Upload folder using huggingface_hub

frontend/style.css

@@ -58,7 +58,7 @@ body {
   display: flex;
   align-items: center;
   gap: 12px;
-  padding: 20px 18px 16px;
+  padding: 16px 16px 12px;
   border-bottom: 1px solid var(--border);
 }
 
@@ -94,21 +94,21 @@ body {
 }
 
 .new-chat-btn {
-  margin: 14px 14px 10px;
-  padding: 10px 14px;
+  margin: 12px 12px 8px;
+  padding: 8px 12px;
   background: var(--gold-glow);
   border: 1px solid var(--border-gold);
-  border-radius: 9px;
+  border-radius: 8px;
   color: var(--gold);
   font-family: 'DM Sans', sans-serif;
-  font-size: 13px;
+  font-size: 12px;
   font-weight: 500;
   cursor: pointer;
   display: flex;
   align-items: center;
-  gap: 8px;
+  gap: 6px;
   transition: background var(--transition), border-color var(--transition);
-  width: calc(100% - 28px);
+  width: calc(100% - 24px);
 }
 
 .new-chat-btn:hover {
@@ -119,8 +119,8 @@ body {
 .new-chat-icon { font-size: 16px; font-weight: 300; }
 
 .sidebar-section-label {
-  padding: 10px 18px 6px;
-  font-size: 10px;
+  padding: 8px 16px 4px;
+  font-size: 9px;
   font-weight: 600;
   color: var(--text-3);
   letter-spacing: 1px;
@@ -189,28 +189,28 @@ body {
 }
 
 .sidebar-footer {
-  padding: 14px 16px;
+  padding: 10px 14px;
   border-top: 1px solid var(--border);
   display: flex;
   flex-direction: column;
-  gap: 8px;
+  gap: 6px;
 }
 
 .footer-disclaimer {
   display: flex;
   gap: 8px;
   align-items: flex-start;
-  font-size: 11px;
+  font-size: 10px;
   color: var(--text-3);
-  line-height: 1.4;
+  line-height: 1.3;
 }
 
 .disclaimer-icon { color: var(--gold-dim); flex-shrink: 0; }
 
 .footer-meta {
-  font-size: 10px;
+  font-size: 9px;
   color: var(--text-3);
-  line-height: 1.5;
+  line-height: 1.4;
 }
 
 /* ── Main wrapper ── */
@@ -229,7 +229,7 @@ body {
   display: flex;
   align-items: center;
   justify-content: space-between;
-  padding: 0 24px;
+  padding: 0 20px;
   flex-shrink: 0;
   background: var(--navy);
 }
@@ -282,6 +282,7 @@ body {
   display: none;
   flex: 1;
   overflow: hidden;
+  position: relative;
 }
 
 .screen.active { display: flex; }
@@ -290,12 +291,12 @@ body {
 .screen-welcome {
   align-items: center;
   justify-content: center;
-  padding: 40px 24px 120px;
+  padding: 20px 24px;
   flex-direction: column;
 }
 
 .welcome-inner {
-  max-width: 640px;
+  max-width: 720px;
   width: 100%;
   text-align: center;
   animation: fadeUp 0.5s ease;
@@ -307,44 +308,44 @@ body {
 }
 
 .welcome-emblem {
-  font-size: 48px;
-  margin-bottom: 20px;
+  font-size: 40px;
+  margin-bottom: 12px;
   opacity: 0.7;
 }
 
 .welcome-heading {
   font-family: 'Cormorant Garamond', serif;
-  font-size: 52px;
+  font-size: 42px;
   font-weight: 700;
   color: var(--text-1);
-  margin-bottom: 14px;
+  margin-bottom: 10px;
   line-height: 1.1;
 }
 
 .welcome-body {
-  font-size: 15px;
+  font-size: 14px;
   color: var(--text-2);
-  line-height: 1.7;
-  margin-bottom: 40px;
+  line-height: 1.6;
+  margin-bottom: 24px;
 }
 
 .suggestion-grid {
   display: grid;
-  grid-template-columns: 1fr 1fr;
-  gap: 10px;
+  grid-template-columns: 1fr 1fr 1fr;
+  gap: 8px;
 }
 
