualLink] = useState('');
   const [manualCaption, setManualCaption] = useState('');
-  const [manualTags, setManualTags] = useState('');
+  const[manualTags, setManualTags] = useState('');
   const[manualReasoning, setManualReasoning] = useState('');
   const[manualScores, setManualScores] = useState({
       visual: 5, audio: 5, source: 5, logic: 5, emotion: 5,
       va: 5, vc: 5, ac: 5, final: 50
   });
-  const [showRubric, setShowRubric] = useState(false);
+  const[showRubric, setShowRubric] = useState(false);
   const[aiReference, setAiReference] = useState<any>(null);
   const[labelBrowserMode, setLabelBrowserMode] = useState<'queue' | 'dataset'>('queue');
   const[labelFilter, setLabelFilter] = useState('');
@@ -77,7 +77,7 @@ function App() {
   const [agentConfig, setAgentConfig] = useState({ use_search: true, use_code: false });
 
   // Resampling configuration
-  const [resampleCount, setResampleCount] = useState<number>(1);
+  const[resampleCount, setResampleCount] = useState<number>(1);
   
   // Drag Selection references
   const isDraggingQueueRef = useRef(false);
@@ -85,10 +85,10 @@ function App() {
 
   // Quick Demo State
   const[demoLink, setDemoLink] = useState('');
-  const [demoLogs, setDemoLogs] = useState('');
+  const[demoLogs, setDemoLogs] = useState('');
   const[demoIsProcessing, setDemoIsProcessing] = useState(false);
   const[demoResult, setDemoResult] = useState<any>(null);
-  const [showDemoConfig, setShowDemoConfig] = useState(false);
+  const[showDemoConfig, setShowDemoConfig] = useState(false);
   const demoLogContainerRef = useRef<HTMLDivElement>(null);
 
   useEffect(() => {
@@ -696,8 +696,9 @@ function App() {
   return (
     <div className="flex h-screen w-full bg-[#09090b] text-slate-200 font-sans overflow-hidden">
       <datalist id="modelSuggestions">
-          <option value="gemini-1.5-pro-preview-0409" />
-          <option value="gemini-2.0-flash-exp" />
+          <option value="gemini-1.5-pro" />
+          <option value="gemini-1.5-flash" />
+          <option value="gemini-2.0-flash" />
           <option value="qwen3" />
           <option value="gpt-oss" />
           <option value="kimi" />

src/agent_logic.py

@@ -42,7 +42,7 @@ async def _analyze_video_async(video_url: str, context: str, agent_config: dict)
         api_key = agent_config.get("api_key", os.getenv("GEMINI_API_KEY", ""))
         project_id = agent_config.get("project_id", os.getenv("VERTEX_PROJECT_ID", ""))
         location = agent_config.get("location", os.getenv("VERTEX_LOCATION", "us-central1"))
-        model_name = agent_config.get("model_name", os.getenv("VERTEX_MODEL_NAME", "gemini-1.5-pro-preview-0409"))
+        model_name = agent_config.get("model_name", os.getenv("VERTEX_MODEL_NAME", "gemini-1.5-pro"))
         reasoning_method = agent_config.get("reasoning_method", "cot")
         prompt_template = agent_config.get("prompt_template", "standard")
         
@@ -325,7 +325,7 @@ def create_a2a_app():
                     )
                     
                     if provider == 'vertex' and not project_id:
-                        reply = f"Welcome to the LiarMP4 Agent Nexus!\n\nIt looks like you haven't configured **Vertex AI** yet. Please enter your Google Cloud Project ID in the 'Inference Config' panel on the left, or tell me directly: *'set project id to [YOUR_PROJECT]'*.\n\n{base_capabilities}"
+                        reply = f"Welcome to the LiarMP4 Agent Nexus!\n\nIt looks like you haven't configured **Vertex AI** yet. Please enter your Google Cloud Project ID in the 'Inference Config' panel on the left, or tell me directly: *'set project id to[YOUR_PROJECT]'*.\n\n{base_capabilities}"
                     elif provider == 'gemini' and not api_key:
                         reply = f"👋 Welcome to the LiarMP4 Agent Nexus!\n\nIt looks like you haven't configured **Gemini** yet. Please enter your API Key in the 'Inference Config' panel on the left, or tell me directly: *'set api key to[YOUR_KEY]'*.\n\n{base_capabilities}"
                     else:

