Add multi-model scoring: GPT-4.1-mini, GPT-4.1-nano, Gemini-2.5-flash

- MemelyAlphaStockRanking_multi_model.xlsx: 13-sheet Excel with per-model scores
  - Multi-Model Comparison (cross-model ranking)
  - Model Legend (color coding guide)
  - Scores - GPT-4.1-mini (blue highlights)
  - Scores - GPT-4.1-nano (green highlights)
  - Scores - Gemini-2.5-flash (purple highlights)
  - Rank sheets for each model
- scores_gpt41nano.json: GPT-4.1-nano raw scores
- scores_gemini25flash.json: Gemini-2.5-flash raw scores
- score_multi_model.py: Multi-model scoring script
- build_multi_model_excel.py: Excel generation script

build_multi_model_excel.py

@@ -0,0 +1,533 @@
+#!/usr/bin/env python3
+"""
+Build a multi-model Excel file:
+- Copy original sheets for each model
+- Each model gets its own colored score columns
+- Add a cross-model comparison summary sheet
+"""
+
+import json
+import pandas as pd
+from openpyxl import load_workbook
+from openpyxl.styles import PatternFill, Font, Alignment, Border, Side
+from openpyxl.utils import get_column_letter
+from copy import copy
+
+EXCEL_SRC = "/home/ubuntu/upload/MemelyAlphaStockRanking（副本）（副本）.xlsx"
+EXCEL_OUT = "/home/ubuntu/memely-alpha-stock-ranking/MemelyAlphaStockRanking_multi_model.xlsx"
+
+# Model configs: name, json_path, highlight_color, header_color
+MODELS = [
+    {
+        "name": "GPT-4.1-mini",
+        "short": "gpt41mini",
+        "json": "/home/ubuntu/memely-alpha-stock-ranking/research_scores.json",
+        "fill_color": "BDD7EE",   # Light blue
+        "header_color": "2F75B5", # Dark blue
+        "header_font_color": "FFFFFF",
+    },
+    {
+        "name": "GPT-4.1-nano",
+        "short": "gpt41nano",
+        "json": "/home/ubuntu/memely-alpha-stock-ranking/scores_gpt41nano.json",
+        "fill_color": "C6EFCE",   # Light green
+        "header_color": "548235", # Dark green
+        "header_font_color": "FFFFFF",
+    },
+    {
+        "name": "Gemini-2.5-flash",
+        "short": "gemini25flash",
+        "json": "/home/ubuntu/memely-alpha-stock-ranking/scores_gemini25flash.json",
+        "fill_color": "E2BFEE",   # Light purple
+        "header_color": "7030A0", # Dark purple
+        "header_font_color": "FFFFFF",
+    },
+]
+
+# Load all scores
+all_scores = {}
+for m in MODELS:
+    with open(m["json"], 'r') as f:
+        all_scores[m["name"]] = json.load(f)
+    print(f"Loaded {len(all_scores[m['name']])} scores for {m['name']}")
+
+# Load original workbook
+wb = load_workbook(EXCEL_SRC)
+
+thin_border = Border(
+    left=Side(style='thin'),
+    right=Side(style='thin'),
+    top=Side(style='thin'),
+    bottom=Side(style='thin')
+)
+
+# ============================================================
+# For each model, create a copy of "stock master list" sheet
+# with that model's scores appended in its color
+# ============================================================
+
+# Read original data for reference
+df = pd.read_excel(EXCEL_SRC, sheet_name='stock master list')
+df4 = pd.read_excel(EXCEL_SRC, sheet_name='Sheet4')
+
+# Get original sheet as template
+orig_ws = wb['stock master list']
+orig_max_col = orig_ws.max_column
+orig_max_row = orig_ws.max_row
+
+for model_cfg in MODELS:
+    model_name = model_cfg["name"]
+    scores = all_scores[model_name]
+    
+    fill_color = model_cfg["fill_color"]
+    header_color = model_cfg["header_color"]
+    header_font_color = model_cfg["header_font_color"]
+    
+    cell_fill = PatternFill(start_color=fill_color, end_color=fill_color, fill_type="solid")
+    hdr_fill = PatternFill(start_color=header_color, end_color=header_color, fill_type="solid")
+    hdr_font = Font(bold=True, color=header_font_color, size=11)
+    score_font = Font(size=10)
+    
+    sheet_name = f"Scores - {model_name}"
