trump-code-market-signals — Прогнозирование S&P 500 по публикациям Трампа

--- name: trump-code-market-signals description: AI-powered analysis of Trump's social media posts to predict stock market movements using 31.5M brute-force tested rules triggers: - analyze trump posts for market signals - run trump code prediction - check today's trading signals from trump - decode trump truth social posts - trump market signal analysis - run overnight brute force model search - check surviving trading rules - trump code cli predict --- # Trump Code — Market Signal Analysis > Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection. Trump Code is an open-source system that applies brute-force computation to find statistically significant patterns between Trump's Truth Social/X posting behavior and S&P 500 movements. It has tested 31.5M model combinations, maintains 551 surviving rules, and has a verified 61.3% hit rate across 566 predictions (z=5.39, p<0.05). ## Installation ```bash git clone https://github.com/sstklen/trump-code.git cd trump-code pip install -r requirements.txt ``` ### Environment Variables ```bash # Required for AI briefing and chatbot export GEMINI_KEYS="key1,key2,key3" # Comma-separated Gemini API keys # Optional: for Claude Opus deep analysis export ANTHROPIC_API_KEY="your-key-here" # Optional: for Polymarket/Kalshi integration export POLYMARKET_API_KEY="your-key-here" ``` ## CLI — Key Commands ```bash # Today's detected signals from Trump's posts python3 trump_code_cli.py signals # Model performance leaderboard (all 11 named models) python3 trump_code_cli.py models # Get LONG/SHORT consensus prediction python3 trump_code_cli.py predict # Prediction market arbitrage opportunities python3 trump_code_cli.py arbitrage # System health check (circuit breaker state) python3 trump_code_cli.py health # Full daily report (trilingual) python3 trump_code_cli.py report # Dump all data as JSON python3 trump_code_cli.py json ``` ## Core Scripts ```bash # Real-time Trump post monitor (polls every 5 min) python3 realtime_loop.py # Brute-force model search (~25 min, tests millions of combos) python3 overnight_search.py # Individual analyses python3 analysis_06_market.py # Posts vs S&P 500 correlation python3 analysis_09_combo_score.py # Multi-signal combo scoring # Web dashboard + AI chatbot on port 8888 export GEMINI_KEYS="key1,key2,key3" python3 chatbot_server.py # → http://localhost:8888 ``` ## REST API (Live at trumpcode.washinmura.jp) ```python import requests BASE = "https://trumpcode.washinmura.jp" # All dashboard data in one call data = requests.get(f"{BASE}/api/dashboard").json() # Today's signals + 7-day history signals = requests.get(f"{BASE}/api/signals").json() # Model performance rankings models = requests.get(f"{BASE}/api/models").json() # Latest 20 Trump posts with signal tags posts = requests.get(f"{BASE}/api/recent-posts").json() # Live Polymarket Trump prediction markets (316+) markets = requests.get(f"{BASE}/api/polymarket-trump").json() # LONG/SHORT playbooks playbook = requests.get(f"{BASE}/api/playbook").json() # System health / circuit breaker state status = requests.get(f"{BASE}/api/status").json() ``` ### AI Chatbot API ```python import requests response = requests.post( "https://trumpcode.washinmura.jp/api/chat", json={"message": "What signals fired today and what's the consensus?"} ) print(response.json()["reply"]) ``` ## MCP Server (Claude Code / Cursor Integration) Add to `~/.claude/settings.json`: ```json { "mcpServers": { "trump-code": { "command": "python3", "args": ["/path/to/trump-code/mcp_server.py"] } } } ``` Available MCP tools: `signals`, `models`, `predict`, `arbitrage`, `health`, `events`, `dual_platform`, `crowd`, `full_report` ## Open Data Files All data lives in `data/` and is updated daily: ```python import json, pathlib DATA = pathlib.Path("data") # 44,000+ Truth Social posts posts = json.loads((DATA / "trump_posts_all.json").read_text()) # Posts with signals pre-tagged posts_lite = json.loads((DATA / "trump_posts_lite.json").read_text()) # 566 verified predictions with outcomes predictions = json.loads((DATA / "predictions_log.json").read_text()) # 551 active rules (brute-force + evolved) rules = json.loads((DATA / "surviving_rules.json").read_text()) # 384 features × 414 trading days features = json.loads((DATA / "daily_features.json").read_text()) # S&P 500 OHLC history market = json.loads((DATA / "market_SP500.json").read_text()) # Circuit breaker / system health cb = json.loads((DATA / "circuit_breaker_state.json").read_text()) # Rule evolution log (crossover/mutation) evo = json.loads((DATA / "evolution_log.json").read_text()) ``` ## Download Data via API ```python import requests BASE = "https://trumpcode.washinmura.jp" # List available datasets catalog = requests.get(f"{BASE}/api/data").json() # Download a specific file raw = requests.get(f"{BASE}/api/data/surviving_rules.json").content rules = json.loads(raw) ``` ## Real Code Examples ### Parse Today's Signals ```python import requests signals_data = requests.get("https://trumpcode.washinmura.jp/api/signals").json() today = signals_data.get("today", {}) print("Signals fired today:", today.get("signals", [])) print("Consensus:", today.get("consensus")) # "LONG" / "SHORT" / "NEUTRAL" print("Confidence:", today.get("confidence")) # 0.0–1.0 print("Active models:", today.get("active_models", [])) ``` ### Find Top Performing Rules from Surviving Rules ```python import json rules = json.loads(open("data/surviving_rules.json").read()) # Sort by hit rate descending top_rules = sorted(rules, key=lambda r: r.get("hit_rate", 0), reverse=True) for rule in top_rules[:10]: print(f"Rule: {rule['id']} | Hit Rate: {rule['hit_rate']:.1%} | " f"Trades: {rule['n_trades']} | Avg Return: {rule['avg_return']:.3%}") ``` ### Check Prediction Market Opportunities ```python import requests arb = requests.get("https://trumpcode.washinmura.jp/api/insights").json() markets = requests.get("https://trumpcode.washinmura.jp/api/polymarket-trump").json() # Markets sorted by volume active = [m for m in markets.get("markets", []) if m.get("active")] by_volume = sorted(active, key=lambda m: m.get("volume", 0), reverse=True) for m in by_volume[:5]: print(f"{m['title']}: YES={m['yes_price']:.0%} | Vol=${m['volume']:,.0f}") ``` ### Correlate Post Features with Returns ```python import json import numpy as np features = json.loads(open("data/daily_features.json").read()) market = json.loads(open("data/market_SP500.json").read()) # Build date-indexed return map returns = {d["date"]: d["close_pct"] for d in market} # Example: correlate post_count with next-day return xs, ys = [], [] for day in features: date = day["date"] if date in returns: xs.append(day.get("post_count", 0)) ys.append(returns[date]) correlation = np.corrcoef(xs, ys)[0, 1] print(f"Post count vs same-day return: r={correlation:.3f}") ``` ### Run a Backtest on a Custom Signal ```python import json posts = json.loads(open("data/trump_posts_lite.json").read()) market = json.loads(open("data/market_SP500.json").read()) returns = {d["date"]: d["close_pct"] for d in market} # Find days with RELIEF signal before 9:30 AM ET relief_days = [ p["date"] for p in posts if "RELIEF" in p.get("signals", []) and p.get("hour", 24) < 9 ] hits = [returns[d] for d in relief_days if d in returns] if hits: print(f"RELIEF pre-market: n={len(hits)}, " f"avg={sum(hits)/len(hits):.3%}, " f"hit_rate={sum(1 for h in hits if h > 0)/len(hits):.1%}") ``` ## Key Signal Types | Signal | Description | Typical Impact | |--------|-------------|----------------| | `RELIEF` pre-market | "Relief" language before 9:30 AM | Avg +1.12% same-day | | `TARIFF` market hours | Tariff mention during trading | Avg -0.758% next day | | `DEAL` | Deal/agreement language | 52.2% hit rate | | `CHINA` (Truth Social only) | China mentions (never on X) | 1.5× weight boost | | `SILENCE` | Zero-post day | 80% bullish, avg +0.409% | | Burst → silence | Rapid posting then goes quiet | 65.3% LONG signal | ## Model Reference | Model | Strategy | Hit Rate | Avg Return | |-------|----------|----------|------------| | A3 | Pre-market RELIEF → surge | 72.7% | +1.206% | | D3 | Volume spike → panic bottom | 70.2% | +0.306% | | D2 | Signature switch → formal statement | 70.0% | +0.472% | | C1 | Burst → long silence → LONG | 65.3% | +0.145% | | C3 ⚠️ | Late-night tariff (anti-indicator) | 37.5% | −0.414% | > **Note:** C3 is an anti-indicator — if it fires, the circuit breaker auto-inverts it to LONG (62% accuracy after inversion). ## System Architecture Flow ``` Truth Social post detected (every 5 min) → Classify signals (RELIEF / TARIFF / DEAL / CHINA / etc.) → Dual-platform boost (TS-only China = 1.5× weight) → Snapshot Polymarket + S&P 500 → Run 551 surviving rules → generate prediction → Track at 1h / 3h / 6h → Verify outcome → update rule weights → Circuit breaker: if system degrades → pause/invert → Daily: evolve rules (crossover / mutation / distillation) → Sync data to GitHub ``` ## Troubleshooting **`realtime_loop.py` not detecting new posts** - Check your network access to Truth Social scraper endpoints - Verify `data/trump_posts_all.json` timestamp is recent - Run `python3 trump_code_cli.py health` to see circuit breaker state **`chatbot_server.py` fails to start** - Ensure `GEMINI_KEYS` env var is set: `export GEMINI_KEYS="key1,key2"` - Port 8888 may be in use: `lsof -i :8888` **`overnight_search.py` runs out of memory** - Runs ~31.5M combinations — needs ~4GB RAM - Run on a machine with 8GB+ or reduce search space in script config **Hit rate dropping below 55%** - Check `data/circuit_breaker_state.json` — system may have auto-paused - Review `data/learning_report.json` for demoted rules - Re-run `overnight_search.py` to refresh surviving rules **Stale data in `data/` directory** - Daily pipeline syncs to GitHub automatically if running - Manually trigger: `python3 trump_code_cli.py report` to force refresh - Or pull latest from remote: `git pull origin main`