src/inference_logic.py

@@ -140,7 +140,9 @@ async def generate_simple_text(prompt: str, model_type: str, config: dict):
         if model_type == 'gemini':
             if genai_legacy is None: return "Error: Legacy SDK missing."
             genai_legacy.configure(api_key=config.get("api_key"))
-            model = genai_legacy.GenerativeModel(config.get("model_name", "models/gemini-2.0-flash-exp"))
+            model_name = config.get("model_name", "gemini-1.5-pro")
+            if not model_name: model_name = "gemini-1.5-pro"
+            model = genai_legacy.GenerativeModel(model_name)
             response = await loop.run_in_executor(
                 None, 
                 lambda: model.generate_content(prompt, generation_config={"temperature": 0.0})
@@ -157,7 +159,7 @@ async def generate_simple_text(prompt: str, model_type: str, config: dict):
             response = await loop.run_in_executor(
                 None,
                 lambda: cl.models.generate_content(
-                    model=config['model_name'],
+                    model=config.get('model_name', 'gemini-1.5-pro'),
                     contents=prompt,
                     config=GenerateContentConfig(temperature=0.0)
                 )
@@ -298,7 +300,10 @@ async def run_gemini_labeling_pipeline(video_path: str, caption: str, transcript
             active_tools.append({"code_execution": {}})
             system_persona += "\n- You MUST use the Code Execution tool for any necessary calculations, data processing, or statistical verifications."
 
-        model = genai_legacy.GenerativeModel("models/gemini-2.0-flash-exp", tools=active_tools if active_tools else None)
+        model_name = gemini_config.get("model_name", "gemini-1.5-pro")
+        if not model_name: model_name = "gemini-1.5-pro"
+        model = genai_legacy.GenerativeModel(model_name, tools=active_tools if active_tools else None)
+        
         toon_schema = SCHEMA_REASONING if include_comments else SCHEMA_SIMPLE
         score_instructions = SCORE_INSTRUCTIONS_REASONING if include_comments else SCORE_INSTRUCTIONS_SIMPLE
         tag_list_text = get_formatted_tag_list()
@@ -337,7 +342,7 @@ async def run_gemini_labeling_pipeline(video_path: str, caption: str, transcript
                     
                     macro_prompt = FCOT_MACRO_PROMPT.format(system_persona=system_persona, caption=caption, transcript=transcript)
                     save_debug_log(request_id, 'prompt', macro_prompt, attempt, 'fcot_macro')
-                    inputs1 = [macro_prompt]
+                    inputs1 =[macro_prompt]
                     if uploaded_file and uploaded_file.state.name != "FAILED": inputs1.insert(0, uploaded_file)
                     res1 = await loop.run_in_executor(None, lambda: chat.send_message(inputs1))
                     macro_hypothesis = res1.text
@@ -406,7 +411,8 @@ async def run_vertex_labeling_pipeline(video_path: str, caption: str, transcript
 
     project_id = vertex_config.get("project_id")
     location = vertex_config.get("location", "us-central1")
-    model_name = vertex_config.get("model_name", "gemini-1.5-pro-preview-0409")
+    model_name = vertex_config.get("model_name", "gemini-1.5-pro")
+    if not model_name: model_name = "gemini-1.5-pro"
     max_retries = int(vertex_config.get("max_retries", 1))
     api_key = vertex_config.get("api_key")
 