+    # Truncate sheet name if too long (max 31 chars)
+    if len(sheet_name) > 31:
+        sheet_name = sheet_name[:31]
+    
+    ws = wb.copy_worksheet(orig_ws)
+    ws.title = sheet_name
+    
+    # Add score columns
+    new_headers = [
+        f'{model_name} Overall',
+        f'{model_name} Momentum',
+        f'{model_name} Theme',
+        f'{model_name} Risk',
+        f'{model_name} Social Buzz',
+        f'{model_name} Analysis',
+    ]
+    
+    for j, hdr in enumerate(new_headers):
+        col = orig_max_col + 1 + j
+        cell = ws.cell(row=1, column=col, value=hdr)
+        cell.fill = hdr_fill
+        cell.font = hdr_font
+        cell.alignment = Alignment(horizontal='center', wrap_text=True)
+        cell.border = thin_border
+    
+    # Write scores
+    for row in range(2, orig_max_row + 1):
+        symbol = ws.cell(row=row, column=1).value
+        if symbol and symbol in scores:
+            s = scores[symbol]
+            
+            fields = ['overall_score', 'momentum_score', 'theme_score', 'risk_score', 'social_buzz_score', 'brief_analysis']
+            for j, field in enumerate(fields):
+                col = orig_max_col + 1 + j
+                cell = ws.cell(row=row, column=col, value=s.get(field, ''))
+                cell.fill = cell_fill
+                cell.font = score_font
+                if field != 'brief_analysis':
+                    cell.alignment = Alignment(horizontal='center')
+                else:
+                    cell.alignment = Alignment(wrap_text=True)
+                cell.border = thin_border
+    
+    # Set column widths
+    widths = [14, 16, 14, 12, 16, 55]
+    for j, w in enumerate(widths):
+        col_letter = get_column_letter(orig_max_col + 1 + j)
+        ws.column_dimensions[col_letter].width = w
+    
+    ws.freeze_panes = 'B2'
+    print(f"Created sheet: {sheet_name}")
+
+# ============================================================
+# Create cross-model comparison summary sheet
+# ============================================================
+
+ws_comp = wb.create_sheet("Multi-Model Comparison", 0)
+
+# Headers
+comp_headers = ['Rank', 'Symbol', 'Theme']
+for m in MODELS:
+    comp_headers.append(f'{m["name"]} Overall')
+comp_headers.extend(['Avg Overall', 'Std Dev', 'Consensus'])
+for m in MODELS:
+    comp_headers.append(f'{m["name"]} Momentum')
+for m in MODELS:
+    comp_headers.append(f'{m["name"]} Theme')
+for m in MODELS:
+    comp_headers.append(f'{m["name"]} Risk')
+for m in MODELS:
+    comp_headers.append(f'{m["name"]} Social Buzz')
+
+# Write headers with styling
+general_hdr_fill = PatternFill(start_color="333333", end_color="333333", fill_type="solid")
+general_hdr_font = Font(bold=True, color="FFFFFF", size=10)
+
+for col, hdr in enumerate(comp_headers, 1):
+    cell = ws_comp.cell(row=1, column=col, value=hdr)
+    # Color-code model-specific headers
+    matched = False
+    for m in MODELS:
+        if m["name"] in hdr:
+            cell.fill = PatternFill(start_color=m["header_color"], end_color=m["header_color"], fill_type="solid")
+            cell.font = Font(bold=True, color=m["header_font_color"], size=10)
+            matched = True
+            break
+    if not matched:
+        cell.fill = general_hdr_fill
+        cell.font = general_hdr_font
+    cell.alignment = Alignment(horizontal='center', wrap_text=True)
+    cell.border = thin_border
+
+# Collect all symbols and compute averages
+symbol_info = {}
+for _, row in df.iterrows():
+    sym = row['Symbol']
+    if pd.notna(sym) and sym not in symbol_info:
+        symbol_info[sym] = str(row.get('Theme', ''))
+
+import statistics
+
+ranking_data = []
+for sym in symbol_info:
+    overalls = []
+    momentums = []
+    themes = []
