Super-Trader — System Architecture

Current deployed state · momentum rotation (rule-based) · DRY-RUN · generated 2026-06-22 EST
Market Data (yfinance) daily OHLCV · VIX · 30-name candidate pool Meta-Orchestrator orchestrator/meta.py · daily cycle respects per-agent enabled flag Universe Selector signals/universe.py point-in-time liquidity rank Strategy Agent momentum rotation + SPY-200d filter no leverage · emits target weights Risk Agent drawdown / VIX / geo gates scales book · gross cap Executor Agent DRY-RUN — no live orders multi-leg rebalance · Alpaca stub Monitor P&L / health Reporter email / alert Research geo risk score → risk gates Shared State (JSON) state/ · signals · heartbeats · snapshots Dashboard Generator site/generate.py · honest/OOS only Cloudflare Pages super-trader-brn.pages.dev + GitHub (mcvalor-ai/super-trader) QUARANTINED / NOT IN LIVE PATH drop_sniper Agent disabled — 9–26 trade noise 4 XGBoost Classifiers (.models/) crash / bounce / fwd-regime / regime orphaned · weak/overfit · not wired in self_improving_engine lookahead-fixed · legacy hybrid not driving live strategy core agent signal/IO risk

Live Trading Path

  • • Data → Orchestrator → Strategy → Risk → Executor
  • • Strategy: momentum rotation, dynamic liquidity universe
  • • SPY>200d-SMA regime filter, no leverage
  • • OOS 2000–26: 13.4% CAGR / 0.93 Sharpe / −28% DD
  • • Executor in DRY-RUN — no live orders placed

State & Publishing

  • • Agents write JSON to shared state/
  • • Dashboard reads honest/OOS metrics only
  • • Biased Sharpe-2.9 source purged from site
  • • Deploy → Cloudflare Pages + push to GitHub
  • • 24 unit tests incl. lookahead guards

Quarantined / Orphaned

  • • drop_sniper: disabled (statistical noise)
  • • 4 trained classifiers: not wired in (weak/overfit)
  • • crash filter 119% CAGR = in-sample; OOS 18.8% / −80% '22
  • • self_improving_engine: lookahead fixed, not live
  • • See CLASSIFIERS.md & CRITIQUE.md for audits