src/user_analysis_logic.py

@@ -1,147 +1,147 @@
-import os
-import csv
-import json
-import logging
-import asyncio
-from pathlib import Path
-import inference_logic
-
-# Configure Logging
-logger = logging.getLogger(__name__)
-
-# --- Prompts for User Analysis ---
-
-PROMPT_USER_PROFILING = """
-You are an Expert Intelligence Analyst specializing in Information Integrity and Social Influence Operations.
-
-**TASK:**
-Analyze the following timeline of social media posts from a single user: "@{username}".
-Your goal is to construct a "Credibility & Bias Profile" based on their historical behavior.
-
-**INPUT DATA (Recent Posts):**
-{timeline_text}
-
-**ANALYSIS REQUIREMENTS:**
-1.  **Thematic Clusters:** What subjects does this user repeatedly post about? (e.g., "Crypto", "US Politics", "Climate Skepticism").
-2.  **Echo Chamber Indicators:** Does the user frequently repost specific domains or engage with specific narratives without adding nuance?
-3.  **Emotional Valence:** Analyze the aggregate emotional tone (Alarmist, Neutral, Aggressive, Satirical).
-4.  **Bias Detection:** Identify explicit political or ideological biases based on the text.
-5.  **Credibility Weighting:** Based on the content, assign a "Historical Credibility Score" (0.0 to 1.0).
-    *   0.0 = High frequency of inflammatory/unverified claims.
-    *   1.0 = Consistently neutral or verified sourcing.
-
-**OUTPUT FORMAT (Strict JSON):**
-{{
-  "username": "@{username}",
-  "thematic_clusters": ["Topic A", "Topic B"],
-  "echo_chamber_detected": boolean,
-  "bias_assessment": "Description of bias...",
-  "emotional_valence": "Dominant tone...",
-  "credibility_score": float,
-  "summary_profile": "A concise paragraph summarizing the user's role in the information ecosystem."
-}}
-"""
-
-async def load_user_history(username: str, limit: int = 50) -> str:
-    """
-    Reads the user's history.csv and formats it into a text block for the LLM.
-    """
-    csv_path = Path(f"data/profiles/{username}/history.csv")
-    if not csv_path.exists():
-        return ""
-
-    timeline_entries =[]
-    try:
-        with open(csv_path, 'r', encoding='utf-8', errors='replace') as f:
-            reader = csv.DictReader(f)
-            # Read all, sort by date descending if needed, but scraper usually does desc
-            rows = list(reader)
-            # Take latest 'limit' posts
-            recent_rows = rows[-limit:] 
-            
-            for row in recent_rows:
-                entry = (
-                    f"[{row['timestamp']}] "
-                    f"{'REPOST' if row.get('is_reply')=='True' else 'POST'}: "
-                    f"\"{row['text']}\" "
-                    f"(Likes: {row['metric_likes']}, Views: {row['metric_views']})"
-                )
-                timeline_entries.append(entry)
-    except Exception as e:
-        logger.error(f"Error reading history for {username}: {e}")
-        return ""
-
-    return "\n".join(timeline_entries)
-
-async def generate_user_profile_report(username: str):
-    """
-    Orchestrates the analysis pipeline:
-    1. Load History.
-    2. Construct Prompt.
-    3. Call LLM (using Vertex/Gemini config from environment or default).
-    4. Save JSON Report.
-    """
-    logger.info(f"Starting analysis for user: {username}")
-    
-    timeline_text = await load_user_history(username)
-    if not timeline_text:
-        return {"error": "No history found or empty timeline."}
-
-    # Format Prompt
-    prompt = PROMPT_USER_PROFILING.format(username=username, timeline_text=timeline_text)
-
-    # Use Vertex AI by default if configured, else try Gemini Legacy
-    # For now, we reuse the pipeline functions in inference_logic if available, 
-    # or create a direct call here for simplicity.
-    
-    # We'll assume Vertex is the primary backend for this advanced analysis
-    # This requires valid credentials in the environment or passed config.
-    # Fallback to a placeholder if no model is loaded.
-    
-    report_json = {}