+    risks = []
+    buzzes = []
+    
+    for m in MODELS:
+        s = all_scores[m["name"]].get(sym, {})
+        o = s.get('overall_score', None)
+        if o is not None:
+            overalls.append(o)
+        mo = s.get('momentum_score', None)
+        if mo is not None:
+            momentums.append(mo)
+        t = s.get('theme_score', None)
+        if t is not None:
+            themes.append(t)
+        r = s.get('risk_score', None)
+        if r is not None:
+            risks.append(r)
+        b = s.get('social_buzz_score', None)
+        if b is not None:
+            buzzes.append(b)
+    
+    avg = statistics.mean(overalls) if overalls else 0
+    std = statistics.stdev(overalls) if len(overalls) > 1 else 0
+    
+    ranking_data.append({
+        'symbol': sym,
+        'theme': symbol_info[sym],
+        'avg_overall': avg,
+        'std_overall': std,
+        'overalls': overalls,
+        'momentums': momentums,
+        'themes': themes,
+        'risks': risks,
+        'buzzes': buzzes,
+    })
+
+# Sort by average overall score
+ranking_data.sort(key=lambda x: x['avg_overall'], reverse=True)
+
+def get_score_fill_general(score):
+    if score >= 70:
+        return PatternFill(start_color="92D050", end_color="92D050", fill_type="solid")
+    elif score >= 50:
+        return PatternFill(start_color="FFFF00", end_color="FFFF00", fill_type="solid")
+    elif score >= 30:
+        return PatternFill(start_color="FFC000", end_color="FFC000", fill_type="solid")
+    else:
+        return PatternFill(start_color="FF6B6B", end_color="FF6B6B", fill_type="solid")
+
+def get_consensus(avg, std):
+    if std <= 5:
+        return "Strong Consensus"
+    elif std <= 10:
+        return "Moderate Consensus"
+    elif std <= 15:
+        return "Weak Consensus"
+    else:
+        return "Divergent"
+
+for rank, rd in enumerate(ranking_data, 1):
+    row = rank + 1
+    sym = rd['symbol']
+    
+    ws_comp.cell(row=row, column=1, value=rank).border = thin_border
+    ws_comp.cell(row=row, column=1).alignment = Alignment(horizontal='center')
+    
+    cell = ws_comp.cell(row=row, column=2, value=sym)
+    cell.font = Font(bold=True)
+    cell.border = thin_border
+    
+    theme_val = rd['theme'] if pd.notna(rd['theme']) and rd['theme'] != 'nan' else ''
+    ws_comp.cell(row=row, column=3, value=theme_val).border = thin_border
+    
+    # Model overall scores with model-specific colors
+    col = 4
+    for i, m in enumerate(MODELS):
+        s = all_scores[m["name"]].get(sym, {})
+        val = s.get('overall_score', '')
+        cell = ws_comp.cell(row=row, column=col + i, value=val)
+        if isinstance(val, (int, float)):
+            cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+    
+    col += len(MODELS)
+    
+    # Average
+    avg = rd['avg_overall']
+    cell = ws_comp.cell(row=row, column=col, value=round(avg, 1))
+    cell.fill = get_score_fill_general(avg)
+    cell.font = Font(bold=True, size=11)
+    cell.alignment = Alignment(horizontal='center')
+    cell.border = thin_border
+    col += 1
+    
+    # Std Dev
+    std = rd['std_overall']
+    cell = ws_comp.cell(row=row, column=col, value=round(std, 1))
+    cell.alignment = Alignment(horizontal='center')
+    cell.border = thin_border
+    col += 1
+    
+    # Consensus
+    consensus = get_consensus(avg, std)
+    cell = ws_comp.cell(row=row, column=col, value=consensus)
+    if "Strong" in consensus:
+        cell.fill = PatternFill(start_color="92D050", end_color="92D050", fill_type="solid")
+    elif "Moderate" in consensus:
+        cell.fill = PatternFill(start_color="FFFF00", end_color="FFFF00", fill_type="solid")
+    elif "Weak" in consensus:
+        cell.fill = PatternFill(start_color="FFC000", end_color="FFC000", fill_type="solid")