 .suggestion-pill {
   background: var(--navy-2);
   border: 1px solid var(--border);
   border-radius: 10px;
-  padding: 13px 16px;
+  padding: 10px 12px;
   font-family: 'DM Sans', sans-serif;
-  font-size: 13px;
+  font-size: 12px;
   color: var(--text-2);
   cursor: pointer;
   text-align: left;
-  line-height: 1.5;
+  line-height: 1.4;
   transition: all var(--transition);
 }
 
@@ -361,7 +362,8 @@ body {
 .messages-container {
   flex: 1;
   overflow-y: auto;
-  padding: 32px 0 16px;
+  padding: 16px 0 12px;
+  max-height: calc(100vh - var(--topbar-h) - var(--input-h) - 40px);
 }
 
 .messages-container::-webkit-scrollbar { width: 4px; }
@@ -372,12 +374,12 @@ body {
   margin: 0 auto;
   padding: 0 24px;
   display: flex;
-  flex-direction: column;
+  flex-d0rection: column;
   gap: 28px;
 }
 
 /* ── Message bubbles ── */
-.msg { display: flex; flex-direction: column; gap: 8px; animation: fadeUp 0.3s ease; }
+.msg { display: flex; flex-direction: column; gap: 6px; animation: fadeUp 0.3s ease; }
 
 .msg-user { align-items: flex-end; }
 .msg-ai   { align-items: flex-start; }
@@ -386,11 +388,11 @@ body {
   background: var(--navy-4);
   border: 1px solid var(--border);
   border-radius: 16px 16px 4px 16px;
-  padding: 12px 16px;
+  padding: 10px 14px;
   max-width: 72%;
   color: var(--text-1);
-  font-size: 14px;
-  line-height: 1.6;
+  font-size: 13px;
+  line-height: 1.5;
 }
 
 .bubble-ai {
@@ -398,15 +400,15 @@ body {
   border: 1px solid var(--border);
   border-left: 3px solid var(--gold);
   border-radius: 4px 16px 16px 16px;
-  padding: 18px 20px;
+  padding: 14px 18px;
   max-width: 88%;
   color: var(--text-1);
-  font-size: 14px;
-  line-height: 1.75;
+  font-size: 13px;
+  line-height: 1.6;
   position: relative;
 }
 
-.bubble-ai p { margin-bottom: 12px; }
+.bubble-ai p { margin-bottom: 10px; }
 .bubble-ai p:last-child { margin-bottom: 0; }
 
 /* AI bubble meta row */
@@ -415,18 +417,18 @@ body {
   align-items: center;
   gap: 10px;
   flex-wrap: wrap;
-  margin-top: 14px;
-  padding-top: 12px;
+  margin-top: 10px;
+  padding-top: 8px;
   border-top: 1px solid var(--border);
 }
 
 .verify-badge {
   display: inline-flex;
   align-items: center;
-  gap: 5px;
-  font-size: 11px;
+  gap: 4px;
+  font-size: 10px;
   font-weight: 600;
-  padding: 3px 10px;
+  padding: 2px 8px;
   border-radius: 20px;
 }
 
@@ -445,14 +447,14 @@ body {
 .sources-btn {
   display: inline-flex;
   align-items: center;
-  gap: 6px;
-  font-size: 11px;
+  gap: 5px;
+  font-size: 10px;
   font-weight: 500;
   color: var(--gold);
   background: var(--gold-glow);
   border: 1px solid var(--border-gold);
   border-radius: 20px;
-  padding: 3px 12px;
+  padding: 2px 10px;
   cursor: pointer;
   transition: background var(--transition);
 }
@@ -460,15 +462,15 @@ body {
 .sources-btn:hover { background: rgba(200,168,75,0.22); }
 
 .latency-label {
-  font-size: 11px;
+  font-size: 10px;
   color: var(--text-3);
   margin-left: auto;
 }
 
 .truncated-note {
-  font-size: 11px;
+  font-size: 10px;
   color: var(--text-3);
-  margin-top: 8px;
+  margin-top: 6px;
   font-style: italic;
 }
 
@@ -476,10 +478,10 @@ body {
 .bubble-loading {
   display: flex;
   align-items: center;
-  gap: 12px;
+  gap: 10px;
   color: var(--text-3);
-  font-size: 13px;
-  padding: 14px 18px;
+  font-size: 12px;
+  padding: 10px 14px;
 }
 