-    
-    try:
-        # Attempt to use the existing Vertex Client in inference_logic if initialized
-        # Otherwise, we instantiate a quick one if env vars exist
-        project_id = os.getenv("VERTEX_PROJECT_ID")
-        location = os.getenv("VERTEX_LOCATION", "us-central1")
-        api_key = os.getenv("VERTEX_API_KEY")
-        
-        if inference_logic.genai and project_id:
-            from google.genai import Client
-            from google.genai.types import GenerateContentConfig
-            
-            if api_key:
-                client = Client(vertexai=True, project=project_id, location=location, api_key=api_key)
-            else:
-                client = Client(vertexai=True, project=project_id, location=location)
-                
-            response = client.models.generate_content(
-                model="gemini-1.5-pro-preview-0409",
-                contents=prompt,
-                config=GenerateContentConfig(response_mime_type="application/json")
-            )
-            report_text = response.text
-            report_json = json.loads(report_text)
-            
-        else:
-            # Fallback Mock for Demo/LITE mode
-            logger.warning("Vertex AI credentials not found. Generating Mock Analysis.")
-            report_json = {
-                "username": f"@{username}",
-                "thematic_clusters":["Simulated Topic 1", "Simulated Topic 2"],
-                "bias_assessment": "System running in LITE mode. Configure Vertex AI for real analysis.",
-                "credibility_score": 0.5,
-                "summary_profile": "Mock profile generated because AI backend is not active."
-            }
-
-    except Exception as e:
-        logger.error(f"LLM Analysis failed: {e}")
-        report_json = {"error": str(e)}
-
-    # Save Report
-    output_path = Path(f"data/profiles/{username}/analysis_report.json")
-    with open(output_path, 'w', encoding='utf-8') as f:
-        json.dump(report_json, f, indent=2)
-
-    return report_json
+import os
+import csv
+import json
+import logging
+import asyncio
+from pathlib import Path
+import inference_logic
+
+# Configure Logging
+logger = logging.getLogger(__name__)
+
+# --- Prompts for User Analysis ---
+
+PROMPT_USER_PROFILING = """
+You are an Expert Intelligence Analyst specializing in Information Integrity and Social Influence Operations.
+
+**TASK:**
+Analyze the following timeline of social media posts from a single user: "@{username}".
+Your goal is to construct a "Credibility & Bias Profile" based on their historical behavior.
+
+**INPUT DATA (Recent Posts):**
+{timeline_text}
+
+**ANALYSIS REQUIREMENTS:**
+1.  **Thematic Clusters:** What subjects does this user repeatedly post about? (e.g., "Crypto", "US Politics", "Climate Skepticism").
+2.  **Echo Chamber Indicators:** Does the user frequently repost specific domains or engage with specific narratives without adding nuance?
+3.  **Emotional Valence:** Analyze the aggregate emotional tone (Alarmist, Neutral, Aggressive, Satirical).
+4.  **Bias Detection:** Identify explicit political or ideological biases based on the text.
+5.  **Credibility Weighting:** Based on the content, assign a "Historical Credibility Score" (0.0 to 1.0).
+    *   0.0 = High frequency of inflammatory/unverified claims.
+    *   1.0 = Consistently neutral or verified sourcing.
+
+**OUTPUT FORMAT (Strict JSON):**
+{{
+  "username": "@{username}",
+  "thematic_clusters": ["Topic A", "Topic B"],
+  "echo_chamber_detected": boolean,
+  "bias_assessment": "Description of bias...",
+  "emotional_valence": "Dominant tone...",
+  "credibility_score": float,
+  "summary_profile": "A concise paragraph summarizing the user's role in the information ecosystem."
+}}
+"""
+
+async def load_user_history(username: str, limit: int = 50) -> str:
+    """
+    Reads the user's history.csv and formats it into a text block for the LLM.
+    """
+    csv_path = Path(f"data/profiles/{username}/history.csv")