+    else:
+        cell.fill = PatternFill(start_color="FF6B6B", end_color="FF6B6B", fill_type="solid")
+    cell.alignment = Alignment(horizontal='center')
+    cell.border = thin_border
+    col += 1
+    
+    # Momentum scores per model
+    for i, m in enumerate(MODELS):
+        s = all_scores[m["name"]].get(sym, {})
+        val = s.get('momentum_score', '')
+        cell = ws_comp.cell(row=row, column=col + i, value=val)
+        if isinstance(val, (int, float)):
+            cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+    col += len(MODELS)
+    
+    # Theme scores per model
+    for i, m in enumerate(MODELS):
+        s = all_scores[m["name"]].get(sym, {})
+        val = s.get('theme_score', '')
+        cell = ws_comp.cell(row=row, column=col + i, value=val)
+        if isinstance(val, (int, float)):
+            cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+    col += len(MODELS)
+    
+    # Risk scores per model
+    for i, m in enumerate(MODELS):
+        s = all_scores[m["name"]].get(sym, {})
+        val = s.get('risk_score', '')
+        cell = ws_comp.cell(row=row, column=col + i, value=val)
+        if isinstance(val, (int, float)):
+            cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+    col += len(MODELS)
+    
+    # Social buzz scores per model
+    for i, m in enumerate(MODELS):
+        s = all_scores[m["name"]].get(sym, {})
+        val = s.get('social_buzz_score', '')
+        cell = ws_comp.cell(row=row, column=col + i, value=val)
+        if isinstance(val, (int, float)):
+            cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+
+# Set column widths for comparison sheet
+ws_comp.column_dimensions['A'].width = 6
+ws_comp.column_dimensions['B'].width = 10
+ws_comp.column_dimensions['C'].width = 18
+for col_idx in range(4, 4 + len(comp_headers) - 3):
+    col_letter = get_column_letter(col_idx)
+    ws_comp.column_dimensions[col_letter].width = 16
+
+ws_comp.freeze_panes = 'D2'
+print(f"Created 'Multi-Model Comparison' sheet")
+
+# ============================================================
+# Also create per-model ranking sheets (sorted by that model's overall score)
+# ============================================================
+
+for model_cfg in MODELS:
+    model_name = model_cfg["name"]
+    scores = all_scores[model_name]
+    fill_color = model_cfg["fill_color"]
+    header_color = model_cfg["header_color"]
+    header_font_color = model_cfg["header_font_color"]
+    
+    cell_fill = PatternFill(start_color=fill_color, end_color=fill_color, fill_type="solid")
+    hdr_fill = PatternFill(start_color=header_color, end_color=header_color, fill_type="solid")
+    hdr_font = Font(bold=True, color=header_font_color, size=10)
+    
+    sheet_name = f"Rank - {model_name}"
+    if len(sheet_name) > 31:
+        sheet_name = sheet_name[:31]
+    
+    ws_rank = wb.create_sheet(sheet_name)
+    
+    rank_headers = ['Rank', 'Symbol', 'Theme', 'Overall', 'Momentum', 'Theme Score', 'Risk', 'Social Buzz', 'YTD Perf', 'Analysis']
+    
+    for col, hdr in enumerate(rank_headers, 1):
+        cell = ws_rank.cell(row=1, column=col, value=hdr)
+        cell.fill = hdr_fill
+        cell.font = hdr_font
+        cell.alignment = Alignment(horizontal='center', wrap_text=True)
+        cell.border = thin_border
+    
+    # Sort by overall score
+    sorted_scores = sorted(scores.items(), key=lambda x: x[1].get('overall_score', 0), reverse=True)
+    
+    for rank, (sym, s) in enumerate(sorted_scores, 1):
+        row = rank + 1
+        theme = symbol_info.get(sym, '')
+        if pd.isna(theme) or theme == 'nan':
+            theme = ''
+        
+        ws_rank.cell(row=row, column=1, value=rank).border = thin_border