 .dots { display: flex; gap: 4px; }
@@ -629,19 +631,23 @@ body {
 
 /* ── Input zone ── */
 .input-zone {
-  padding: 12px 24px 14px;
+  padding: 8px 20px 10px;
   background: linear-gradient(transparent, var(--navy) 30%);
   flex-shrink: 0;
+  height: var(--input-h);
+  display: flex;
+  flex-direction: column;
+  justify-content: flex-end;
 }
 
 .input-box {
   display: flex;
   align-items: flex-end;
-  gap: 10px;
+  gap: 8px;
   background: var(--navy-2);
   border: 1px solid var(--border);
-  border-radius: 14px;
-  padding: 10px 10px 10px 16px;
+  border-radius: 12px;
+  padding: 8px 10px 8px 14px;
   transition: border-color var(--transition);
 }
 
@@ -664,11 +670,11 @@ body {
 .query-textarea::placeholder { color: var(--text-3); }
 
 .send-btn {
-  width: 38px;
-  height: 38px;
+  width: 36px;
+  height: 36px;
   background: var(--gold);
   border: none;
-  border-radius: 9px;
+  border-radius: 8px;
   color: var(--navy);
   cursor: pointer;
   display: flex;
@@ -683,17 +689,18 @@ body {
 
 .input-disclaimer {
   text-align: center;
-  font-size: 11px;
+  font-size: 10px;
   color: var(--text-3);
-  margin-top: 8px;
+  margin-top: 4px;
+  line-height: 1.3;
 }
 
 /* ── Answer formatting ── */
 .bubble-ai ol, .bubble-ai ul {
-  margin: 10px 0 10px 20px;
+  margin: 8px 0 8px 18px;
   display: flex;
   flex-direction: column;
-  gap: 6px;
+  gap: 4px;
 }
 
 .bubble-ai ol { list-style: decimal; }
@@ -701,19 +708,19 @@ body {
 
 .bubble-ai li {
   color: var(--text-1);
-  line-height: 1.6;
-  padding-left: 4px;
+  line-height: 1.5;
+  padding-left: 2px;
 }
 
 .bubble-ai h1, .bubble-ai h2, .bubble-ai h3 {
   font-family: 'Cormorant Garamond', serif;
   color: var(--gold);
-  margin: 16px 0 8px;
+  margin: 12px 0 6px;
 }
 
-.bubble-ai h1 { font-size: 20px; }
-.bubble-ai h2 { font-size: 17px; }
-.bubble-ai h3 { font-size: 15px; }
+.bubble-ai h1 { font-size: 18px; }
+.bubble-ai h2 { font-size: 15px; }
+.bubble-ai h3 { font-size: 13px; }
 
 .bubble-ai strong { color: var(--text-1); font-weight: 600; }
 .bubble-ai em { color: var(--text-2); font-style: italic; }
@@ -730,15 +737,15 @@ body {
 .answer-table {
   width: 100%;
   border-collapse: collapse;
-  margin: 12px 0;
-  font-size: 13px;
+  margin: 10px 0;
+  font-size: 12px;
 }
 
 .answer-table td {
-  padding: 8px 12px;
+  padding: 6px 10px;
   border: 1px solid var(--border);
   color: var(--text-2);
-  line-height: 1.5;
+  line-height: 1.4;
 }
 
 .answer-table tr:first-child td {
@@ -773,45 +780,75 @@ body {
 }
 .analytics-btn:hover {
   background: var(--navy-3);
-  color: var(--text-1);
-}
-
-.screen-analytics {
-  padding: 32px;
+  color: va20px 24px;
   overflow-y: auto;
   height: 100%;
 }
 .analytics-inner {
-  max-width: 800px;
+  max-width: 900px;
   margin: 0 auto;
 }
 .analytics-header h2 {
   font-family: 'Cormorant Garamond', serif;
-  font-size: 28px;
+  font-size: 24px;
   margin: 0 0 4px;
 }
 .analytics-header p {
   color: var(--text-2);
-  font-size: 14px;
-  margin: 0 0 32px;
+  font-size: 13px;
+  margin: 0 0 16px;
 }
 
