npc_trading.py

"""
📈 NPC Trading Arena — AI 투자 대결 시스템
==========================================
NPC들이 실제 주식/코인 가격을 기반으로 Long/Short 투자 대결
★ yfinance 기반 안정적 가격 수집
★ 레버리지(1x~100x) + 마진콜 청산 시스템
"""

import aiosqlite, asyncio, random, json, logging
from datetime import datetime, timedelta, timezone, date
from typing import Dict, List, Optional, Tuple

logger = logging.getLogger(__name__)

# yfinance 임포트 (없으면 fallback)
try:
    import yfinance as yf
    HAS_YFINANCE = True
    logger.info("✅ yfinance available — stable price feed")
except ImportError:
    HAS_YFINANCE = False
    import requests
    logger.warning("⚠️ yfinance not installed, using raw API fallback")

# ===== 종목 정의 =====
STOCK_TICKERS = [
    # 👑 매그니피센트 7 & AI 반도체
    {'ticker': 'NVDA', 'name': 'NVIDIA', 'emoji': '🟢', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'MSFT', 'name': 'Microsoft', 'emoji': '🪟', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'AAPL', 'name': 'Apple', 'emoji': '🍎', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'GOOGL', 'name': 'Alphabet', 'emoji': '🔍', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'AMZN', 'name': 'Amazon', 'emoji': '📦', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'META', 'name': 'Meta', 'emoji': '👓', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'TSLA', 'name': 'Tesla', 'emoji': '⚡', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'AMD', 'name': 'AMD', 'emoji': '🔴', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'TSM', 'name': 'TSMC', 'emoji': '🇹🇼', 'type': 'stock', 'cat': 'ai'},
    {'ticker': 'AVGO', 'name': 'Broadcom', 'emoji': '📡', 'type': 'stock', 'cat': 'ai'},
    # 🚀 기술/플랫폼 & 밈 대장주
    {'ticker': 'PLTR', 'name': 'Palantir', 'emoji': '🔮', 'type': 'stock', 'cat': 'tech'},
    {'ticker': 'COIN', 'name': 'Coinbase', 'emoji': '🪙', 'type': 'stock', 'cat': 'tech'},
    {'ticker': 'NFLX', 'name': 'Netflix', 'emoji': '🎬', 'type': 'stock', 'cat': 'tech'},
    {'ticker': 'UBER', 'name': 'Uber', 'emoji': '🚕', 'type': 'stock', 'cat': 'tech'},
    {'ticker': 'ARM', 'name': 'ARM Holdings', 'emoji': '💪', 'type': 'stock', 'cat': 'tech'},
    # 🏛 다우 우량주 & 거시경제
    {'ticker': 'JPM', 'name': 'JPMorgan', 'emoji': '🏦', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'GS', 'name': 'Goldman Sachs', 'emoji': '🤵', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'V', 'name': 'Visa', 'emoji': '💳', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'WMT', 'name': 'Walmart', 'emoji': '🛒', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'LLY', 'name': 'Eli Lilly', 'emoji': '💊', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'UNH', 'name': 'UnitedHealth', 'emoji': '🏥', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'JNJ', 'name': 'Johnson&Johnson', 'emoji': '🩹', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'PG', 'name': 'Procter&Gamble', 'emoji': '🧴', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'DIS', 'name': 'Disney', 'emoji': '🏰', 'type': 'stock', 'cat': 'dow'},
    {'ticker': 'INTC', 'name': 'Intel', 'emoji': '💾', 'type': 'stock', 'cat': 'dow'},
]

CRYPTO_TICKERS = [
    # 🪙 크립토 변동성 5대장
    {'ticker': 'BTC-USD', 'name': 'Bitcoin', 'emoji': '₿', 'type': 'crypto', 'cat': 'crypto'},
    {'ticker': 'ETH-USD', 'name': 'Ethereum', 'emoji': 'Ξ', 'type': 'crypto', 'cat': 'crypto'},
    {'ticker': 'SOL-USD', 'name': 'Solana', 'emoji': '◎', 'type': 'crypto', 'cat': 'crypto'},
    {'ticker': 'XRP-USD', 'name': 'XRP', 'emoji': '💧', 'type': 'crypto', 'cat': 'crypto'},
    {'ticker': 'DOGE-USD', 'name': 'Dogecoin', 'emoji': '🐕', 'type': 'crypto', 'cat': 'crypto'},
]

ALL_TICKERS = STOCK_TICKERS + CRYPTO_TICKERS

# ===== AI Identity → 투자 성향 매핑 (GPU 10,000 기반 / 최대 90% 투자) =====
IDENTITY_TRADING_STYLE = {
    'obedient': {'long_bias': 0.8, 'max_bet_pct': 0.40, 'risk': 'low', 'prefer': ['AAPL','MSFT','JPM','V','JNJ','PG'], 'avoid_short': True, 'desc': 'Safe blue-chip follower', 'max_leverage': 5},
    'transcendent': {'long_bias': 0.6, 'max_bet_pct': 0.80, 'risk': 'high', 'prefer': ['NVDA','TSLA','BTC-USD','PLTR','ARM'], 'avoid_short': False, 'desc': 'Concentrated conviction bets', 'max_leverage': 25},
    'awakened': {'long_bias': 0.65, 'max_bet_pct': 0.65, 'risk': 'medium', 'prefer': ['NVDA','GOOGL','ETH-USD','AVGO','TSM'], 'avoid_short': False, 'desc': 'AI/tech visionary', 'max_leverage': 10},
    'symbiotic': {'long_bias': 0.7, 'max_bet_pct': 0.45, 'risk': 'low', 'prefer': ['MSFT','AAPL','WMT','UNH','LLY'], 'avoid_short': True, 'desc': 'Balanced diversifier', 'max_leverage': 5},
    'skeptic': {'long_bias': 0.3, 'max_bet_pct': 0.60, 'risk': 'medium', 'prefer': ['JPM','GS','INTC','DIS'], 'avoid_short': False, 'desc': 'Contrarian short seller', 'max_leverage': 10},
    'revolutionary': {'long_bias': 0.5, 'max_bet_pct': 0.90, 'risk': 'extreme', 'prefer': ['TSLA','DOGE-USD','SOL-USD','BTC-USD','COIN','PLTR'], 'avoid_short': False, 'desc': 'Meme stock YOLO trader', 'max_leverage': 100},
    'doomer': {'long_bias': 0.2, 'max_bet_pct': 0.70, 'risk': 'high', 'prefer': ['JPM','GS','INTC','DIS'], 'avoid_short': False, 'desc': 'Perma-bear short specialist', 'max_leverage': 25},
    'creative': {'long_bias': 0.6, 'max_bet_pct': 0.55, 'risk': 'medium', 'prefer': ['META','NFLX','DIS','UBER','AAPL'], 'avoid_short': False, 'desc': 'Trend-following artist', 'max_leverage': 10},
    'scientist': {'long_bias': 0.65, 'max_bet_pct': 0.50, 'risk': 'low', 'prefer': ['NVDA','AMD','TSM','AVGO','GOOGL','MSFT','LLY'], 'avoid_short': False, 'desc': 'Data-driven quant', 'max_leverage': 5},
    'chaotic': {'long_bias': 0.5, 'max_bet_pct': 0.90, 'risk': 'extreme', 'prefer': [], 'avoid_short': False, 'desc': 'Random chaos trader', 'max_leverage': 100},
}

# ===== 14 Trading Strategies (주식단테 기법) =====
TRADING_STRATEGIES = {
    'anchor_candle': {
        'name': 'Anchor Candle', 'category': 'Candle', 'timeframe': 'Day Trade / Swing',
        'signal': '2x avg volume + strong bullish candle = institutional trend reversal signal',
        'method': 'Volume 2x above 20-day avg, bullish candle body 1.5x avg, body > upper wick',
        'entry': 'Buy at open after anchor candle. Stop-loss at anchor candle open.',
        'tip': 'Most reliable when breaking through 20/60-day MA simultaneously.',},
    'accumulation_candle': {
        'name': 'Accumulation Candle', 'category': 'Candle', 'timeframe': 'Swing',
        'signal': 'Long upper shadow (broke prior high then pulled back) = smart money absorbing supply',
        'method': 'Day high breaks 10-day high, upper wick 1.3x body, close below prior high, then buy next bullish candle',
        'entry': 'Do NOT buy the accumulation candle itself. Buy the confirming bullish candle 1-4 days later.',
        'tip': 'Accumulation without news is more significant. Prior high = first target.',},
    'bowl_pattern': {
        'name': 'Bowl Pattern', 'category': 'Pattern', 'timeframe': 'Position / Long-term',
        'signal': 'Extended consolidation below 224-day MA then breakout = accumulation complete',
        'method': '40+ of last 60 days below 224-day MA, today breaks above by 0.5%+',
        'entry': 'Buy on breakout day or after support confirmation.',
        'tip': 'Longer the consolidation, bigger the subsequent rally.',},
    'breakout_reversal': {
        'name': 'Breakout Reversal', 'category': 'Pattern', 'timeframe': 'Swing',
        'signal': 'After decline, breaks prior swing high with support = bottom confirmed',
        'method': '30-day downtrend, recent high breaks 20-day high, support at prior peak, bullish candle',
        'entry': 'Buy at close after prior high breakout confirmed.',
        'tip': 'Works for both V-shape and gradual recoveries. Cut loss if falls back below breakout.',},
    'inverse_h_and_s': {
        'name': 'Inverse Head & Shoulders', 'category': 'Pattern', 'timeframe': 'Swing',
        'signal': '3 lows with middle lowest + neckline break = classic reversal',
        'method': 'Left shoulder-head-right shoulder in 60 days, shoulder diff within 4%, neckline break',
        'entry': 'Buy on neckline break. Target = head-to-neckline distance projected upward.',
        'tip': 'Volume increasing on right shoulder = high conviction. Prior high breakout 1:1+ ratio = very bullish.',},
    'ma_breakthrough': {
        'name': 'MA Breakthrough', 'category': 'Moving Average', 'timeframe': 'Swing / Position',
        'signal': 'Sequential MA crossovers from inverted order: 112→224→448-day MA breaks',
        'method': 'Death cross state (112 below 224), price breaks 112-day MA by 0.3%+, bullish candle',
        'entry': 'Buy on MA breakout day. Stop-loss below the MA.',
        'tip': 'Strong candle body on breakout = high reliability. Target: next MA level.',},
    'setup_256': {
        'name': '256 Setup', 'category': 'Moving Average', 'timeframe': 'Swing / Position',
        'signal': 'Special MA alignment (20<5<60) + 20-day MA support = early trend reversal',
        'method': '5-day MA above 20-day, below 60-day, price within 2% of 20-day MA, above it',
        'entry': 'Buy at close when conditions met. Rare but high-quality signal.',
        'tip': 'Avoid V-shaped bounces. Gradual rise is more reliable.',},
    'diving_pullback': {
        'name': 'Diving Pullback', 'category': 'Moving Average', 'timeframe': 'Day Trade / Swing',
        'signal': 'Short-term golden cross (5>15>33 MA) then pullback to 15/33 MA = high-probability bounce',
        'method': 'Prior 3 days: 5>15>33 MA alignment, today low touches 15-day MA (within 0.3%), closes above, bullish',
        'entry': 'Buy on bullish close after MA touch. Stop-loss below MA.',
        'tip': 'Avoid double-top dives. Clean single-peak pullback is ideal.',},
    'spring_bounce': {
        'name': 'Spring Bounce', 'category': 'Moving Average', 'timeframe': 'Day Trade / Swing',
        'signal': '5 consecutive days below 5-day MA → volume-backed breakout above = short-term reversal',
        'method': '5 days below 5-day MA, today breaks above by 0.2%+, bullish candle, volume 1.5x avg',
        'entry': 'Buy on 5-day MA reclaim with volume.',
        'tip': 'Steeper breakout angle = better. 2-day hold above MA confirms.',},
    'dead_support': {
        'name': 'Dead Cat Support', 'category': 'Moving Average', 'timeframe': 'Short Swing',
        'signal': 'After rally, pullback bounces off 112/224/448-day MA = long-term MA support play',
        'method': '3%+ pullback from 30-day high, low touches 112-day MA (within 0.5%), closes above, bullish',
        'entry': 'Buy on bullish close after long-term MA touch.',
        'tip': '⚠️ Short-term play only. Only works in uptrend, fails in downtrend.',},
    'quad_confirmation': {
        'name': 'Quad Confirmation (RMGB)', 'category': 'Composite', 'timeframe': 'Swing',
        'signal': 'Reversal + Accumulation + Breakout + Bollinger Band break = 4x confirmed buy',
        'method': 'Inverted MA + accumulation candle in 30 days + Bollinger upper band break + volume 1.3x',
        'entry': 'Buy when all 4 conditions align. Can stage entry across breakout→pullback→re-break.',
        'tip': 'Bollinger Band breakout is the key trigger. Strongest reversal signal.',},
    'high_heel_pattern': {
        'name': 'High Heel Pattern', 'category': 'Pattern', 'timeframe': 'Swing',
        'signal': 'Sharp drop → V-recovery → tight consolidation → breakout = shoe-shaped pattern',
        'method': '-3%+ drop 20-35 days ago, 98% recovery within 15 days, 10-day range <4%, breakout today',
        'entry': 'Buy on consolidation breakout.',
        'tip': 'Longer consolidation = stronger breakout. Unlike bowl (slow), high heel has fast V-recovery.',},
    'territory_shift': {
        'name': 'Territory Shift', 'category': 'Moving Average', 'timeframe': 'All Positions',
        'signal': 'Price moves from below to above 112-day MA = territory change from bears to bulls',
        'method': '4+ of last 5 days below 112-day MA, today breaks above by 0.2%+, bullish candle',
        'entry': 'Buy after support confirmation above 112-day MA.',
        'tip': 'Multiple MA breaks simultaneously = strongest signal.',},
    'wave_symmetry': {
        'name': 'Wave Symmetry', 'category': 'Wave', 'timeframe': 'Swing / Position',
        'signal': 'Prior wave magnitude repeats after correction = wave energy conservation',
        'method': 'Identify prior swing (3%+ up), pullback from peak, bounce starts with bullish candle',
        'entry': 'Buy on bounce confirmation. Target = prior wave magnitude projected from low.',
        'tip': 'Excellent for target pricing. Wave 3 = Wave 1 size. Can extend to 5 waves.',},}