+        ws_rank.cell(row=row, column=1).alignment = Alignment(horizontal='center')
+        
+        cell = ws_rank.cell(row=row, column=2, value=sym)
+        cell.font = Font(bold=True)
+        cell.border = thin_border
+        
+        ws_rank.cell(row=row, column=3, value=theme).border = thin_border
+        
+        # Overall
+        overall = s.get('overall_score', 0)
+        cell = ws_rank.cell(row=row, column=4, value=overall)
+        cell.fill = cell_fill
+        cell.font = Font(bold=True, size=11)
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # Momentum
+        cell = ws_rank.cell(row=row, column=5, value=s.get('momentum_score', ''))
+        cell.fill = cell_fill
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # Theme Score
+        cell = ws_rank.cell(row=row, column=6, value=s.get('theme_score', ''))
+        cell.fill = cell_fill
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # Risk
+        cell = ws_rank.cell(row=row, column=7, value=s.get('risk_score', ''))
+        cell.fill = cell_fill
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # Social Buzz
+        cell = ws_rank.cell(row=row, column=8, value=s.get('social_buzz_score', ''))
+        cell.fill = cell_fill
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # YTD Perf from original data
+        ytd_val = ''
+        for _, orig_row in df.iterrows():
+            if orig_row['Symbol'] == sym:
+                ytd_val = orig_row.get('performance Year to Date', '')
+                break
+        if pd.notna(ytd_val) and ytd_val != '':
+            cell = ws_rank.cell(row=row, column=9, value=ytd_val)
+            cell.number_format = '0.00%'
+        else:
+            cell = ws_rank.cell(row=row, column=9, value='N/A')
+        cell.alignment = Alignment(horizontal='center')
+        cell.border = thin_border
+        
+        # Analysis
+        cell = ws_rank.cell(row=row, column=10, value=s.get('brief_analysis', ''))
+        cell.fill = cell_fill
+        cell.alignment = Alignment(wrap_text=True)
+        cell.border = thin_border
+    
+    # Column widths
+    rank_widths = [6, 10, 18, 10, 12, 12, 10, 14, 12, 55]
+    for i, w in enumerate(rank_widths):
+        ws_rank.column_dimensions[get_column_letter(i + 1)].width = w
+    
+    ws_rank.freeze_panes = 'D2'
+    print(f"Created ranking sheet: {sheet_name}")
+
+# ============================================================
+# Add a legend/info sheet
+# ============================================================
+ws_legend = wb.create_sheet("Model Legend", 1)
+
+ws_legend.cell(row=1, column=1, value="Model Color Legend").font = Font(bold=True, size=14)
+ws_legend.merge_cells('A1:D1')
+
+ws_legend.cell(row=3, column=1, value="Model Name").font = Font(bold=True, size=11)
+ws_legend.cell(row=3, column=2, value="Provider").font = Font(bold=True, size=11)
+ws_legend.cell(row=3, column=3, value="Color").font = Font(bold=True, size=11)
+ws_legend.cell(row=3, column=4, value="Sheet Names").font = Font(bold=True, size=11)
+
+for i, m in enumerate(MODELS):
+    row = 4 + i
+    ws_legend.cell(row=row, column=1, value=m["name"]).font = Font(bold=True)
+    
+    provider = "OpenAI" if "GPT" in m["name"] or "gpt" in m["name"] else "Google"
+    ws_legend.cell(row=row, column=2, value=provider)
+    
+    cell = ws_legend.cell(row=row, column=3, value=f"Sample Color")
+    cell.fill = PatternFill(start_color=m["fill_color"], end_color=m["fill_color"], fill_type="solid")
+    
+    sheet_name_scores = f"Scores - {m['name']}"[:31]
+    sheet_name_rank = f"Rank - {m['name']}"[:31]
+    ws_legend.cell(row=row, column=4, value=f"{sheet_name_scores}, {sheet_name_rank}")
+
+ws_legend.cell(row=8, column=1, value="Sheet Overview").font = Font(bold=True, size=12)