+    if not csv_path.exists():
+        return ""
+
+    timeline_entries =[]
+    try:
+        with open(csv_path, 'r', encoding='utf-8', errors='replace') as f:
+            reader = csv.DictReader(f)
+            # Read all, sort by date descending if needed, but scraper usually does desc
+            rows = list(reader)
+            # Take latest 'limit' posts
+            recent_rows = rows[-limit:] 
+            
+            for row in recent_rows:
+                entry = (
+                    f"[{row['timestamp']}] "
+                    f"{'REPOST' if row.get('is_reply')=='True' else 'POST'}: "
+                    f"\"{row['text']}\" "
+                    f"(Likes: {row['metric_likes']}, Views: {row['metric_views']})"
+                )
+                timeline_entries.append(entry)
+    except Exception as e:
+        logger.error(f"Error reading history for {username}: {e}")
+        return ""
+
+    return "\n".join(timeline_entries)
+
+async def generate_user_profile_report(username: str):
+    """
+    Orchestrates the analysis pipeline:
+    1. Load History.
+    2. Construct Prompt.
+    3. Call LLM (using Vertex/Gemini config from environment or default).
+    4. Save JSON Report.
+    """
+    logger.info(f"Starting analysis for user: {username}")
+    
+    timeline_text = await load_user_history(username)
+    if not timeline_text:
+        return {"error": "No history found or empty timeline."}
+
+    # Format Prompt
+    prompt = PROMPT_USER_PROFILING.format(username=username, timeline_text=timeline_text)
+
+    # Use Vertex AI by default if configured, else try Gemini Legacy
+    # For now, we reuse the pipeline functions in inference_logic if available, 
+    # or create a direct call here for simplicity.
+    
+    # We'll assume Vertex is the primary backend for this advanced analysis
+    # This requires valid credentials in the environment or passed config.
+    # Fallback to a placeholder if no model is loaded.
+    
+    report_json = {}
+    
+    try:
+        # Attempt to use the existing Vertex Client in inference_logic if initialized
+        # Otherwise, we instantiate a quick one if env vars exist
+        project_id = os.getenv("VERTEX_PROJECT_ID")
+        location = os.getenv("VERTEX_LOCATION", "us-central1")
+        api_key = os.getenv("VERTEX_API_KEY")
+        
+        if inference_logic.genai and project_id:
+            from google.genai import Client
+            from google.genai.types import GenerateContentConfig
+            
+            if api_key:
+                client = Client(vertexai=True, project=project_id, location=location, api_key=api_key)
+            else:
+                client = Client(vertexai=True, project=project_id, location=location)
+                
+            response = client.models.generate_content(
+                model="gemini-1.5-pro",
+                contents=prompt,
+                config=GenerateContentConfig(response_mime_type="application/json")
+            )
+            report_text = response.text
+            report_json = json.loads(report_text)
+            
+        else:
+            # Fallback Mock for Demo/LITE mode
+            logger.warning("Vertex AI credentials not found. Generating Mock Analysis.")
+            report_json = {
+                "username": f"@{username}",
+                "thematic_clusters":["Simulated Topic 1", "Simulated Topic 2"],
+                "bias_assessment": "System running in LITE mode. Configure Vertex AI for real analysis.",
+                "credibility_score": 0.5,
+                "summary_profile": "Mock profile generated because AI backend is not active."
+            }
+
+    except Exception as e:
+        logger.error(f"LLM Analysis failed: {e}")
+        report_json = {"error": str(e)}
+
+    # Save Report
+    output_path = Path(f"data/profiles/{username}/analysis_report.json")
+    with open(output_path, 'w', encoding='utf-8') as f:
+        json.dump(report_json, f, indent=2)
+
+    return report_json