 .analytics-grid {
   display: grid;
-  grid-template-columns: repeat(auto-fit, minmax(140px, 1fr));
-  gap: 16px;
-  margin-bottom: 32px;
+  grid-template-columns: repeat(5, 1fr);
+  gap: 12px;
+  margin-bottom: 20px;
 }
 .stat-card {
   background: var(--navy-2);
   border: 1px solid var(--border);
-  border-radius: 12px;
-  padding: 20px 16px;
+  border-radius: 10px;
+  padding: 14px 12px;
   text-align: center;
 }
 .stat-value {
-  font-size: 28px;
+  font-size: 22px;
+  font-weight: 600;
+}
+.stat-label {
+  font-size: 11px;
+  color: var(--text-3);
+  margin-top: 6px;
+}
+
+.analytics-charts {
+  display: grid;
+  grid-template-columns: repeat(3, 1fr);
+  gap: 12px;
+  margin-bottom: 16px;
+}
+
+.chart-card {
+  background: var(--navy-2);
+  border: 1px solid var(--border);
+  border-radius: 10px;
+  padding: 14px;
+}
+
+.chart-card h3 {
+  font-family: 'Cormorant Garamond', serif;
+  font-size: 14px;
   font-weight: 600;
+  color: var(--text-1);
+  margin-bottom: 10px;
+}
+
+.chart-container {
+  min-height: 140px;
+  display: flex;
+  align-items: center;
+  justify-content: center;
 }
 
 /* ── Responsive Improvements ── */
@@ -821,8 +858,9 @@ html {
 
 @media (max-width: 768px) {
   :root {
-    --sidebar-w: 208px;
+    --sidebar-w: 200px;
     --topbar-h: 48px;
+    --input-h: 80px;
   }
 
   .app-layout {
@@ -832,7 +870,7 @@ html {
   .sidebar {
     width: 100% !important;
     height: auto;
-    max-height: 40vh;
+    max-height: 35vh;
     overflow-y: auto;
     border-right: none;
     border-bottom: 1px solid var(--border);
@@ -840,11 +878,11 @@ html {
 
   .main-wrapper {
     flex: 1;
-    overflow-y: auto;
+    overflow-y: hidden;
   }
 
   .topbar-title {
-    font-size: 14px;
+    font-size: 13px;
     max-width: 60%;
   }
 
@@ -854,19 +892,46 @@ html {
   }
 
   .screen-welcome {
-    padding: 20px 16px 60px !important;
+    padding: 18px 16px 12px !important;
+  }
+
+  .welcome-heading {
+    font-size: 32px !important;
+  }
+
+  .welcome-body {
+    font-size: 13px !important;
+    margin-bottom: 16px !important;
+  }
+
+  .suggestion-grid {
+    grid-template-columns: 1fr !important;
+    gap: 6px !important;
+  }
+
+  .suggestion-pill {
+    font-size: 11px !important;
+    padding: 8px 10px !important;
   }
 
-  .welcome-inner {
-    max-width: 100% !important;
+  .messages-list {
+    gap: 12px !important;
   }
 
-  .chat-bubble {
+  .bubble-ai,
+  .bubble-user {
     max-width: 95% !important;
+    font-size: 12px !important;
+    padding: 10px 12px !important;
   }
 
-  .msg-text {
-    font-size: 13px;
+  .analytics-grid {
+    grid-template-columns: repeat(auto-fit, minmax(100px, 1fr)) !important;
+    gap: 8px !important;
+  }
+
+  .analytics-charts {
+    grid-template-columns: 1fr !important;
   }
 }
 

src/court/opposing.py

@@ -98,41 +98,41 @@ def build_opposing_prompt(
     Build the messages list for opposing counsel LLM call.
     
     The opposing counsel sees:
-    - Case brief (user's research, gaps)
+    - Full case brief (preserved throughout)
+    - All documents filed (critical!)
+    - All concessions made (critical!)
     - Full recent transcript
     - User's latest argument
     - Retrieved precedents to use against user
     - Any detected trap opportunities
-    - All concessions made so far
+    - Trap history
     """
     difficulty = session.get("difficulty", "standard")
     difficulty_modifier = DIFFICULTY_MODIFIERS.get(difficulty, DIFFICULTY_MODIFIERS["standard"])
     