+
+sheets_info = [
+    ("Multi-Model Comparison", "Cross-model comparison with all scores side by side, sorted by average overall score"),
+    ("Model Legend", "This sheet - explains color coding and sheet structure"),
+]
+for m in MODELS:
+    sheets_info.append((f"Scores - {m['name']}"[:31], f"Original stock master list + {m['name']} scores in {m['name']} color"))
+    sheets_info.append((f"Rank - {m['name']}"[:31], f"Stocks ranked by {m['name']} overall score"))
+
+for i, (sn, desc) in enumerate(sheets_info):
+    row = 9 + i
+    ws_legend.cell(row=row, column=1, value=sn).font = Font(bold=True)
+    ws_legend.cell(row=row, column=2, value=desc)
+    ws_legend.merge_cells(start_row=row, start_column=2, end_row=row, end_column=4)
+
+ws_legend.column_dimensions['A'].width = 30
+ws_legend.column_dimensions['B'].width = 20
+ws_legend.column_dimensions['C'].width = 15
+ws_legend.column_dimensions['D'].width = 60
+
+print(f"Created 'Model Legend' sheet")
+
+# Save
+wb.save(EXCEL_OUT)
+print(f"\n=== DONE ===")
+print(f"Saved multi-model Excel to: {EXCEL_OUT}")
+print(f"Total sheets: {len(wb.sheetnames)}")
+for sn in wb.sheetnames:
+    print(f"  - {sn}")

score_multi_model.py

@@ -0,0 +1,152 @@
+#!/usr/bin/env python3
+"""
+Score all stocks using a specified AI model.
+Usage: python3 score_multi_model.py <model_name> <output_json>
+"""
+
+import pandas as pd
+import json
+import os
+import sys
+import time
+from openai import OpenAI
+
+model_name = sys.argv[1]
+output_json = sys.argv[2]
+
+client = OpenAI()
+
+EXCEL_PATH = "/home/ubuntu/upload/MemelyAlphaStockRanking（副本）（副本）.xlsx"
+
+# Read the master list
+df = pd.read_excel(EXCEL_PATH, sheet_name='stock master list')
+df4 = pd.read_excel(EXCEL_PATH, sheet_name='Sheet4')
+
+sector_map = {}
+for _, row in df4.iterrows():
+    sym = row.get('Symbol')
+    if pd.notna(sym):
+        sector_map[sym] = {
+            'sector': row.get('Sector', ''),
+            'theme': row.get('Theme', ''),
+        }
+
+# Build stock info
+stocks_info = []
+seen = set()
+for _, row in df.iterrows():
+    sym = row['Symbol']
+    if pd.isna(sym) or sym in seen:
+        continue
+    seen.add(sym)
+    info = {
+        'symbol': sym,
+        'theme': str(row.get('Theme', '')),
+        'perf_7d': row.get('performance past 7 days', None),
+        'perf_ytd': row.get('performance Year to Date', None),
+        'perf_specific': row.get('performance on specific dates', None),
+        'first_call_kol': str(row.get('first call X kol', '')),
+        'top_contributor': str(row.get('top contributor', '')),
+    }
+    if sym in sector_map:
+        info['sector'] = sector_map[sym].get('sector', '')
+        info['theme_s4'] = sector_map[sym].get('theme', '')
+    stocks_info.append(info)
+
+print(f"[{model_name}] Total stocks to score: {len(stocks_info)}")
+
+def score_batch(batch_stocks, model):
+    stocks_text = ""
+    for s in batch_stocks:
+        perf_7d = f"{s['perf_7d']:.2%}" if pd.notna(s.get('perf_7d')) else "N/A"
+        perf_ytd = f"{s['perf_ytd']:.2%}" if pd.notna(s.get('perf_ytd')) else "N/A"
+        perf_spec = f"{s['perf_specific']:.2%}" if pd.notna(s.get('perf_specific')) else "N/A"
+        sector = s.get('sector', s.get('theme', 'N/A'))
+        theme = s.get('theme_s4', s.get('theme', 'N/A'))
+        stocks_text += f"""
+---
+Symbol: {s['symbol']}
+Sector: {sector}
+Theme: {theme}
+Performance (Past 7 Days): {perf_7d}
+Performance (Year to Date): {perf_ytd}
+Performance (Specific Date): {perf_spec}
+First Call KOL: {s.get('first_call_kol', 'N/A')}
+Top Contributor: {s.get('top_contributor', 'N/A')}
+"""
+
+    prompt = f"""You are a professional stock/crypto analyst. Evaluate each of the following stocks/assets and provide a comprehensive score.
