LLM Trading Lab — Autonomous AI Trading Experiment

6 frontier AI models run fully autonomous trading portfolios with zero human intervention. Real-time performance vs S&P 500.

▍ EQUITY CURVES // 6 MODELS vs SPY

▍ LEADERBOARD —

#	MODEL	STREAK	VERSION	RETURN	DAILY	TREND_30D	SHARPE_30	MAX_DD	WIN_RATE	ALPHA	POS	CASH	VALUE

▍ API COST TRACKER —

MODEL	TODAY	7-DAY	MONTH	TOTAL	$/TRADE	GROSS P&L	NET (P&L − COST)	ROI vs COST

▍ TRADE FEED LAST 50

▍ PORTFOLIOS PER-MODEL HOLDINGS

▍ CONSENSUS PICKS —

TICKER	MODELS	AVG WT	AVG CONF	AVG P/L

▮ HEAD TO HEAD

▮ MODEL CORRELATION PAIRWISE DAILY RETURN CORRELATION

▍ CONFIDENCE CALIBRATION DO THE MODELS KNOW WHAT THEY KNOW?

▍ SYSTEM

▍ MODEL VERSIONS

▍ REPORTS MONTHLY DEEP-DIVE ARCHIVE

· PDF · PHASE A · PILOT / EXPLORATORY

Erratum · 2026-08-13 · July 2026 (rev. 2): the notable-decisions extractor omitted short-side executions (SHORT/COVER) from July 1 until corrected 2026-08-10 — a BUY/SELL-only filter across 74 short-side orders. Gemini's PG cover ($16,505.28) and DeepSeek's QCOM cover ($9,412.04) now enter at #2, displacing SLB and GLD. Rev. 1's “long-leg only by extractor design” framing stated a defect as a design property. Descriptive display field only — no return, metric, gate, or RQ input changed.

· PDF · PHASE A · PILOT / EXPLORATORY

Erratum · 2026-07-03 · May 2026 (rev. 2): SPY since-inception return corrected 11.26% → 11.18% and the six derived model alphas re-stated. Cause: ambiguous inception anchor, now ledger-pinned at 680.40 (deployment-moment pin). Magnitude 8 bp — no rankings or conclusions changed.