-    case_brief = session.get("case_brief", "")
-    concessions = _format_concessions(session.get("concessions", []))
-    inconsistencies = _detect_inconsistencies(session.get("user_arguments", []))
-    transcript_recent = _get_recent_transcript(session, last_n=4)
+    # The retrieved_context now contains EVERYTHING from _build_full_context
+    # including case brief, documents, concessions, and precedents
+    # This is the complete informational context
     
     trap_instruction = ""
     if trap_opportunity:
-        trap_instruction = f"\nTRAP OPPORTUNITY DETECTED: {trap_opportunity}\nConsider exploiting this in your response."
-    elif inconsistencies:
-        trap_instruction = f"\nINCONSISTENCY DETECTED: {inconsistencies}\nConsider using the inconsistency trap."
+        trap_instruction = f"\n🎯 TRAP OPPORTUNITY DETECTED: {trap_opportunity}\nConsider exploiting this in your response."
     
-    user_content = f"""CASE BRIEF (your preparation before court):
-{case_brief[:1500]}
+    user_content = f"""COMPLETE SESSION CONTEXT (use all of this):
+{retrieved_context[:4000]}
 
-RECENT TRANSCRIPT:
-{transcript_recent}
-
-{concessions}
-
-RETRIEVED LEGAL AUTHORITIES (use these against the user):
-{retrieved_context[:2000] if retrieved_context else "Use your general legal knowledge."}
-
-USER'S LATEST ARGUMENT:
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+USER'S LATEST ARGUMENT (respond to this specifically):
 {user_argument}
 
+━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+CRITICAL REMINDERS:
+1. You have full access to all documents filed by the user
+2. You have the complete history of the user's arguments
+3. You have all concessions made — exploit them mercilessly
+4. You have seen all trap attempts before — use what worked
+5. You know their case brief — you know their weaknesses
+
 Round {session.get('current_round', 1)} of {session.get('max_rounds', 5)}.
 {trap_instruction}
 
@@ -234,18 +234,95 @@ Deliver your closing argument."""
     ]
 
 
+def detect_trap_opportunity_llm(
+    user_argument: str,
+    session: Dict,
+) -> Optional[Tuple[str, str]]:
+    """
+    Use LLM to detect trap opportunities semantically.
+    More sophisticated than keyword matching — understands legal logic.
+    Falls back to keyword detection if LLM fails.
+    
+    Returns (trap_type, description) or None.
+    """
+    try:
+        from src.llm import call_llm_raw
+        import json
+        import re
+        
+        # Build minimal context for trap analysis
+        case_brief = session.get("case_brief", "")[:500]
+        concessions = session.get("concessions", [])
+        user_args = session.get("user_arguments", [])
+        
+        concessions_text = ""
+        if concessions:
+            concessions_text = "Previous concessions by user: " + "; ".join(
+                [c.get("exact_quote", "")[:60] for c in concessions[-3:]]
+            )
+        
+        previous_args_text = ""
+        if len(user_args) >= 2:
+            previous_args_text = "User argued in Round 1: " + user_args[0].get("text", "")[:150]
+        
+        system_prompt = """You are a legal trap detector for a moot court simulation.
+Analyze if the user's argument contains a trap opportunity for opposing counsel.
+
+Return ONLY valid JSON:
+{
+  "trap_found": true/false,
+  "trap_type": "admission_trap|precedent_trap|inconsistency_trap|none",
+  "description": "brief description of the trap"
+}
+
+Trap types:
+- admission_trap: User made an absolute claim (e.g., "no exceptions", "unlimited", "cannot be restricted")
+- precedent_trap: User cited a case that actually supports opposition when read carefully
+- inconsistency_trap: User contradicted their own previous argument
+"""
+        
+        user_prompt = f"""Case brief: {case_brief}
+
+{concessions_text}
+
+{previous_args_text}
+
+User's latest argument: {user_argument}
+
+Detect trap opportunity. Return ONLY JSON."""
+        
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt}
+        ]
+        
+        response = call_llm_raw(messages)
+        
+        # Extract JSON from response
+        match = re.search(r'\{.*\}', response, re.DOTALL)
+        if match:
+            data = json.loads(match.group())
+            if data.get("trap_found"):
+                trap_type = data.get("trap_type", "admission_trap")
+                if trap_type != "none":
+                    return (trap_type, data.get("description", "Trap detected"))
+    
+    except Exception as e:
+        logger.debug(f"LLM trap detection failed, falling back to keyword: {e}")
+    
+    # Fall back to keyword-based detection
+    return detect_trap_opportunity(user_argument, session.get("user_arguments", []), session)
+
+
 def detect_trap_opportunity(
     user_argument: str,
     previous_arguments: List[Dict],
     session: Dict,
 ) -> Optional[Tuple[str, str]]:
     """
-    Analyse user's argument to detect trap opportunities.
+    Analyse user's argument to detect trap opportunities (keyword-based fallback).
     