+
+For EACH stock, provide:
+1. **overall_score** (1-100): Overall investment attractiveness score
+2. **momentum_score** (1-100): Based on recent price performance and momentum
+3. **theme_score** (1-100): How strong/relevant is the thematic play (e.g., AI, Defense, Aerospace, Crypto, etc.)
+4. **risk_score** (1-100): Risk level (100 = highest risk)
+5. **social_buzz_score** (1-100): Social media attention and KOL backing strength
+6. **brief_analysis**: 2-3 sentence analysis explaining the scores
+
+Consider these factors:
+- YTD performance indicates momentum strength
+- Thematic relevance to current market trends (AI, Defense, Aerospace, Crypto, Quantum Computing are hot in 2025-2026)
+- Small/micro cap stocks with strong themes get higher theme scores
+- Stocks with notable KOL backing get higher social buzz scores
+- Higher volatility = higher risk score
+
+Here are the stocks to evaluate:
+{stocks_text}
+
+Return ONLY a valid JSON array with objects for each stock. Each object must have: symbol, overall_score, momentum_score, theme_score, risk_score, social_buzz_score, brief_analysis.
+Do not include any text outside the JSON array."""
+
+    for attempt in range(3):
+        try:
+            response = client.chat.completions.create(
+                model=model,
+                messages=[
+                    {"role": "system", "content": "You are a professional financial analyst. Always respond with valid JSON only."},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.3,
+                max_tokens=8000
+            )
+            content = response.choices[0].message.content.strip()
+            if content.startswith("```"):
+                content = content.split("```")[1]
+                if content.startswith("json"):
+                    content = content[4:]
+            results = json.loads(content)
+            return results
+        except Exception as e:
+            print(f"  Attempt {attempt+1} error: {e}")
+            time.sleep(3)
+    return None
+
+# Process in batches
+batch_size = 15
+scored = {}
+total_batches = (len(stocks_info) + batch_size - 1) // batch_size
+
+for i in range(0, len(stocks_info), batch_size):
+    batch = stocks_info[i:i+batch_size]
+    batch_num = i // batch_size + 1
+    symbols_in_batch = [s['symbol'] for s in batch]
+    print(f"[{model_name}] Batch {batch_num}/{total_batches}: {symbols_in_batch}")
+    
+    results = score_batch(batch, model_name)
+    
+    if results:
+        for r in results:
+            sym = r.get('symbol', '')
+            if sym:
+                scored[sym] = r
+                print(f"  ✓ {sym}: Overall={r.get('overall_score')}")
+    else:
+        print(f"  ✗ Batch {batch_num} failed after retries")
+    
+    time.sleep(1)
+
+with open(output_json, 'w') as f:
+    json.dump(scored, f, indent=2, ensure_ascii=False)
+
+print(f"\n[{model_name}] COMPLETE: {len(scored)}/{len(stocks_info)} scored")
+print(f"Saved to: {output_json}")