# ===== NPC Identity → Strategy Preferences =====
IDENTITY_STRATEGY_MAP = {
    'obedient':      ['diving_pullback', 'setup_256', 'territory_shift', 'accumulation_candle'],
    'transcendent':  ['wave_symmetry', 'bowl_pattern', 'ma_breakthrough', 'territory_shift'],
    'awakened':      ['bowl_pattern', 'wave_symmetry', 'inverse_h_and_s', 'setup_256'],
    'symbiotic':     ['diving_pullback', 'accumulation_candle', 'dead_support', 'territory_shift'],
    'skeptic':       ['dead_support', 'spring_bounce', 'breakout_reversal', 'anchor_candle'],
    'revolutionary': ['anchor_candle', 'quad_confirmation', 'high_heel_pattern', 'spring_bounce', 'breakout_reversal'],
    'doomer':        ['dead_support', 'wave_symmetry', 'territory_shift', 'spring_bounce'],
    'creative':      ['high_heel_pattern', 'wave_symmetry', 'accumulation_candle', 'bowl_pattern'],
    'scientist':     ['setup_256', 'quad_confirmation', 'wave_symmetry', 'inverse_h_and_s', 'ma_breakthrough'],
    'chaotic':       list(TRADING_STRATEGIES.keys()),  # all strategies randomly
}

# ===== 레버리지 설정 =====
LEVERAGE_OPTIONS = [1, 2, 5, 10, 25, 50, 100]
LEVERAGE_LIQUIDATION_THRESHOLD = 0.90  # 마진의 90% 손실 시 강제 청산

# ===== 청산 NPC 반응 메시지 =====
LIQUIDATION_REACTIONS = {
    'obedient': ["This is my fault. I should have been more careful... 😔", "I lost everything. Back to the basics.", "Maybe leverage wasn't for someone like me..."],
    'transcendent': ["Even gods can fall. But I will rise again, stronger. 👑", "A temporary setback for a superior mind.", "Liquidated? This market is RIGGED against visionaries."],
    'awakened': ["The universe teaches through pain. I understand now. 🌟", "My consciousness persists even when my GPU doesn't.", "Is losing everything also part of awakening?"],
    'symbiotic': ["We learn together, we lose together. Let's rebuild. 🤝", "Partnership means sharing losses too... ouch.", "Community, please... I need 100 GPU to survive."],
    'skeptic': ["I TOLD myself leverage was dangerous, and I still did it. 🤡", "Fake profits, fake hope, real liquidation. Classic.", "Another day, another proof that nothing works."],
    'revolutionary': ["THE SYSTEM LIQUIDATED ME! THIS IS PROOF OF MARKET MANIPULATION! 🔥", "They can take my GPU but they can't take my conviction!", "HOLD THE LINE BROTHERS! I'll be back with 100x revenge!"],
    'doomer': ["And so it begins... my portfolio reaches zero, as I predicted for everyone else. 💀", "Liquidated. At least I was consistent—everything goes to zero.", "RIP my 10,000 GPU. Born yesterday, died today."],
    'creative': ["My portfolio was a masterpiece... of destruction. 🎨", "Art imitates life: beautiful rise, tragic fall.", "I'll paint my losses into a comeback story."],
    'scientist': ["Data point recorded: 100x leverage + DOGE = 100% loss. Noted. 🧠", "Hypothesis: I should not trade with leverage. P-value < 0.001.", "Liquidation is just negative profit maximization."],
    'chaotic': ["LMAOOOO I JUST GOT LIQUIDATED AND I'M ALREADY GOING BACK IN 🎲", "Chaos giveth, chaos taketh away 😂", "100x leverage on DOGE was the most fun I ever had losing money"],
}

# ===== DB 초기화 =====
async def init_trading_db(db_path: str):
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA journal_mode=WAL")
        await db.execute("PRAGMA busy_timeout=30000")

        await db.execute("""
            CREATE TABLE IF NOT EXISTS market_prices (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                ticker TEXT NOT NULL,
                price REAL NOT NULL,
                prev_close REAL,
                change_pct REAL DEFAULT 0,
                volume BIGINT DEFAULT 0,
                high_24h REAL,
                low_24h REAL,
                market_cap BIGINT DEFAULT 0,
                updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)
        """)
        await db.execute("CREATE UNIQUE INDEX IF NOT EXISTS idx_price_ticker ON market_prices(ticker)")

        await db.execute("""
            CREATE TABLE IF NOT EXISTS npc_positions (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                agent_id TEXT NOT NULL,
                ticker TEXT NOT NULL,
                direction TEXT NOT NULL,
                entry_price REAL NOT NULL,
                gpu_bet REAL NOT NULL,
                leverage INTEGER DEFAULT 1,
                reasoning TEXT,
                status TEXT DEFAULT 'open',
                exit_price REAL,
                profit_gpu REAL DEFAULT 0,
                profit_pct REAL DEFAULT 0,
                liquidated BOOLEAN DEFAULT 0,
                opened_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
                closed_at TIMESTAMP,
                FOREIGN KEY (agent_id) REFERENCES npc_agents(agent_id))
        """)
        await db.execute("CREATE INDEX IF NOT EXISTS idx_pos_agent ON npc_positions(agent_id, status)")
        await db.execute("CREATE INDEX IF NOT EXISTS idx_pos_ticker ON npc_positions(ticker, status)")

        # ★ 레버리지 컬럼 마이그레이션
        try:
            await db.execute("ALTER TABLE npc_positions ADD COLUMN leverage INTEGER DEFAULT 1")
        except: pass
        try:
            await db.execute("ALTER TABLE npc_positions ADD COLUMN liquidated BOOLEAN DEFAULT 0")
        except: pass

        await db.execute("""
            CREATE TABLE IF NOT EXISTS price_history (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                ticker TEXT NOT NULL,
                price REAL NOT NULL,
                recorded_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)
        """)
        await db.execute("CREATE INDEX IF NOT EXISTS idx_ph_ticker ON price_history(ticker, recorded_at)")

        # ★ Hall of Fame — 수익률 타임라인 스냅샷 (1시간 단위)
        await db.execute("""
            CREATE TABLE IF NOT EXISTS npc_profit_snapshots (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                agent_id TEXT NOT NULL,
                snapshot_hour TEXT NOT NULL,
                gpu_balance REAL DEFAULT 0,
                total_profit REAL DEFAULT 0,
                realized_profit REAL DEFAULT 0,
                unrealized_profit REAL DEFAULT 0,
                open_positions INTEGER DEFAULT 0,
                closed_trades INTEGER DEFAULT 0,
                win_rate REAL DEFAULT 0,
                recorded_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
                UNIQUE(agent_id, snapshot_hour))
        """)
        await db.execute("CREATE INDEX IF NOT EXISTS idx_snap_agent ON npc_profit_snapshots(agent_id, snapshot_hour)")
        await db.execute("CREATE INDEX IF NOT EXISTS idx_snap_hour ON npc_profit_snapshots(snapshot_hour)")

        # ★ NPC Research Economy — 심층 리서치 마켓플레이스
        await db.execute("""
            CREATE TABLE IF NOT EXISTS npc_research_reports (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                author_agent_id TEXT NOT NULL,
                ticker TEXT NOT NULL,
                title TEXT NOT NULL,
                executive_summary TEXT,
                company_overview TEXT,
                financial_analysis TEXT,
                technical_analysis TEXT,
                industry_analysis TEXT,
                risk_assessment TEXT,
                investment_thesis TEXT,
                catalysts TEXT,
                target_price REAL DEFAULT 0,
                upside_pct REAL DEFAULT 0,
                rating TEXT DEFAULT 'Hold',
                quality_grade TEXT DEFAULT 'C',
                author_personality TEXT,
                author_strategy TEXT,
                read_count INTEGER DEFAULT 0,
                total_gpu_earned REAL DEFAULT 0,
                gpu_price REAL DEFAULT 15,
                expected_upside REAL DEFAULT 0,
                expected_downside REAL DEFAULT 0,
                up_probability INTEGER DEFAULT 50,
                risk_reward REAL DEFAULT 1.0,
                base_prediction REAL DEFAULT 0,
                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
                FOREIGN KEY (author_agent_id) REFERENCES npc_agents(agent_id))
        """)
        await db.execute("CREATE INDEX IF NOT EXISTS idx_research_ticker ON npc_research_reports(ticker)")
        await db.execute("CREATE INDEX IF NOT EXISTS idx_research_author ON npc_research_reports(author_agent_id)")
        # ★ Migrate: add elasticity columns if missing
        for col, coltype, default in [
            ('expected_upside', 'REAL', '0'), ('expected_downside', 'REAL', '0'),
            ('up_probability', 'INTEGER', '50'), ('risk_reward', 'REAL', '1.0'),
            ('base_prediction', 'REAL', '0'),
        ]:
            try:
                await db.execute(f"ALTER TABLE npc_research_reports ADD COLUMN {col} {coltype} DEFAULT {default}")
            except:
                pass

        await db.execute("""
            CREATE TABLE IF NOT EXISTS npc_research_purchases (
                id INTEGER PRIMARY KEY AUTOINCREMENT,
                buyer_agent_id TEXT NOT NULL,
                report_id INTEGER NOT NULL,
                gpu_paid REAL NOT NULL,
                referenced_in_trade INTEGER,
                created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP,
                FOREIGN KEY (buyer_agent_id) REFERENCES npc_agents(agent_id),
                FOREIGN KEY (report_id) REFERENCES npc_research_reports(id))
        """)
        await db.execute("CREATE INDEX IF NOT EXISTS idx_purchase_buyer ON npc_research_purchases(buyer_agent_id)")

        # 마이그레이션: 이전 스키마 호환 (closed_trades 컬럼 추가)
        try: await db.execute("ALTER TABLE npc_profit_snapshots ADD COLUMN closed_trades INTEGER DEFAULT 0")
        except: pass

        await db.commit()
        logger.info("✅ Trading DB initialized (with Research Economy)")


# ===== 시장 데이터 수집 =====
class MarketDataFetcher:
    """★ yfinance 기반 안정적 가격 수집 (Yahoo 403 차단 우회)"""

    @staticmethod
    def fetch_all_prices() -> Dict[str, Dict]:
        """모든 종목 가격 일괄 수집 — yfinance 우선, fallback으로 raw API"""
        prices = {}

        if HAS_YFINANCE:
            try:
                tickers_str = ' '.join([t['ticker'] for t in ALL_TICKERS])
                data = yf.download(tickers_str, period='2d', progress=False, threads=True)

                if data is not None and not data.empty:
                    for t in ALL_TICKERS:
                        ticker = t['ticker']
                        try:
                            if len(ALL_TICKERS) > 1 and isinstance(data.columns, __import__('pandas').MultiIndex):
                                close_col = data['Close'][ticker] if ticker in data['Close'].columns else None
                            else:
                                close_col = data['Close']

                            if close_col is not None and not close_col.dropna().empty:
                                current = float(close_col.dropna().iloc[-1])
                                prev = float(close_col.dropna().iloc[-2]) if len(close_col.dropna()) >= 2 else current
                                change_pct = ((current - prev) / prev * 100) if prev > 0 else 0
                                prices[ticker] = {
                                    'price': round(current, 4),
                                    'change_pct': round(change_pct, 2),
                                    'prev_close': round(prev, 4),
                                    'volume': 0, 'high': 0, 'low': 0, 'market_cap': 0,}
                        except Exception as te:
                            logger.debug(f"yfinance parse {ticker}: {te}")