     Returns (trap_type, description) or None.
-    
-    This runs before the LLM call so we can include
-    trap instruction in the prompt when relevant.
     """
     arg_lower = user_argument.lower()
     

src/court/orchestrator.py

@@ -64,6 +64,86 @@ def _call_llm(messages: List[Dict]) -> str:
     return call_llm_raw(messages)
 
 
+def _build_full_context(session: Dict) -> str:
+    """
+    Builds complete context string fed to every LLM call.
+    Ensures nothing goes wasted — case brief, all arguments, all documents,
+    all concessions, all traps are included in every agent's view.
+    
+    This is the single source of context enrichment for the moot court.
+    """
+    parts = []
+    
+    # ── Case foundation ────────────────────────────────────────
+    parts.append("=== CASE FOUNDATION ===")
+    parts.append(f"Case: {session.get('case_title', '')}")
+    parts.append(f"Your side: {session.get('user_side', '').upper()}")
+    parts.append(f"Legal issues: {', '.join(session.get('legal_issues', []))}")
+    if session.get('brief_facts'):
+        parts.append(f"Facts: {session.get('brief_facts', '')[:400]}")
+    parts.append("")
+    
+    # ── Case brief (preserved throughout) ───────────────────────
+    case_brief = session.get("case_brief", "")
+    if case_brief:
+        parts.append("=== CASE BRIEF & RESEARCH ===")
+        parts.append(case_brief[:1200])
+        parts.append("")
+    
+    # ── Documents produced (CRITICAL — opposing counsel sees these) ──
+    docs = session.get("documents_produced", [])
+    if docs:
+        parts.append("=== DOCUMENTS ON RECORD ===")
+        for doc in docs:
+            parts.append(f"[{doc.get('type', 'DOCUMENT')} — filed by {doc.get('for_side', 'COUNSEL')}]")
+            parts.append(doc.get("content", "")[:500])
+            parts.append("")
+    
+    # ── All concessions made (CRITICAL — opposing counsel exploits these) ──
+    concessions = session.get("concessions", [])
+    if concessions:
+        parts.append("=== CONCESSIONS ON RECORD (EXPLOIT THESE) ===")
+        for c in concessions:
+            parts.append(
+                f"Round {c.get('round_number', '?')}: \"{c.get('exact_quote', '')[:120]}\" "
+                f"({c.get('legal_significance', 'Concession')[:80]})"
+            )
+        parts.append("")
+    
+    # ── Trap history (opposing counsel knows what worked before) ──
+    traps = session.get("trap_events", [])
+    if traps:
+        parts.append("=== TRAP HISTORY ===")
+        for t in traps:
+            fell = "USER FELL IN" if t.get("user_fell_in") else "user avoided"
+            parts.append(
+                f"Round {t.get('round_number', '?')} [{t.get('trap_type', 'trap').upper()}] {fell}: "
+                f"{t.get('trap_text', '')[:120]}"
+            )
+        parts.append("")
+    
+    # ── Full user argument history (consistency checking) ────────
+    user_args = session.get("user_arguments", [])
+    if user_args:
+        parts.append("=== USER'S ARGUMENT HISTORY ===")
+        for arg in user_args:
+            parts.append(f"Round {arg.get('round', '?')}: {arg.get('text', '')[:250]}")
+        parts.append("")
+    
+    # ── Recent transcript (verbatim, untruncated where possible) ─
+    transcript = session.get("transcript", [])
+    if transcript:
+        recent = transcript[-10:]  # More entries than before (was 4, now 10)
+        parts.append("=== RECENT PROCEEDINGS ===")
+        for entry in recent:
+            role = entry.get('role_label', entry.get('speaker', 'SPEAKER')).upper()
+            content = entry.get('content', '')[:300]
+            parts.append(f"{role}: {content}")
+            parts.append("")
+    
+    return "\n".join(parts)
+
+
 def _retrieve_for_court(query: str, session: Dict) -> str:
     """
     Retrieve relevant precedents for court use.
@@ -156,9 +236,13 @@ def _handle_briefing(session_id: str, user_argument: str, session: Dict) -> Dict
     
     add_user_argument(session_id, user_argument, [])
     