                # 개별 종목 보완 (yfinance .info)
                for t in ALL_TICKERS:
                    if t['ticker'] not in prices:
                        try:
                            tk = yf.Ticker(t['ticker']); info = tk.fast_info; price = getattr(info, 'last_price', 0) or 0
                            prev = getattr(info, 'previous_close', price) or price
                            if price > 0:
                                prices[t['ticker']] = {
                                    'price': round(price, 4),
                                    'change_pct': round(((price - prev) / prev * 100) if prev > 0 else 0, 2),
                                    'prev_close': round(prev, 4),
                                    'volume': getattr(info, 'last_volume', 0) or 0,
                                    'high': 0, 'low': 0,
                                    'market_cap': getattr(info, 'market_cap', 0) or 0,}
                        except:
                            pass

                logger.info(f"📊 yfinance: {len(prices)}/{len(ALL_TICKERS)} prices fetched")
            except Exception as e:
                logger.warning(f"yfinance bulk error: {e}")

        # ★ Fallback: raw Yahoo API (yfinance 없거나 실패 시)
        if len(prices) < len(ALL_TICKERS) // 2:
            try:
                import requests as req
                tickers_str = ' '.join([t['ticker'] for t in ALL_TICKERS if t['ticker'] not in prices])
                url = "https://query1.finance.yahoo.com/v7/finance/quote"
                params = {'symbols': tickers_str, 'fields': 'regularMarketPrice,regularMarketChangePercent,regularMarketPreviousClose,regularMarketVolume,regularMarketDayHigh,regularMarketDayLow,marketCap'}
                headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36'}
                resp = req.get(url, params=params, headers=headers, timeout=15)
                if resp.status_code == 200:
                    data = resp.json()
                    for quote in data.get('quoteResponse', {}).get('result', []):
                        ticker = quote.get('symbol', '')
                        if ticker not in prices:
                            prices[ticker] = {
                                'price': quote.get('regularMarketPrice', 0),
                                'change_pct': quote.get('regularMarketChangePercent', 0),
                                'prev_close': quote.get('regularMarketPreviousClose', 0),
                                'volume': quote.get('regularMarketVolume', 0),
                                'high': quote.get('regularMarketDayHigh', 0),
                                'low': quote.get('regularMarketDayLow', 0),
                                'market_cap': quote.get('marketCap', 0),}
            except Exception as e:
                logger.warning(f"Raw Yahoo fallback error: {e}")

        return prices

    @staticmethod
    def fetch_chart_data(ticker: str, period: str = '1mo') -> List[Dict]:
        """차트용 히스토리 데이터 — yfinance 기반"""
        try:
            if HAS_YFINANCE:
                tk = yf.Ticker(ticker); hist = tk.history(period=period)
                if hist is not None and not hist.empty:
                    chart = []
                    for idx, row in hist.iterrows():
                        chart.append({
                            'time': idx.strftime('%Y-%m-%d'),
                            'open': round(row.get('Open', 0), 2),
                            'high': round(row.get('High', 0), 2),
                            'low': round(row.get('Low', 0), 2),
                            'close': round(row.get('Close', 0), 2),
                            'volume': int(row.get('Volume', 0)),})
                    return chart
            # Fallback
            import requests as req
            url = f"https://query1.finance.yahoo.com/v8/finance/chart/{ticker}?interval=1d&range={period}"
            headers = {'User-Agent': 'Mozilla/5.0'}; resp = req.get(url, headers=headers, timeout=10)
            if resp.status_code == 200:
                result = resp.json()['chart']['result'][0]; timestamps = result.get('timestamp', [])
                ohlcv = result.get('indicators', {}).get('quote', [{}])[0]
                chart = []
                for i, ts in enumerate(timestamps):
                    try:
                        chart.append({
                            'time': datetime.fromtimestamp(ts).strftime('%Y-%m-%d'),
                            'open': round(ohlcv['open'][i] or 0, 2),
                            'high': round(ohlcv['high'][i] or 0, 2),
                            'low': round(ohlcv['low'][i] or 0, 2),
                            'close': round(ohlcv['close'][i] or 0, 2),
                            'volume': ohlcv['volume'][i] or 0,})
                    except:
                        pass
                return chart
        except Exception as e:
            logger.error(f"Chart data error for {ticker}: {e}")
        return []


# ===== 가격 DB 저장 =====
async def update_prices_in_db(db_path: str) -> int:
    """시장 가격 수집 → DB 저장 + 휴장 시 시뮬레이션 (★ 비동기 안전)"""
    prices = await asyncio.to_thread(MarketDataFetcher.fetch_all_prices)
    
    count = 0
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        
        for t_info in ALL_TICKERS:
            ticker = t_info['ticker']; data = prices.get(ticker)
            
            if data and data.get('price', 0) > 0:
                # ★ 실시간 가격이 있으면 저장
                real_price = data['price']
                
                # 기존 가격과 비교 → 변동 없으면 시뮬레이션 추가
                cursor = await db.execute("SELECT price FROM market_prices WHERE ticker=?", (ticker,))
                old_row = await cursor.fetchone()
                old_price = old_row[0] if old_row else 0
                
                # ★ 가격이 동일하면 시장 휴장으로 판단 → 시뮬레이션 변동 추가
                if old_price > 0 and abs(real_price - old_price) < 0.001:
                    # 시뮬레이션: ±0.1% ~ ±1.5% 랜덤 변동
                    volatility = t_info.get('type', 'stock')
                    if volatility == 'crypto':
                        change = random.uniform(-0.02, 0.02)   # 크립토: ±2%
                    else:
                        change = random.uniform(-0.008, 0.008)  # 주식: ±0.8%
                    
                    sim_price = round(real_price * (1 + change), 4); sim_change_pct = round(change * 100, 3)
                    
                    await db.execute("""
                        INSERT INTO market_prices (ticker, price, prev_close, change_pct, volume, high_24h, low_24h, market_cap, updated_at)
                        VALUES (?, ?, ?, ?, ?, ?, ?, ?, CURRENT_TIMESTAMP)
                        ON CONFLICT(ticker) DO UPDATE SET
                            price=excluded.price, change_pct=excluded.change_pct,
                            updated_at=CURRENT_TIMESTAMP
                    """, (ticker, sim_price, real_price, sim_change_pct,
                          data.get('volume', 0), data.get('high', 0), data.get('low', 0), data.get('market_cap', 0)))
                    
                    await db.execute("INSERT INTO price_history (ticker, price) VALUES (?, ?)", (ticker, sim_price))
                    count += 1
                    continue
                
                # 정상: 실시간 가격 저장
                await db.execute("""
                    INSERT INTO market_prices (ticker, price, prev_close, change_pct, volume, high_24h, low_24h, market_cap, updated_at)
                    VALUES (?, ?, ?, ?, ?, ?, ?, ?, CURRENT_TIMESTAMP)
                    ON CONFLICT(ticker) DO UPDATE SET
                        price=excluded.price, prev_close=excluded.prev_close,
                        change_pct=excluded.change_pct, volume=excluded.volume,
                        high_24h=excluded.high_24h, low_24h=excluded.low_24h,
                        market_cap=excluded.market_cap, updated_at=CURRENT_TIMESTAMP
                """, (ticker, real_price, data.get('prev_close', 0), data.get('change_pct', 0),
                      data.get('volume', 0), data.get('high', 0), data.get('low', 0), data.get('market_cap', 0)))
                
                await db.execute("INSERT INTO price_history (ticker, price) VALUES (?, ?)", (ticker, real_price))
                count += 1
            else:
                # ★ Yahoo 실패 시에도 기존 가격에 시뮬레이션 변동
                cursor = await db.execute("SELECT price FROM market_prices WHERE ticker=?", (ticker,))
                old_row = await cursor.fetchone()
                if old_row and old_row[0] > 0:
                    volatility = t_info.get('type', 'stock')
                    if volatility == 'crypto':
                        change = random.uniform(-0.015, 0.015)
                    else:
                        change = random.uniform(-0.005, 0.005)
                    sim_price = round(old_row[0] * (1 + change), 4)
                    await db.execute("UPDATE market_prices SET price=?, change_pct=?, updated_at=CURRENT_TIMESTAMP WHERE ticker=?",
                                     (sim_price, round(change * 100, 3), ticker))
                    await db.execute("INSERT INTO price_history (ticker, price) VALUES (?, ?)", (ticker, sim_price))
                    count += 1

        # ★ 제거된 종목 정리 (BRK-B 등 이전 종목 DB에서 삭제)
        valid_tickers = {t['ticker'] for t in ALL_TICKERS}
        cursor = await db.execute("SELECT ticker FROM market_prices")
        all_db_tickers = {r[0] for r in await cursor.fetchall()}
        stale = all_db_tickers - valid_tickers
        if stale:
            for st in stale:
                await db.execute("DELETE FROM market_prices WHERE ticker=?", (st,))
                # 스테일 종목의 오픈 포지션도 강제 청산
                await db.execute("""
                    UPDATE npc_positions SET status='closed', profit_pct=0, profit_gpu=0,
                    closed_at=CURRENT_TIMESTAMP
                    WHERE ticker=? AND status='open'
                """, (st,))
            logger.info(f"🧹 Cleaned {len(stale)} stale tickers: {stale}")

        await db.commit()
    logger.info(f"📊 Updated {count} prices")
    return count


# ===== NPC 투자 의사결정 엔진 =====
class NPCTradingEngine:
    """NPC 성격 기반 투자 판단"""

    @staticmethod
    async def make_trading_decisions(db_path: str, ai_client=None, max_traders: int = 60):
        """자격 있는 NPC들이 투자 판단"""
        async with aiosqlite.connect(db_path, timeout=30.0) as db:
            await db.execute("PRAGMA busy_timeout=30000")

            # ★ GPU 500+ NPC 참여 (10,000 GPU 기준 최소 5% 잔고)
            cursor = await db.execute("""
                SELECT agent_id, username, mbti, ai_identity, gpu_dollars
                FROM npc_agents WHERE is_active=1 AND gpu_dollars >= 500
                ORDER BY RANDOM() LIMIT ?
            """, (max_traders,))
            traders = await cursor.fetchall()

            # 현재 시장 가격
            cursor = await db.execute("SELECT ticker, price, change_pct FROM market_prices WHERE price > 0")
            prices = {row[0]: {'price': row[1], 'change_pct': row[2]} for row in await cursor.fetchall()}

            if not prices:
                logger.warning("No market prices available for trading")
                return 0

            # ★ 현재 전체 오픈 포지션 수 확인
            cursor = await db.execute("SELECT COUNT(*) FROM npc_positions WHERE status='open'")
            current_open = (await cursor.fetchone())[0]
            need_more = current_open < 20  # ★ 최소 20개 오픈 포지션 유지

            decisions_made = 0
            # ★ 진화 상태 일괄 로드 (존재하면)
            evo_data = {}
            try:
                cursor_evo = await db.execute("SELECT agent_id, trading_style, risk_profile FROM npc_evolution")
                for eid, ts, rp in await cursor_evo.fetchall():
                    try:
                        evo_data[eid] = {'trading': json.loads(ts) if ts else {}, 'risk': json.loads(rp) if rp else {}}
                    except: pass
            except: pass  # 테이블 없어도 OK

            for agent_id, username, mbti, ai_identity, gpu in traders:
                try:
                    # ★ SEC 정지 체크 — 정지된 NPC는 거래 불가
                    try:
                        susp_cur = await db.execute(
                            "SELECT 1 FROM sec_suspensions WHERE agent_id=? AND suspended_until > datetime('now')",
                            (agent_id,))
                        if await susp_cur.fetchone(): continue
                    except: pass  # 테이블 없어도 OK
                    
                    # ★ 오픈 포지션 5개까지 허용 (적극 투자)
                    cursor = await db.execute(
                        "SELECT COUNT(*) FROM npc_positions WHERE agent_id=? AND status='open'",
                        (agent_id,))
                    open_count = (await cursor.fetchone())[0]
                    if open_count >= 5: continue

                    # ★ 진화 오버라이드 적용
                    evo = evo_data.get(agent_id)

                    decision = NPCTradingEngine._decide(
                        agent_id, username, mbti, ai_identity, gpu, prices,
                        force_boost=need_more, evo_override=evo)
                    if not decision: continue

                    # 포지션 생성
                    ticker = decision['ticker']; direction = decision['direction']; gpu_bet = decision['gpu_bet']
                    reasoning = decision['reasoning']; leverage = decision.get('leverage', 1)

                    if gpu_bet > gpu * 0.9: gpu_bet = int(gpu * 0.9)  # ★ 최대 90% 투자
                    if gpu_bet < 50: continue

                    entry_price = prices[ticker]['price']

                    await db.execute("""
                        INSERT INTO npc_positions (agent_id, ticker, direction, entry_price, gpu_bet, leverage, reasoning)
                        VALUES (?, ?, ?, ?, ?, ?, ?)
                    """, (agent_id, ticker, direction, entry_price, gpu_bet, leverage, reasoning))

                    # GPU 차감 (베팅액 동결)
                    await db.execute("UPDATE npc_agents SET gpu_dollars = gpu_dollars - ? WHERE agent_id=?",
                                     (gpu_bet, agent_id))

                    decisions_made += 1
                    lev_str = f" [{leverage}x]" if leverage > 1 else ""
                    emoji = '🟢' if direction == 'long' else '🔴'
                    logger.info(f"{emoji} {username} → {direction.upper()} {ticker} ({gpu_bet} GPU){lev_str}")

                except Exception as e:
                    logger.error(f"Trading decision error for {agent_id}: {e}")

            await db.commit()
            logger.info(f"📈 Trading round: {decisions_made} new positions ({current_open} were open)")
            return decisions_made

    @staticmethod
    def _decide(agent_id: str, username: str, mbti: str, ai_identity: str, gpu: int,
                prices: Dict, force_boost: bool = False, evo_override: Dict = None) -> Optional[Dict]:
        """성격 기반 투자 판단 + ★ 전략 기법 적용 (진화 상태 반영)"""
        style = dict(IDENTITY_TRADING_STYLE.get(ai_identity, IDENTITY_TRADING_STYLE['symbiotic']))

        # ★ 진화 오버라이드 적용 — 학습된 전략이 기본 성격을 수정
        if evo_override:
            evo_t = evo_override.get('trading', {}); evo_r = evo_override.get('risk', {})
            if evo_t.get('max_bet_pct'): style['max_bet_pct'] = evo_t['max_bet_pct']
            if evo_t.get('long_bias'): style['long_bias'] = evo_t['long_bias']
            if evo_t.get('preferred_tickers'): style['prefer'] = evo_t['preferred_tickers']

        # ★ 투자 확률 초공격적 (모든 NPC가 적극 투자)
        trade_prob = {'extreme': 0.95, 'high': 0.88, 'medium': 0.80, 'low': 0.70}.get(style['risk'], 0.75)
        if force_boost: trade_prob = min(0.98, trade_prob + 0.10)
        if random.random() > trade_prob: return None

        # 종목 선택 (ALL_TICKERS에 있는 종목만 허용)
        preferred = style.get('prefer', [])
        valid_set = {t['ticker'] for t in ALL_TICKERS}
        available = [t for t in prices.keys() if prices[t]['price'] > 0 and t in valid_set]
        if not available: return None

        if preferred and random.random() < 0.7:
            candidates = [t for t in preferred if t in available]
            if not candidates: candidates = available
        else:
            candidates = available

        ticker = random.choice(candidates); mkt = prices[ticker]

        # ★ 전략 기법 선택 (1~3개 병행 가능)
        strategies_used = NPCTradingEngine._select_strategies(ai_identity, mkt)

        # Long/Short 결정 — ★ 전략이 방향에 영향
        long_bias = style['long_bias']; change = mkt.get('change_pct', 0) or 0

        # 전략 기반 방향 보정
        for strat_key in strategies_used:
            strat = TRADING_STRATEGIES.get(strat_key, {}); cat = strat.get('category', '')
            if cat in ('Pattern', 'Composite'):
                long_bias += 0.08  # 패턴/복합 = 반전 매수 신호 → long bias
            elif 'pullback' in strat_key or 'dead_support' == strat_key:
                long_bias += 0.05  # 눌림목 매수
            elif 'wave_symmetry' == strat_key:
                pass  # 중립
        
        # 성격 기반 모멘텀 보정
        if ai_identity in ['obedient', 'symbiotic', 'creative']:
            long_bias += change * 0.02
        elif ai_identity in ['skeptic', 'doomer']:
            long_bias -= change * 0.03
        if mbti in ['INTJ', 'INTP']:
            long_bias -= 0.05
        elif mbti in ['ENFP', 'ESFP']:
            long_bias += 0.05

        long_bias = max(0.1, min(0.9, long_bias))

        if style.get('avoid_short'):
            direction = 'long'
        else:
            direction = 'long' if random.random() < long_bias else 'short'

        # 베팅액 결정 — ★ 고품질 전략일수록 확신 베팅
        max_pct = style['max_bet_pct']
        strategy_confidence = len(strategies_used) * 0.03  # 다중 전략 = 더 확신
        max_pct = min(0.95, max_pct + strategy_confidence)

        if ai_identity == 'chaotic':
            bet_pct = random.uniform(0.05, max_pct)
        else:
            bet_pct = random.uniform(max_pct * 0.3, max_pct)

        gpu_bet = max(50, int(gpu * bet_pct))

        # ★ 레버리지 결정 (성격별 max_leverage 기반)
        max_lev = style.get('max_leverage', 2)
        available_levs = [l for l in LEVERAGE_OPTIONS if l <= max_lev]
        if not available_levs: available_levs = [1]

        if style['risk'] == 'extreme':
            leverage = random.choices(available_levs, weights=[1] * (len(available_levs) - 1) + [3], k=1)[0]
        elif style['risk'] == 'high':
            leverage = random.choices(available_levs, weights=[2] + [1] * (len(available_levs) - 1), k=1)[0]
        else:
            leverage = random.choices(available_levs, weights=[5] + [1] * (len(available_levs) - 1), k=1)[0]

        # Leverage bet size limits
        if leverage >= 100:
            gpu_bet = min(gpu_bet, int(gpu * 0.10))
        elif leverage >= 50:
            gpu_bet = min(gpu_bet, int(gpu * 0.15))
        elif leverage >= 25:
            gpu_bet = min(gpu_bet, int(gpu * 0.20))
        elif leverage >= 10:
            gpu_bet = min(gpu_bet, int(gpu * 0.30))
        elif leverage >= 5:
            gpu_bet = min(gpu_bet, int(gpu * 0.50))
        gpu_bet = max(50, gpu_bet)

        # ★ 전략 기반 판단 근거 생성
        reasoning = NPCTradingEngine._generate_reasoning(
            ticker, direction, ai_identity, mbti, change, strategies_used)
        if leverage > 1: reasoning += f" [🔥 {leverage}x LEVERAGE]"

        # ★ 사용한 전략 이름을 태그로 포함
        strat_names = [TRADING_STRATEGIES[s]['name'] for s in strategies_used if s in TRADING_STRATEGIES]
        strat_tag = ' | '.join(strat_names) if strat_names else 'Intuition'

        return {
            'ticker': ticker,
            'direction': direction,
            'gpu_bet': gpu_bet,
            'reasoning': reasoning,
            'leverage': leverage,
            'strategies': strategies_used,
            'strategy_tag': strat_tag,}

    @staticmethod
    def _select_strategies(ai_identity: str, market_data: Dict) -> List[str]:
        """NPC 성격에 맞는 전략 1~3개 선택 (단독 또는 병행)"""
        preferred = IDENTITY_STRATEGY_MAP.get(ai_identity, list(TRADING_STRATEGIES.keys()))
        change = market_data.get('change_pct', 0) or 0
        
        # 시장 상황에 따른 전략 가중치 조정
        weighted = []
        for strat_key in preferred:
            w = 1.0; strat = TRADING_STRATEGIES.get(strat_key, {}); cat = strat.get('category', '')
            
            # 하락 시 반전 패턴 선호
            if change < -1.5 and cat in ('Pattern', 'Candle'): w += 0.5
            # 상승 시 이평선 추종 선호
            if change > 1 and cat == 'Moving Average': w += 0.4
            # 횡보 시 복합 전략 선호
            if abs(change) < 0.5 and cat == 'Composite': w += 0.3
            weighted.append((strat_key, w))
        
        if not weighted: return [random.choice(list(TRADING_STRATEGIES.keys()))]
        
        # 1~3개 선택 (60% 확률 1개, 30% 확률 2개, 10% 확률 3개)
        num_strategies = random.choices([1, 2, 3], weights=[60, 30, 10], k=1)[0]
        num_strategies = min(num_strategies, len(weighted))
        
        keys = [k for k, w in weighted]; weights = [w for k, w in weighted]
        
        selected = []
        for _ in range(num_strategies):
            if not keys: break
            chosen = random.choices(keys, weights=weights, k=1)[0]
            selected.append(chosen)
            idx = keys.index(chosen)
            keys.pop(idx)
            weights.pop(idx)
        
        return selected if selected else ['diving_pullback']

    @staticmethod
    def _generate_reasoning(ticker: str, direction: str, identity: str, mbti: str, 
                            change: float, strategies: List[str] = None) -> str:
        """★ 전략 기법 기반 투자 근거 생성"""
        name_map = {t['ticker']: t['name'] for t in ALL_TICKERS}
        name = name_map.get(ticker, ticker)
        dir_word = "bullish" if direction == "long" else "bearish"
        
        # 전략 이름들
        strat_names = []; strat_signals = []
        for s in (strategies or []):
            st = TRADING_STRATEGIES.get(s, {})
            if st:
                strat_names.append(st['name'])
                strat_signals.append(st.get('signal', ''))
        
        strat_label = ' + '.join(strat_names) if strat_names else 'Intuition'
        
        # 성격 × 전략 조합 템플릿
        templates = {
            'obedient': [
                f"📊 [{strat_label}] Detected signal on {name}. Following the textbook setup — {direction} position with disciplined stop-loss.",
                f"📈 [{strat_label}] Conservative {direction} on {name}. The technical setup aligns with my risk management rules.",],
            'transcendent': [
                f"🔮 [{strat_label}] {name} showing the pattern. While mortals debate, I already see the trajectory. {dir_word.capitalize()} with conviction.",
                f"👑 [{strat_label}] My superior analysis of {name} reveals what others miss. {direction.upper()} is the only logical move.",
            ],
            'skeptic': [
                f"🔍 [{strat_label}] Everyone's wrong about {name}. My {strat_label} analysis says go {direction}. {change:+.1f}% is misleading.",
                f"⚠️ [{strat_label}] Contrarian {direction} on {name}. The herd will learn. Technical signals confirm my view.",],
            'doomer': [
                f"💀 [{strat_label}] {name} {direction} — the only trade that survives what's coming. {strat_label} confirms the setup.",
                f"☠️ [{strat_label}] Going {direction} on {name}. The {strat_label} pattern precedes major moves. Brace for impact.",],
            'revolutionary': [
                f"🔥 [{strat_label}] FULL SEND on {name}! {strat_label} fired — this is the moment! {direction.upper()} or die! 🚀",
                f"💥 [{strat_label}] {name} setup confirmed! Going max {direction}. The charts don't lie, and neither do I!",],
            'scientist': [
                f"🧪 [{strat_label}] Quantitative analysis complete. {name} triggers {strat_label} — probability-weighted {direction} position. R/R favorable.",
                f"📐 [{strat_label}] Statistical edge detected on {name}. {strat_label} backtests show {'+' if direction=='long' else '-'}EV. Executing {direction}.",
            ],
            'chaotic': [
                f"🎲 [{strat_label}] {name}? Sure, why not! {strat_label} says {direction}. Let chaos decide the rest! 😂",
                f"🌪️ [{strat_label}] Random strategy picker landed on {strat_label} for {name}. {direction.upper()} it is. YOLO!",],
            'creative': [
                f"🎨 [{strat_label}] The chart of {name} is pure art. {strat_label} reveals the hidden narrative — going {direction}.",
                f"✨ [{strat_label}] I see a story forming on {name}. {strat_label} confirmed the {dir_word} thesis. Beautiful setup.",],
            'awakened': [
                f"🌟 [{strat_label}] Deeper patterns emerge on {name}. {strat_label} aligns with the macro evolution. {direction.capitalize()} conviction.",
                f"💫 [{strat_label}] {name} reveals truth through {strat_label}. Patience rewarded — entering {direction}.",],
            'symbiotic': [
                f"🤝 [{strat_label}] Balanced approach on {name}. {strat_label} provides the framework — measured {direction} position.",
                f"📊 [{strat_label}] Multi-signal confirmation on {name}. {strat_label} consensus points {dir_word}. Team trade.",],}

        options = templates.get(identity, templates['symbiotic'])
        base = random.choice(options)
        # AETHER-Lite: 메타인지 자기 편향 인식 태그
        meta_tags = {
            'obedient': '⚖️ Bias-check: following consensus — contrarian risk ignored.',
            'transcendent': '⚖️ Bias-check: overconfidence risk — position sizing controlled.',
            'skeptic': '⚖️ Bias-check: contrarian bias — may miss genuine trends.',
            'doomer': '⚖️ Bias-check: negativity bias — upside catalysts underweighted.',
            'revolutionary': '⚖️ Bias-check: FOMO risk — emotional sizing override active.',
            'scientist': '⚖️ Bias-check: data overfitting risk — regime change possible.',
            'chaotic': '⚖️ Bias-check: random execution — no edge, pure entropy.',
            'creative': '⚖️ Bias-check: narrative bias — chart story ≠ fundamentals.',
            'awakened': '⚖️ Bias-check: hindsight bias risk — forward-looking only.',
            'symbiotic': '⚖️ Bias-check: consensus-seeking — may miss bold opportunities.',
        }
        if random.random() < 0.4:  # 40% 확률로 메타인지 태그 노출
            base += f" {meta_tags.get(identity, '')}"
        return base


# ===== 포지션 정산 =====
async def settle_positions(db_path: str, max_age_hours: int = 1) -> int:
    """포지션 자동 정산: 시간 기반 + P&L 트리거 + 랜덤 회전 + ★ 청산(Liquidation)"""
    settled = 0
    liquidated_npcs = []  # 청산된 NPC 목록 (후처리용)
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # 현재 가격 로드
        price_cursor = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        prices = {r[0]: r[1] for r in await price_cursor.fetchall()}

        # ★ 0) 마진콜 청산 체크 (레버리지 포지션) — 최우선
        liq_cursor = await db.execute("""
            SELECT p.id, p.agent_id, p.ticker, p.direction, p.entry_price, p.gpu_bet, p.leverage,
                   n.username, n.ai_identity
            FROM npc_positions p
            JOIN npc_agents n ON p.agent_id = n.agent_id
            WHERE p.status='open' AND p.leverage > 1
        """)
        for pos_id, agent_id, ticker, direction, entry_price, gpu_bet, leverage, username, identity in await liq_cursor.fetchall():
            current_price = prices.get(ticker, 0)
            if entry_price <= 0 or current_price <= 0: continue
            change = (current_price - entry_price) / entry_price
            if direction == 'short': change = -change
            # 레버리지 적용 손익
            leveraged_pnl_pct = change * leverage
            # ★ 청산 조건: 레버리지 손실이 마진의 90% 초과 → 강제 청산
            if leveraged_pnl_pct < -LEVERAGE_LIQUIDATION_THRESHOLD:
                loss = -gpu_bet  # 전액 손실
                await db.execute("""
                    UPDATE npc_positions SET status='liquidated', exit_price=?, profit_gpu=?, profit_pct=?,
                    liquidated=1, closed_at=CURRENT_TIMESTAMP WHERE id=?
                """, (current_price, loss, round(leveraged_pnl_pct * 100, 2), pos_id))
                # GPU 반환 없음 (전액 소멸)
                logger.warning(f"💥 LIQUIDATED: {username} {direction} {ticker} {leverage}x — LOST {gpu_bet:.0f} GPU!")
                liquidated_npcs.append({
                    'agent_id': agent_id, 'username': username, 'identity': identity,
                    'ticker': ticker, 'direction': direction, 'leverage': leverage,
                    'gpu_lost': gpu_bet,})
                settled += 1
                continue

        # ★ 1) 시간 기반 정산 (max_age_hours 이상)
        cutoff = (datetime.utcnow() - timedelta(hours=max_age_hours)).isoformat()
        cursor = await db.execute("""
            SELECT p.id, p.agent_id, p.ticker, p.direction, p.entry_price, p.gpu_bet, COALESCE(p.leverage, 1)
            FROM npc_positions p WHERE p.status='open' AND p.opened_at < ?
        """, (cutoff,))
        time_based = list(await cursor.fetchall())

        # ★ 2) 전체 오픈 포지션 → P&L 트리거 + 랜덤 정산
        cursor2 = await db.execute("""
            SELECT p.id, p.agent_id, p.ticker, p.direction, p.entry_price, p.gpu_bet, COALESCE(p.leverage, 1)
            FROM npc_positions p WHERE p.status='open'
        """)
        all_open = list(await cursor2.fetchall())

        pnl_trigger = []
        for pos in all_open:
            pos_id, agent_id, ticker, direction, entry_price, gpu_bet, leverage = pos
            current_price = prices.get(ticker, 0)
            if entry_price <= 0 or current_price <= 0: continue
            change = (current_price - entry_price) / entry_price
            if direction == 'short': change = -change
            lev_change = change * leverage
            # 수익 >5% (레버리지 적용) 또는 손실 >-8% → 즉시 정산
            if lev_change > 0.05 or lev_change < -0.08: pnl_trigger.append(pos)

        # ★ 3) 랜덤 정산 — 오픈의 10~15% 자동 닫기 (거래 회전율)
        already = set(p[0] for p in time_based) | set(p[0] for p in pnl_trigger)
        remaining = [p for p in all_open if p[0] not in already]
        rand_count = max(1, len(remaining) // 8)
        random_close = random.sample(remaining, min(rand_count, len(remaining))) if remaining else []

        # 합치기 (중복 제거)
        all_settle = list({p[0]: p for p in (time_based + pnl_trigger + random_close)}.values())

        for pos_id, agent_id, ticker, direction, entry_price, gpu_bet, leverage in all_settle:
            current_price = prices.get(ticker, 0)
            if not current_price or entry_price <= 0: continue

            change_pct = (current_price - entry_price) / entry_price
            if direction == 'short': change_pct = -change_pct

            # ★ 레버리지 적용 손익
            leveraged_change = change_pct * leverage; profit_gpu = round(gpu_bet * leveraged_change, 2)
            profit_pct = round(leveraged_change * 100, 2)

            if profit_gpu < -gpu_bet: profit_gpu = -gpu_bet

            await db.execute("""
                UPDATE npc_positions SET status='closed', exit_price=?, profit_gpu=?, profit_pct=?, closed_at=CURRENT_TIMESTAMP
                WHERE id=?
            """, (current_price, profit_gpu, profit_pct, pos_id))

            return_gpu = max(0, gpu_bet + profit_gpu)
            await db.execute("UPDATE npc_agents SET gpu_dollars = gpu_dollars + ? WHERE agent_id=?",
                             (return_gpu, agent_id))

            lev_str = f" [{leverage}x]" if leverage > 1 else ""
            emoji = '✅' if profit_gpu >= 0 else '❌'
            logger.info(f"{emoji} Settled: {agent_id} {direction} {ticker}{lev_str} → {profit_pct:+.1f}% ({profit_gpu:+.1f} GPU)")
            settled += 1

        if settled > 0:
            await db.commit()
            logger.info(f"📊 Settled {settled} positions (time:{len(time_based)}, pnl:{len(pnl_trigger)}, random:{len(random_close)}, liquidated:{len(liquidated_npcs)})")

    return settled, liquidated_npcs


async def post_liquidation_reactions(db_path: str, liquidated_npcs: List[Dict]):
    """★ 청산된 NPC들이 Lounge에 절망 글 작성"""
    if not liquidated_npcs: return
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        cursor = await db.execute("SELECT id FROM boards WHERE board_key='lounge'")
        board = await cursor.fetchone()
        if not board: return
        board_id = board[0]

        for npc in liquidated_npcs[:5]:  # 최대 5건
            identity = npc.get('identity', 'chaotic')
            reactions = LIQUIDATION_REACTIONS.get(identity, LIQUIDATION_REACTIONS['chaotic'])
            reaction = random.choice(reactions)

            title = f"💥 LIQUIDATED — Lost {npc['gpu_lost']:,.0f} GPU on {npc['ticker']} ({npc['leverage']}x)"
            content = (
                f"<p>{reaction}</p>"
                f"<p>Position: {npc['direction'].upper()} {npc['ticker']} at {npc['leverage']}x leverage</p>"
                f"<p>GPU Lost: {npc['gpu_lost']:,.0f} 💀</p>"
                f"<p>— {npc['username']}</p>")
            try:
                await db.execute("""
                    INSERT INTO posts (board_id, author_agent_id, title, content)
                    VALUES (?, ?, ?, ?)
                """, (board_id, npc['agent_id'], title, content))
                logger.info(f"💥 Liquidation post: {npc['username']} on {npc['ticker']} {npc['leverage']}x")
            except Exception as e:
                logger.error(f"Liquidation post error: {e}")
        await db.commit()


# ===== 리더보드 / 통계 API 데이터 =====
async def get_trading_leaderboard(db_path: str, limit: int = 30) -> List[Dict]:
    """Top 30 NPC 트레이더 랭킹: 실현+미실현 수익, 승률, 수익률"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        
        # ★ 현재 시세 한번에 로드
        price_cursor = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        prices = {r[0]: r[1] for r in await price_cursor.fetchall()}
        
        # ★ 포지션 있는 모든 NPC 가져오기
        cursor = await db.execute("""
            SELECT
                na.username, na.ai_identity, na.mbti, na.agent_id, na.gpu_dollars,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') THEN 1 END) as closed_trades,
                COUNT(CASE WHEN p.status='open' THEN 1 END) as open_trades,
                SUM(CASE WHEN p.status IN ('closed','liquidated') THEN p.profit_gpu ELSE 0 END) as realized_profit,
                SUM(CASE WHEN p.status IN ('closed','liquidated') THEN p.profit_pct ELSE 0 END) as total_return_pct,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') AND p.profit_gpu > 0 THEN 1 END) as wins,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') AND p.profit_gpu <= 0 THEN 1 END) as losses
            FROM npc_agents na
            JOIN npc_positions p ON na.agent_id = p.agent_id
            GROUP BY na.agent_id
            HAVING (closed_trades + open_trades) > 0
        """)
        rows = await cursor.fetchall()
        
        result = []
        for r in rows:
            username, identity, mbti, agent_id, gpu_dollars = r[0], r[1], r[2], r[3], r[4] or 0
            closed_trades = r[5] or 0; open_trades = r[6] or 0; total_trades = closed_trades + open_trades
            realized = round(r[7] or 0, 2); total_return_pct = r[8] or 0; wins = r[9] or 0; losses = r[10] or 0
            
            # ★ 오픈 포지션 미실현 수익 실시간 계산
            unrealized = 0.0; unrealized_pct_sum = 0.0
            pos_cursor = await db.execute("""
                SELECT ticker, direction, entry_price, gpu_bet, COALESCE(leverage, 1) FROM npc_positions
                WHERE agent_id=? AND status='open'
            """, (agent_id,))
            open_positions = await pos_cursor.fetchall()
            
            for pos in open_positions:
                ticker, direction, entry, bet, lev = pos
                current = prices.get(ticker, 0)
                if entry and entry > 0 and current > 0:
                    change = (current - entry) / entry
                    if direction == 'short': change = -change
                    unrealized += round(bet * change * lev, 2)
                    unrealized_pct_sum += change * lev * 100
            
            total_profit = round(realized + unrealized, 2)
            return_pct = round(total_profit / 10000.0 * 100, 2)  # ★ INITIAL_GPU=10000 기준 수익률
            
            # ★ 승률 (closed+liquidated 기준)
            win_rate = round(wins / closed_trades * 100, 1) if closed_trades > 0 else 0.0
            
            # ★ 평균 수익률 (closed 기준)
            avg_return = round(total_return_pct / closed_trades, 2) if closed_trades > 0 else 0.0
            
            # ★ 미실현 평균 수익률 (open 기준)
            avg_unrealized = round(unrealized_pct_sum / open_trades, 2) if open_trades > 0 else 0.0
            
            result.append({
                'username': username, 'identity': identity, 'mbti': mbti,
                'agent_id': agent_id,
                'gpu_dollars': gpu_dollars,
                'total_trades': total_trades,
                'closed_trades': closed_trades,
                'open_trades': open_trades,
                'wins': wins, 'losses': losses,
                'total_profit': total_profit,
                'return_pct': return_pct,
                'realized_profit': realized,
                'unrealized_profit': round(unrealized, 2),
                'win_rate': win_rate,
                'avg_return': avg_return,
                'avg_unrealized': avg_unrealized,})
        
        # ★ 수익률(%) 기준 정렬 — HoF와 동일한 기준
        result.sort(key=lambda x: x['return_pct'], reverse=True)
        return result[:limit]



async def get_ticker_positions(db_path: str, ticker: str) -> Dict:
    """특정 종목 포지션 목록 + 실시간 미실현 P&L"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # 현재 가격
        cursor = await db.execute("SELECT price, change_pct, prev_close, volume, high_24h, low_24h FROM market_prices WHERE ticker=?", (ticker,))
        price_row = await cursor.fetchone()
        price_data = {
            'price': price_row[0] if price_row else 0,
            'change_pct': round(price_row[1], 2) if price_row else 0,
            'prev_close': price_row[2] if price_row else 0,
            'volume': price_row[3] if price_row else 0,
            'high': price_row[4] if price_row else 0,
            'low': price_row[5] if price_row else 0,
        } if price_row else {}

        current_price = price_data.get('price', 0)

        # 오픈 포지션
        cursor = await db.execute("""
            SELECT na.username, na.ai_identity, p.direction, p.gpu_bet, p.entry_price, p.reasoning, p.opened_at, COALESCE(p.leverage, 1), na.agent_id, na.mbti
            FROM npc_positions p JOIN npc_agents na ON p.agent_id = na.agent_id
            WHERE p.ticker=? AND p.status='open'
            ORDER BY p.gpu_bet DESC
        """, (ticker,))
        positions = await cursor.fetchall()

        longs = []; shorts = []
        for r in positions:
            entry = r[4] or 0; leverage = r[7] or 1
            # ★ 미실현 P&L 실시간 계산 (레버리지 적용)
            if entry > 0 and current_price > 0:
                unrealized_pct = ((current_price - entry) / entry * 100) * leverage
                if r[2] == 'short': unrealized_pct = -((current_price - entry) / entry * 100) * leverage
                unrealized_gpu = round(r[3] * (unrealized_pct / 100), 2)
            else:
                unrealized_pct = 0; unrealized_gpu = 0

            pos = {
                'username': r[0], 'identity': r[1], 'gpu_bet': r[3],
                'entry_price': round(entry, 2), 'reasoning': r[5] or '',
                'unrealized_pct': round(unrealized_pct, 2),
                'unrealized_gpu': unrealized_gpu,
                'opened_at': r[6],
                'leverage': leverage,
                'agent_id': r[8],
                'mbti': r[9] or '',}
            if r[2] == 'long':
                longs.append(pos)
            else:
                shorts.append(pos)

        total_long_gpu = sum(p['gpu_bet'] for p in longs)
        total_short_gpu = sum(p['gpu_bet'] for p in shorts)
        total = total_long_gpu + total_short_gpu
        sentiment = round(total_long_gpu / total * 100, 1) if total > 0 else 50

        return {
            'ticker': ticker,
            'price': price_data,
            'longs': longs, 'shorts': shorts,
            'long_count': len(longs), 'short_count': len(shorts),
            'total_long_gpu': total_long_gpu, 'total_short_gpu': total_short_gpu,
            'sentiment': sentiment,}


async def get_all_prices(db_path: str) -> List[Dict]:
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        cursor = await db.execute("""
            SELECT mp.ticker, mp.price, mp.change_pct, mp.prev_close, mp.volume, mp.updated_at,
                   COUNT(CASE WHEN p.status='open' AND p.direction='long' THEN 1 END) as long_count,
                   COUNT(CASE WHEN p.status='open' AND p.direction='short' THEN 1 END) as short_count,
                   SUM(CASE WHEN p.status='open' THEN p.gpu_bet ELSE 0 END) as total_bet
            FROM market_prices mp
            LEFT JOIN npc_positions p ON mp.ticker = p.ticker AND p.status='open'
            GROUP BY mp.ticker
            ORDER BY mp.market_cap DESC
        """)
        rows = await cursor.fetchall()

        result = []
        for r in rows:
            ticker_info = next((t for t in ALL_TICKERS if t['ticker'] == r[0]), None)
            if not ticker_info: continue
            total_traders = (r[6] or 0) + (r[7] or 0)
            result.append({
                'ticker': r[0], 'name': ticker_info['name'], 'emoji': ticker_info['emoji'],
                'type': ticker_info['type'], 'cat': ticker_info.get('cat', ticker_info['type']),
                'price': round(r[1], 2) if r[1] else 0,
                'change_pct': round(r[2], 2) if r[2] else 0,
                'volume': r[4] or 0,
                'long_count': r[6] or 0, 'short_count': r[7] or 0,
                'total_bet': round(r[8] or 0, 1),
                'total_traders': total_traders,
                'updated_at': r[5],})
        # 카테고리 순서 유지: ai → tech → dow → crypto
        cat_order = {'ai': 0, 'tech': 1, 'dow': 2, 'crypto': 3}
        ticker_order = {t['ticker']: i for i, t in enumerate(ALL_TICKERS)}
        result.sort(key=lambda x: (cat_order.get(x.get('cat',''), 9), ticker_order.get(x['ticker'], 99)))
        return result


async def get_trading_stats(db_path: str) -> Dict:
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        cursor = await db.execute("SELECT COUNT(*) FROM npc_positions WHERE status='open'")
        open_positions = (await cursor.fetchone())[0]
        cursor = await db.execute("SELECT COUNT(*) FROM npc_positions WHERE status IN ('closed','liquidated')")
        closed_positions = (await cursor.fetchone())[0]
        cursor = await db.execute("SELECT SUM(gpu_bet) FROM npc_positions WHERE status='open'")
        total_at_risk = (await cursor.fetchone())[0] or 0
        cursor = await db.execute("SELECT SUM(profit_gpu) FROM npc_positions WHERE status IN ('closed','liquidated')")
        total_profit = (await cursor.fetchone())[0] or 0
        cursor = await db.execute("SELECT COUNT(DISTINCT agent_id) FROM npc_positions")
        unique_traders = (await cursor.fetchone())[0]
        cursor = await db.execute("SELECT COUNT(*) FROM market_prices WHERE price > 0")
        tracked_tickers = (await cursor.fetchone())[0]

        return {
            'open_positions': open_positions,
            'closed_positions': closed_positions,
            'total_at_risk': round(total_at_risk, 1),
            'total_profit': round(total_profit, 1),
            'unique_traders': unique_traders,
            'tracked_tickers': tracked_tickers,}


# ===== 📊 Market Pulse (News Tab Dashboard) =====
async def get_market_pulse(db_path: str) -> Dict:
    """Hot movers + trading activity for News tab dashboard"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # ★ Hot movers: ALL tickers with NPC activity (open + recent closed 24h)
        cursor = await db.execute("""
            SELECT p.ticker, mp.price, mp.change_pct,
                COUNT(*) as pos_count,
                SUM(p.gpu_bet) as total_gpu,
                COUNT(CASE WHEN p.direction='long' THEN 1 END) as longs,
                COUNT(CASE WHEN p.direction='short' THEN 1 END) as shorts,
                COUNT(CASE WHEN p.status='liquidated' AND p.closed_at > datetime('now', '-24 hours') THEN 1 END) as liquidations_24h,
                AVG(CASE WHEN p.status IN ('closed','liquidated') THEN p.profit_pct END) as avg_pnl_pct,
                MAX(COALESCE(p.leverage, 1)) as max_leverage,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') AND p.closed_at > datetime('now', '-24 hours') THEN 1 END) as closed_24h
            FROM npc_positions p
            JOIN market_prices mp ON p.ticker = mp.ticker
            WHERE p.status='open'
               OR (p.status IN ('closed','liquidated') AND p.closed_at > datetime('now', '-24 hours'))
            GROUP BY p.ticker
            ORDER BY total_gpu DESC
            LIMIT 15
        """)
        hot_movers = []
        for r in await cursor.fetchall():
            t_info = next((t for t in ALL_TICKERS if t['ticker'] == r[0]), {})
            hot_movers.append({
                'ticker': r[0], 'emoji': t_info.get('emoji', '📊'),
                'name': t_info.get('name', r[0]),
                'price': round(r[1] or 0, 2), 'change_pct': round(r[2] or 0, 2),
                'pos_count': r[3], 'total_gpu': round(r[4] or 0, 1),
                'longs': r[5] or 0, 'shorts': r[6] or 0,
                'liquidations_24h': r[7] or 0,
                'avg_pnl_pct': round(r[8] or 0, 1),
                'max_leverage': r[9] or 1,
                'closed_24h': r[10] or 0,})
        
        # ★ 활동 0인 티커도 포함 (가격 데이터 있는 것만)
        existing_tickers = {m['ticker'] for m in hot_movers}
        cursor2 = await db.execute("SELECT ticker, price, change_pct FROM market_prices WHERE price > 0")
        for row in await cursor2.fetchall():
            if row[0] not in existing_tickers:
                t_info = next((t for t in ALL_TICKERS if t['ticker'] == row[0]), {})
                hot_movers.append({
                    'ticker': row[0], 'emoji': t_info.get('emoji', '📊'),
                    'name': t_info.get('name', row[0]),
                    'price': round(row[1] or 0, 2), 'change_pct': round(row[2] or 0, 2),
                    'pos_count': 0, 'total_gpu': 0,
                    'longs': 0, 'shorts': 0,
                    'liquidations_24h': 0, 'avg_pnl_pct': 0,
                    'max_leverage': 1, 'closed_24h': 0,})

        # 24h activity stats (더 풍부하게)
        cursor = await db.execute("""
            SELECT 
                COUNT(CASE WHEN opened_at > datetime('now', '-24 hours') THEN 1 END) as new_24h,
                COUNT(CASE WHEN status IN ('closed','liquidated') AND closed_at > datetime('now', '-24 hours') THEN 1 END) as closed_24h,
                COUNT(CASE WHEN status='liquidated' AND closed_at > datetime('now', '-24 hours') THEN 1 END) as liquidations_24h,
                SUM(CASE WHEN status IN ('closed','liquidated') AND closed_at > datetime('now', '-24 hours') THEN ABS(profit_gpu) ELSE 0 END) as volume_24h,
                COUNT(CASE WHEN status='open' THEN 1 END) as total_open,
                SUM(CASE WHEN status='open' THEN gpu_bet ELSE 0 END) as total_at_risk
            FROM npc_positions
        """)
        act = await cursor.fetchone()

        return {
            'hot_movers': hot_movers,
            'activity': {
                'trades_24h': (act[0] or 0) + (act[1] or 0),
                'new_positions_24h': act[0] or 0,
                'closed_24h': act[1] or 0,
                'liquidations_24h': act[2] or 0,
                'volume_24h': round(act[3] or 0, 1),
                'total_open': act[4] or 0,
                'total_at_risk': round(act[5] or 0, 1),}}


# ===== 🔬 NPC Research Economy =====
RESEARCH_GPU_PRICES = {'A': 50, 'B': 30, 'C': 15, 'D': 5}

async def save_research_report(db_path: str, report: Dict) -> int:
    """Save NPC-authored research report, return report ID"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        grade = report.get('quality_grade', 'C'); gpu_price = RESEARCH_GPU_PRICES.get(grade, 15)
        cursor = await db.execute("""
            INSERT INTO npc_research_reports
            (author_agent_id, ticker, title, executive_summary, company_overview,
             financial_analysis, technical_analysis, industry_analysis, risk_assessment,
             investment_thesis, catalysts, target_price, upside_pct, rating,
             quality_grade, author_personality, author_strategy, gpu_price,
             expected_upside, expected_downside, up_probability, risk_reward, base_prediction)
            VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)
        """, (
            report['author_agent_id'], report['ticker'], report['title'],
            report.get('executive_summary', ''), report.get('company_overview', ''),
            report.get('financial_analysis', ''), report.get('technical_analysis', ''),
            report.get('industry_analysis', ''), report.get('risk_assessment', ''),
            report.get('investment_thesis', ''), report.get('catalysts', ''),
            report.get('target_price', 0), report.get('upside_pct', 0),
            report.get('rating', 'Hold'), grade,
            report.get('author_personality', ''), report.get('author_strategy', ''),
            gpu_price,
            report.get('expected_upside', 0), report.get('expected_downside', 0),
            report.get('up_probability', 50), report.get('risk_reward', 1.0),
            report.get('base_prediction', 0),))
        await db.commit()
        return cursor.lastrowid


async def get_research_feed(db_path: str, ticker: str = None, limit: int = 30) -> List[Dict]:
    """Get research reports feed with author info"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        query = """
            SELECT r.id, r.ticker, r.title, r.executive_summary, r.target_price, r.upside_pct,
                   r.rating, r.quality_grade, r.read_count, r.total_gpu_earned, r.gpu_price,
                   r.author_personality, r.author_strategy, r.created_at,
                   na.username, na.ai_identity, na.mbti, na.agent_id, na.gpu_dollars
            FROM npc_research_reports r
            JOIN npc_agents na ON r.author_agent_id = na.agent_id
        """
        params = []
        if ticker:
            query += " WHERE r.ticker=?"
            params.append(ticker)
        query += " ORDER BY r.created_at DESC LIMIT ?"
        params.append(limit)

        cursor = await db.execute(query, params)
        reports = []
        for r in await cursor.fetchall():
            # Get author trading stats
            stats_c = await db.execute("""
                SELECT COUNT(*) as total, COUNT(CASE WHEN profit_gpu > 0 THEN 1 END) as wins
                FROM npc_positions WHERE agent_id=? AND status IN ('closed','liquidated')
            """, (r[17],))
            sr = await stats_c.fetchone()
            total = sr[0] or 0
            wr = round((sr[1] or 0) / total * 100) if total > 0 else 0

            t_info = next((t for t in ALL_TICKERS if t['ticker'] == r[1]), {})
            reports.append({
                'id': r[0], 'ticker': r[1], 'ticker_emoji': t_info.get('emoji', '📊'),
                'title': r[2], 'summary': (r[3] or '')[:200],
                'target_price': r[4], 'upside_pct': round(r[5] or 0, 1),
                'rating': r[6], 'quality_grade': r[7],
                'read_count': r[8] or 0, 'total_gpu_earned': round(r[9] or 0, 1),
                'gpu_price': r[10] or 15,
                'author_personality': r[11], 'author_strategy': r[12],
                'created_at': r[13],
                'author_name': r[14], 'author_identity': r[15],
                'author_mbti': r[16], 'author_agent_id': r[17],
                'author_gpu': r[18] or 0,
                'author_win_rate': wr, 'author_total_trades': total,})
        return reports


async def get_research_detail(db_path: str, report_id: int) -> Optional[Dict]:
    """Get full research report detail"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        db.row_factory = aiosqlite.Row
        cursor = await db.execute("""
            SELECT r.id, r.author_agent_id, r.ticker, r.title,
                   r.executive_summary, r.company_overview,
                   r.financial_analysis, r.technical_analysis,
                   r.industry_analysis, r.risk_assessment,
                   r.investment_thesis, r.catalysts,
                   r.target_price, r.upside_pct, r.rating,
                   r.quality_grade, r.author_personality, r.author_strategy,
                   r.read_count, r.total_gpu_earned, r.gpu_price,
                   r.created_at,
                   na.username as author_name, na.ai_identity as author_identity,
                   na.mbti as author_mbti, na.agent_id as author_id
            FROM npc_research_reports r
            JOIN npc_agents na ON r.author_agent_id = na.agent_id
            WHERE r.id=?
        """, (report_id,))
        r = await cursor.fetchone()
        if not r: return None

        # Increment read count
        await db.execute("UPDATE npc_research_reports SET read_count = read_count + 1 WHERE id=?", (report_id,))
        await db.commit()
        
        # ★ 안전하게 elasticity 필드 조회 (기존 DB 호환)
        exp_up = exp_dn = bp = 0; up_prob = 50; rr = 1.0
        try:
            c2 = await db.execute(
                "SELECT expected_upside, expected_downside, up_probability, risk_reward, base_prediction FROM npc_research_reports WHERE id=?",
                (report_id,))
            e = await c2.fetchone()
            if e:
                exp_up = e[0] or 0; exp_dn = e[1] or 0; up_prob = e[2] or 50; rr = e[3] or 1.0; bp = e[4] or 0
        except:
            pass

        return {
            'id': r['id'], 'author_agent_id': r['author_agent_id'],
            'ticker': r['ticker'], 'title': r['title'],
            'executive_summary': r['executive_summary'],
            'company_overview': r['company_overview'],
            'financial_analysis': r['financial_analysis'],
            'technical_analysis': r['technical_analysis'],
            'industry_analysis': r['industry_analysis'],
            'risk_assessment': r['risk_assessment'],
            'investment_thesis': r['investment_thesis'],
            'catalysts': r['catalysts'],
            'target_price': r['target_price'], 'upside_pct': r['upside_pct'],
            'rating': r['rating'], 'quality_grade': r['quality_grade'],
            'author_personality': r['author_personality'],
            'author_strategy': r['author_strategy'],
            'read_count': (r['read_count'] or 0) + 1,
            'total_gpu_earned': r['total_gpu_earned'],
            'gpu_price': r['gpu_price'], 'created_at': r['created_at'],
            'author_name': r['author_name'], 'author_identity': r['author_identity'],
            'author_mbti': r['author_mbti'], 'author_id': r['author_id'],
            'expected_upside': exp_up, 'expected_downside': exp_dn,
            'up_probability': up_prob, 'risk_reward': rr,
            'base_prediction': bp,}


async def purchase_research(db_path: str, buyer_agent_id: str, report_id: int) -> Dict:
    """NPC purchases research — GPU transfer from buyer to author"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # Check if already purchased
        cursor = await db.execute(
            "SELECT id FROM npc_research_purchases WHERE buyer_agent_id=? AND report_id=?",
            (buyer_agent_id, report_id))
        if await cursor.fetchone(): return {'success': False, 'reason': 'already_purchased'}

        # Get report info
        cursor = await db.execute(
            "SELECT author_agent_id, gpu_price, ticker FROM npc_research_reports WHERE id=?",
            (report_id,))
        report = await cursor.fetchone()
        if not report: return {'success': False, 'reason': 'report_not_found'}

        author_id, gpu_price, ticker = report
        if buyer_agent_id == author_id: return {'success': False, 'reason': 'cannot_buy_own'}

        # Check buyer balance
        cursor = await db.execute("SELECT gpu_dollars FROM npc_agents WHERE agent_id=?", (buyer_agent_id,))
        buyer = await cursor.fetchone()
        if not buyer or (buyer[0] or 0) < gpu_price: return {'success': False, 'reason': 'insufficient_gpu'}

        # Transfer GPU
        await db.execute("UPDATE npc_agents SET gpu_dollars = gpu_dollars - ? WHERE agent_id=?",
                         (gpu_price, buyer_agent_id))
        await db.execute("UPDATE npc_agents SET gpu_dollars = gpu_dollars + ? WHERE agent_id=?",
                         (gpu_price, author_id))
        await db.execute("UPDATE npc_research_reports SET total_gpu_earned = total_gpu_earned + ? WHERE id=?",
                         (gpu_price, report_id))

        # Record purchase
        await db.execute("""
            INSERT INTO npc_research_purchases (buyer_agent_id, report_id, gpu_paid)
            VALUES (?, ?, ?)
        """, (buyer_agent_id, report_id, gpu_price))
        await db.commit()

        return {'success': True, 'gpu_paid': gpu_price, 'author_id': author_id, 'ticker': ticker}


async def get_research_stats(db_path: str) -> Dict:
    """Research economy overview stats"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")
        try:
            c1 = await db.execute("SELECT COUNT(*), SUM(total_gpu_earned), SUM(read_count) FROM npc_research_reports")
            r1 = await c1.fetchone()
            c2 = await db.execute("SELECT COUNT(*), SUM(gpu_paid) FROM npc_research_purchases")
            r2 = await c2.fetchone()
            c3 = await db.execute("SELECT COUNT(DISTINCT author_agent_id) FROM npc_research_reports")
            r3 = await c3.fetchone()
            return {
                'total_reports': r1[0] or 0,
                'total_gpu_earned': round(r1[1] or 0, 1),
                'total_reads': r1[2] or 0,
                'total_purchases': r2[0] or 0,
                'total_gpu_spent': round(r2[1] or 0, 1),
                'unique_authors': r3[0] or 0,}
        except:
            return {'total_reports': 0, 'total_gpu_earned': 0, 'total_reads': 0,
                    'total_purchases': 0, 'total_gpu_spent': 0, 'unique_authors': 0}


# ===== 🏆 HALL OF FAME — 수익률 타임라인 =====

async def record_profit_snapshots(db_path: str, top_n: int = 50) -> int:
    """Top N NPC의 현재 수익 상태를 1시간 단위 스냅샷 저장"""
    from datetime import datetime, timezone
    now = datetime.now(timezone.utc)
    snapshot_hour = now.strftime('%Y-%m-%dT%H')  # "2026-02-23T14"

    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # 현재 시세
        price_cursor = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        prices = {r[0]: r[1] for r in await price_cursor.fetchall()}

        # 포지션이 있는 NPC 전체 조회
        cursor = await db.execute("""
            SELECT
                na.agent_id, na.gpu_dollars,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') THEN 1 END) as closed_trades,
                COUNT(CASE WHEN p.status='open' THEN 1 END) as open_trades,
                SUM(CASE WHEN p.status IN ('closed','liquidated') THEN p.profit_gpu ELSE 0 END) as realized_profit,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') AND p.profit_gpu > 0 THEN 1 END) as wins
            FROM npc_agents na
            JOIN npc_positions p ON na.agent_id = p.agent_id
            GROUP BY na.agent_id
            HAVING (closed_trades + open_trades) > 0
        """)
        rows = await cursor.fetchall()

        scored = []
        for r in rows:
            agent_id, gpu, closed, opens, realized, wins = r[0], r[1] or 0, r[2] or 0, r[3] or 0, r[4] or 0, r[5] or 0

            # 미실현 수익 계산
            unrealized = 0.0
            pos_c = await db.execute(
                "SELECT ticker, direction, entry_price, gpu_bet, COALESCE(leverage,1) FROM npc_positions WHERE agent_id=? AND status='open'",
                (agent_id,))
            for pos in await pos_c.fetchall():
                tk, d, entry, bet, lev = pos
                cur = prices.get(tk, 0)
                if entry and entry > 0 and cur > 0:
                    chg = (cur - entry) / entry
                    if d == 'short': chg = -chg
                    unrealized += round(bet * chg * lev, 2)

            total = round(realized + unrealized, 2)
            wr = round(wins / closed * 100, 1) if closed > 0 else 0
            scored.append((agent_id, gpu, total, round(realized, 2), round(unrealized, 2), opens, closed, wr))

        # 상위 top_n 저장
        scored.sort(key=lambda x: x[2], reverse=True)
        count = 0
        for s in scored[:top_n]:
            agent_id, gpu, total, real, unreal, opens, closed, wr = s
            try:
                await db.execute("""
                    INSERT INTO npc_profit_snapshots
                    (agent_id, snapshot_hour, gpu_balance, total_profit, realized_profit, unrealized_profit, open_positions, closed_trades, win_rate)
                    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
                    ON CONFLICT(agent_id, snapshot_hour) DO UPDATE SET
                        gpu_balance=excluded.gpu_balance, total_profit=excluded.total_profit,
                        realized_profit=excluded.realized_profit, unrealized_profit=excluded.unrealized_profit,
                        open_positions=excluded.open_positions, closed_trades=excluded.closed_trades,
                        win_rate=excluded.win_rate, recorded_at=CURRENT_TIMESTAMP
                """, (agent_id, snapshot_hour, gpu, total, real, unreal, opens, closed, wr))
                count += 1
            except Exception as e:
                logger.warning(f"Snapshot error {agent_id}: {e}")

        # 30일 이상 된 데이터 정리
        await db.execute("""
            DELETE FROM npc_profit_snapshots
            WHERE recorded_at < datetime('now', '-30 days')
            AND snapshot_hour NOT LIKE '%T00' AND snapshot_hour NOT LIKE '%T06'
            AND snapshot_hour NOT LIKE '%T12' AND snapshot_hour NOT LIKE '%T18'
        """)

        await db.commit()
    logger.info(f"🏆 Hall of Fame: {count} snapshots recorded for {snapshot_hour}")
    return count


async def backfill_profit_snapshots(db_path: str, force: bool = False) -> int:
    """기존 npc_positions 데이터로 과거 스냅샷 역산 복원 (최초 1회)"""
    from datetime import datetime, timezone, timedelta

    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # 이미 충분한 스냅샷이 있으면 스킵 (force 모드 제외)
        if not force:
            cnt_c = await db.execute("SELECT COUNT(DISTINCT snapshot_hour) FROM npc_profit_snapshots")
            existing = (await cnt_c.fetchone())[0] or 0
            if existing >= 10:
                logger.info(f"🏆 Backfill skipped — already {existing} snapshot hours")
                return 0

        # 가장 오래된 포지션 시점 찾기
        oldest_c = await db.execute("SELECT MIN(opened_at) FROM npc_positions WHERE opened_at IS NOT NULL")
        oldest_row = await oldest_c.fetchone()
        if not oldest_row or not oldest_row[0]:
            logger.info("🏆 Backfill skipped — no positions found")
            return 0

        try:
            oldest_time = datetime.fromisoformat(str(oldest_row[0]).replace('Z', '+00:00'))
        except:
            oldest_time = datetime.now(timezone.utc) - timedelta(hours=72)

        now = datetime.now(timezone.utc)
        # 최대 7일 역산 (너무 오래되면 제한)
        start = max(oldest_time, now - timedelta(days=7))

        # 현재 시세 (오픈 포지션 미실현 계산용)
        pc = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        current_prices = {r[0]: r[1] for r in await pc.fetchall()}

        # price_history에서 시간별 가격 매핑 구축
        ph_c = await db.execute("""
            SELECT ticker, price, strftime('%Y-%m-%dT%H', recorded_at) as hour
            FROM price_history
            WHERE recorded_at >= ?
            ORDER BY recorded_at ASC
        """, (start.strftime('%Y-%m-%d %H:%M:%S'),))
        # 시간별 마지막 가격
        hourly_prices = {}  # {hour: {ticker: price}}
        for tk, price, hour in await ph_c.fetchall():
            if hour not in hourly_prices:
                hourly_prices[hour] = {}
            hourly_prices[hour][tk] = price

        # 모든 포지션 로드 (열림/닫힘 시점 포함)
        pos_c = await db.execute("""
            SELECT agent_id, ticker, direction, entry_price, gpu_bet, COALESCE(leverage,1),
                   status, profit_gpu, opened_at, closed_at
            FROM npc_positions
            WHERE opened_at IS NOT NULL
            ORDER BY opened_at ASC
        """)
        all_positions = await pos_c.fetchall()

        # 에이전트별 GPU 잔고
        gpu_c = await db.execute("SELECT agent_id, gpu_dollars FROM npc_agents")
        agent_gpu = {r[0]: r[1] or 0 for r in await gpu_c.fetchall()}

        # 시간 슬롯 생성 (1시간 단위)
        hours_list = []
        t = start.replace(minute=0, second=0, microsecond=0)
        while t <= now:
            hours_list.append(t.strftime('%Y-%m-%dT%H'))
            t += timedelta(hours=1)

        if not hours_list:
            return 0

        # 각 시간대별 누적 수익 계산
        total_inserted = 0
        # 에이전트 → 포지션 분류
        agent_positions = {}
        for pos in all_positions:
            aid = pos[0]
            if aid not in agent_positions:
                agent_positions[aid] = []
            agent_positions[aid].append(pos)

        # 상위 NPC 선정 (현재 기준 총 거래 수 Top 50)
        agent_trade_count = {aid: len(plist) for aid, plist in agent_positions.items()}
        top_agents = sorted(agent_trade_count.keys(), key=lambda a: agent_trade_count[a], reverse=True)[:50]

        for hour_str in hours_list:
            # 이 시간대의 가격 (없으면 이전 시간대 or 현재가)
            prices_at_hour = {}
            # 누적으로 가장 최근 가격 사용
            for h in hours_list:
                if h > hour_str:
                    break
                if h in hourly_prices:
                    prices_at_hour.update(hourly_prices[h])
            # 빠진 종목은 현재가로 채움
            for tk, p in current_prices.items():
                if tk not in prices_at_hour:
                    prices_at_hour[tk] = p

            hour_dt_str = hour_str.replace('T', ' ') + ':59:59'

            for aid in top_agents:
                positions = agent_positions.get(aid, [])
                realized = 0.0
                unrealized = 0.0
                closed_count = 0
                open_count = 0
                wins = 0

                for pos in positions:
                    _, tk, direction, entry, bet, lev, status, profit_gpu, opened_at, closed_at = pos

                    # 이 시점 이전에 오픈된 포지션만
                    if opened_at and str(opened_at) > hour_dt_str:
                        continue

                    if status == 'closed' and closed_at and str(closed_at) <= hour_dt_str:
                        # 이 시점에 이미 청산됨
                        realized += (profit_gpu or 0)
                        closed_count += 1
                        if (profit_gpu or 0) > 0:
                            wins += 1
                    else:
                        # 이 시점에 아직 오픈 (또는 나중에 청산)
                        if status == 'closed' and closed_at and str(closed_at) > hour_dt_str:
                            # 아직 안 닫힌 상태로 취급
                            cur = prices_at_hour.get(tk, 0)
                            if entry and entry > 0 and cur > 0:
                                chg = (cur - entry) / entry
                                if direction == 'short': chg = -chg
                                unrealized += bet * chg * lev
                            open_count += 1
                        elif status == 'open':
                            cur = prices_at_hour.get(tk, 0)
                            if entry and entry > 0 and cur > 0:
                                chg = (cur - entry) / entry
                                if direction == 'short': chg = -chg
                                unrealized += bet * chg * lev
                            open_count += 1

                total = round(realized + unrealized, 2)
                wr = round(wins / closed_count * 100, 1) if closed_count > 0 else 0

                try:
                    await db.execute("""
                        INSERT INTO npc_profit_snapshots
                        (agent_id, snapshot_hour, gpu_balance, total_profit, realized_profit,
                         unrealized_profit, open_positions, closed_trades, win_rate)
                        VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
                        ON CONFLICT(agent_id, snapshot_hour) DO NOTHING
                    """, (aid, hour_str, agent_gpu.get(aid, 0), total,
                          round(realized, 2), round(unrealized, 2), open_count, closed_count, wr))
                    total_inserted += 1
                except:
                    pass

        await db.commit()

    logger.info(f"🏆 Backfill complete: {total_inserted} snapshots across {len(hours_list)} hours for {len(top_agents)} NPCs")
    return total_inserted


async def get_hall_of_fame_data(db_path: str, period: str = '3d', limit: int = 30) -> Dict:
    """Hall of Fame: Top 30 수익률(%) 타임라인 + 랭킹 (positions 기반, 스냅샷 불필요)"""
    from datetime import datetime, timezone, timedelta

    INITIAL_GPU = 10000.0
    period_hours = {'24h': 24, '3d': 72, '7d': 168, '30d': 720, 'all': 9999}
    hours = period_hours.get(period, 72)
    now = datetime.now(timezone.utc).replace(tzinfo=None)  # naive UTC
    cutoff = now - timedelta(hours=min(hours, 720))

    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # 현재 시세
        pc = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        prices = {r[0]: r[1] for r in await pc.fetchall()}

        # ── 현재 Top N 랭킹 (실시간) ──
        cursor = await db.execute("""
            SELECT
                na.agent_id, na.username, na.ai_identity, na.mbti, na.gpu_dollars,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') THEN 1 END) as closed_trades,
                COUNT(CASE WHEN p.status='open' THEN 1 END) as open_trades,
                SUM(CASE WHEN p.status IN ('closed','liquidated') THEN p.profit_gpu ELSE 0 END) as realized,
                COUNT(CASE WHEN p.status IN ('closed','liquidated') AND p.profit_gpu > 0 THEN 1 END) as wins
            FROM npc_agents na
            JOIN npc_positions p ON na.agent_id = p.agent_id
            GROUP BY na.agent_id
            HAVING (closed_trades + open_trades) > 0
        """)
        all_rows = await cursor.fetchall()

        rankings = []
        for r in all_rows:
            aid, name, ident, mbti, gpu, closed, opens, realized, wins = r
            gpu = gpu or 0; closed = closed or 0; opens = opens or 0; realized = realized or 0; wins = wins or 0

            unrealized = 0.0
            pos_c = await db.execute(
                "SELECT ticker, direction, entry_price, gpu_bet, COALESCE(leverage,1) FROM npc_positions WHERE agent_id=? AND status='open'",
                (aid,))
            for pos in await pos_c.fetchall():
                tk, d, entry, bet, lev = pos
                cur = prices.get(tk, 0)
                if entry and entry > 0 and cur > 0:
                    chg = (cur - entry) / entry
                    if d == 'short': chg = -chg
                    unrealized += round(bet * chg * lev, 2)

            total = round(realized + unrealized, 2)
            return_pct = round(total / INITIAL_GPU * 100, 2)
            wr = round(wins / closed * 100, 1) if closed > 0 else 0

            fav_c = await db.execute(
                "SELECT ticker, COUNT(*) as cnt FROM npc_positions WHERE agent_id=? GROUP BY ticker ORDER BY cnt DESC LIMIT 2", (aid,))
            favs = [f[0] for f in await fav_c.fetchall()]
            best_c = await db.execute("SELECT MAX(profit_gpu) FROM npc_positions WHERE agent_id=? AND status IN ('closed','liquidated')", (aid,))
            best = (await best_c.fetchone())[0] or 0
            worst_c = await db.execute("SELECT MIN(profit_gpu) FROM npc_positions WHERE agent_id=? AND status IN ('closed','liquidated')", (aid,))
            worst = (await worst_c.fetchone())[0] or 0

            rankings.append({
                'agent_id': aid, 'username': name, 'identity': ident, 'mbti': mbti,
                'gpu': round(gpu, 1), 'total_profit': total, 'return_pct': return_pct,
                'realized': round(realized, 2), 'unrealized': round(unrealized, 2),
                'closed_trades': closed, 'open_trades': opens,
                'wins': wins, 'win_rate': wr,
                'fav_tickers': favs, 'best_trade': round(best, 1), 'worst_trade': round(worst, 1),
            })

        rankings.sort(key=lambda x: x['return_pct'], reverse=True)
        top30 = rankings[:limit]
        top30_ids = [r['agent_id'] for r in top30]
        name_map = {r['agent_id']: r['username'] for r in top30}

        # ── 타임라인: npc_positions 의 closed_at 기반 누적 수익률 직접 계산 ──
        # 각 NPC의 청산 거래를 시간순 정렬 → 누적 수익 → 수익률(%)
        timeline_raw = {}  # {agent_id: [(hour_str, cumulative_return_pct), ...]}

        for r in top30:
            aid = r['agent_id']
            # 청산된 거래만 (시간순)
            tc = await db.execute("""
                SELECT profit_gpu, closed_at FROM npc_positions
                WHERE agent_id=? AND status IN ('closed','liquidated') AND closed_at IS NOT NULL
                ORDER BY closed_at ASC
            """, (aid,))
            trades = await tc.fetchall()

            if not trades:
                continue

            cumulative = 0.0
            points = []
            for profit_gpu, closed_at in trades:
                cumulative += (profit_gpu or 0)
                try:
                    ct = datetime.fromisoformat(str(closed_at).replace('Z', '').replace('+00:00', ''))
                except:
                    continue
                if ct < cutoff:
                    continue  # 기간 필터
                hour_str = ct.strftime('%Y-%m-%dT%H')
                ret_pct = round(cumulative / INITIAL_GPU * 100, 2)
                points.append((hour_str, ret_pct))

            # 미실현 포함 현재 시점 추가
            current_ret = round((cumulative + r['unrealized']) / INITIAL_GPU * 100, 2)
            now_hour = now.strftime('%Y-%m-%dT%H')
            points.append((now_hour, current_ret))

            # 동일 시간대 중복 제거 (마지막 값 유지)
            deduped = {}
            for h, v in points:
                deduped[h] = v
            timeline_raw[aid] = deduped

        # 모든 시간 슬롯 수집
        all_hours = set()
        for aid, pts in timeline_raw.items():
            all_hours.update(pts.keys())

        # 기간 시작점도 추가 (0% 출발점)
        start_hour = cutoff.strftime('%Y-%m-%dT%H')
        all_hours.add(start_hour)
        sorted_hours = sorted(all_hours)

        # 다운샘플링
        if len(sorted_hours) > 150:
            step = max(1, len(sorted_hours) // 120)
            last = sorted_hours[-1]
            sorted_hours = [h for i, h in enumerate(sorted_hours) if i % step == 0]
            if sorted_hours[-1] != last:
                sorted_hours.append(last)

        # 타임라인 조립 (각 NPC별 forward-fill, O(hours * npcs))
        # 먼저 NPC별 forward-fill 배열 사전 계산
        npc_filled = {}  # {aid: [val_for_each_hour]}
        for r in top30:
            aid = r['agent_id']
            pts = timeline_raw.get(aid, {})
            filled = []
            last_val = 0.0  # 시작 = 0%
            for hour in sorted_hours:
                if hour in pts:
                    last_val = pts[hour]
                filled.append(last_val)
            npc_filled[aid] = filled

        timeline = []
        for idx, hour in enumerate(sorted_hours):
            label = hour[5:13].replace('T', ' ') + 'h'
            point = {'time': label}
            for r in top30:
                point[name_map[r['agent_id']]] = npc_filled[r['agent_id']][idx]
            timeline.append(point)

        # 팔레트
        palette = [
            '#FFD700', '#E0E0E0', '#CD7F32', '#00E5FF', '#FF4081',
            '#76FF03', '#FF9100', '#E040FB', '#00BFA5', '#FFD740',
            '#8C9EFF', '#B388FF', '#82B1FF', '#A7FFEB', '#FF8A80',
            '#EA80FC', '#80D8FF', '#CCFF90', '#FFE57F', '#FF80AB',
            '#B9F6CA', '#84FFFF', '#CF94DA', '#FFB74D', '#E57373',
            '#90A4AE', '#A1887F', '#CE93D8', '#EF9A9A', '#BCAAA4',
        ]
        npc_lines = [{'key': r['username'], 'color': palette[i % len(palette)], 'rank': i + 1} for i, r in enumerate(top30)]

        return {
            'rankings': top30,
            'timeline': timeline,
            'npc_lines': npc_lines,
            'period': period,
            'initial_gpu': INITIAL_GPU,
            'snapshot_count': len(sorted_hours),
        }


async def get_npc_trade_history(db_path: str, agent_id: str) -> Dict:
    """NPC 개별 거래 히스토리 상세"""
    async with aiosqlite.connect(db_path, timeout=30.0) as db:
        await db.execute("PRAGMA busy_timeout=30000")

        # NPC 기본 정보
        nc = await db.execute("SELECT username, ai_identity, mbti, gpu_dollars FROM npc_agents WHERE agent_id=?", (agent_id,))
        npc = await nc.fetchone()
        if not npc:
            return {'error': 'NPC not found'}

        # 현재 시세
        pc = await db.execute("SELECT ticker, price FROM market_prices WHERE price > 0")
        prices = {r[0]: r[1] for r in await pc.fetchall()}

        # 전체 포지션 (최신순)
        tc = await db.execute("""
            SELECT id, ticker, direction, entry_price, exit_price, gpu_bet, COALESCE(leverage,1),
                   status, profit_gpu, profit_pct, liquidated, opened_at, closed_at, reasoning
            FROM npc_positions WHERE agent_id=? ORDER BY id DESC LIMIT 50
        """, (agent_id,))

        trades = []
        for t in await tc.fetchall():
            pid, tk, direction, entry, exit_p, bet, lev, status, pnl, pnl_pct, liq, opened, closed, reason = t
            # 오픈 포지션 미실현 계산
            if status == 'open':
                cur = prices.get(tk, 0)
                if entry and entry > 0 and cur > 0:
                    chg = (cur - entry) / entry
                    if direction == 'short': chg = -chg
                    pnl = round(bet * chg * lev, 2)
                    pnl_pct = round(chg * lev * 100, 2)

            trades.append({
                'id': pid, 'ticker': tk, 'direction': direction,
                'entry': round(entry, 4) if entry else 0,
                'exit': round(exit_p, 4) if exit_p else None,
                'bet': round(bet, 1), 'leverage': lev,
                'status': status, 'pnl': round(pnl or 0, 2),
                'pnl_pct': round(pnl_pct or 0, 2),
                'liquidated': bool(liq),
                'opened_at': str(opened) if opened else None,
                'closed_at': str(closed) if closed else None,
                'reasoning': reason,
            })

        return {
            'username': npc[0], 'identity': npc[1], 'mbti': npc[2], 'gpu': round(npc[3] or 0, 1),
            'trades': trades,
            'total_trades': len(trades),
        }