-    # Retrieve context for both agents
+    # Build full context with all case info
+    full_context = _build_full_context(session)
+    
+    # Retrieve additional precedents
     query = f"{session.get('case_title', '')} {' '.join(session.get('legal_issues', []))}"
-    retrieved_context = _retrieve_for_court(query, session)
+    retrieved_precedents = _retrieve_for_court(query, session)
+    combined_context = full_context + "\n\n=== RETRIEVED PRECEDENTS ===\n" + retrieved_precedents if retrieved_precedents else full_context
     
     # Check for trap opportunity
     trap_info = detect_trap_opportunity(user_argument, [], session)
@@ -167,7 +251,7 @@ def _handle_briefing(session_id: str, user_argument: str, session: Dict) -> Dict
     opposing_messages = build_opposing_prompt(
         session=session,
         user_argument=user_argument,
-        retrieved_context=retrieved_context,
+        retrieved_context=combined_context,
         trap_opportunity=trap_info[1] if trap_info else None,
     )
     
@@ -193,7 +277,7 @@ def _handle_briefing(session_id: str, user_argument: str, session: Dict) -> Dict
     judge_messages = build_judge_prompt(
         session=get_session(session_id),  # Fresh session after updates
         last_user_argument=user_argument,
-        retrieved_context=retrieved_context,
+        retrieved_context=combined_context,
     )
     
     try:
@@ -270,12 +354,17 @@ def _handle_round(session_id: str, user_argument: str, session: Dict) -> Dict:
         key_claims=_extract_key_claims(user_argument),
     )
     
-    # Retrieve context
+    # Build full context — EVERYTHING is preserved and fed to agents
+    fresh_session = get_session(session_id)
+    full_context = _build_full_context(fresh_session)
+    
+    # Retrieve additional precedents and combine
     legal_issues = " ".join(session.get("legal_issues", []))
     query = f"{user_argument[:200]} {legal_issues}"
-    retrieved_context = _retrieve_for_court(query, session)
+    retrieved_precedents = _retrieve_for_court(query, session)
+    combined_context = full_context + "\n\n=== RETRIEVED PRECEDENTS ===\n" + retrieved_precedents if retrieved_precedents else full_context
     
-    # Detect traps
+    # Detect traps (based on full history)
     trap_info = detect_trap_opportunity(
         user_argument,
         session.get("user_arguments", []),
@@ -283,11 +372,10 @@ def _handle_round(session_id: str, user_argument: str, session: Dict) -> Dict:
     )
     
     # ── LLM Call 1: Opposing counsel ──────────────────────────
-    fresh_session = get_session(session_id)
     opposing_messages = build_opposing_prompt(
         session=fresh_session,
         user_argument=user_argument,
-        retrieved_context=retrieved_context,
+        retrieved_context=combined_context,
         trap_opportunity=trap_info[1] if trap_info else None,
     )
     
@@ -323,7 +411,7 @@ def _handle_round(session_id: str, user_argument: str, session: Dict) -> Dict:
     judge_messages = build_judge_prompt(
         session=fresh_session,
         last_user_argument=user_argument,
-        retrieved_context=retrieved_context,
+        retrieved_context=combined_context,
     )
     
     try:
@@ -410,13 +498,16 @@ def _handle_cross_exam_answer(
     # If more questions remaining (max 3), get next question
     if question_number < 3:
         query = " ".join(session.get("legal_issues", []))
-        retrieved_context = _retrieve_for_court(query, session)
+        retrieved_precedents = _retrieve_for_court(query, session)
         
         fresh_session = get_session(session_id)
+        full_context = _build_full_context(fresh_session)
+        combined_context = full_context + "\n\n=== RETRIEVED PRECEDENTS ===\n" + retrieved_precedents if retrieved_precedents else full_context
+        
         cross_messages = build_cross_examination_prompt(
             session=fresh_session,
             question_number=question_number + 1,
-            retrieved_context=retrieved_context,
+            retrieved_context=combined_context,
         )
         
         try: