The systems
We run multiple independent return engines. Some are proven edges; some are research we're honest about. Every one is labeled for exactly what it is.
Time-series momentum across 27 global futures, volatility-targeted, daily — the most documented public edge in markets, run the way the multi-billion-dollar CTAs do it.
The real insight: average single-market Sharpe is just 0.14 — almost noise. The 27-market portfolio is 0.69. Diversification is a 5× multiplier. Honest: trend has been in an industry-wide drought since ~2023; it shines in dislocations, bleeds in chop, and holds for weeks.
Scores each market from grounded cross-market relationships — gold vs the dollar, equities vs the VIX, copper with growth — with weights evolved by a genetic engine and tested only on years it never saw.
The optimizer put equal weight on cross-market signal (0.41) and a market's own trend (0.41). Honest: the single split looked like 0.78; rolling walk-forward is ~0.5 and the deflated Sharpe is low, so we trade the un-over-optimized exposure, not the most-evolved genome.
Llama (Cloudflare Workers AI) reads the crypto news 24/7 → the model decides → paper book on BTC/ETH, public scoreboard vs Codex. The edge is breadth + 24/7 speed vs other retail; it's a live forward test, not a sold backtest (the published edge decays as everyone adopts LLMs).
Open the live desk →An 8-step method from a genuinely profitable discretionary trader. We did the rare thing and tested whether it mechanizes — honestly. The rules sit at a coin flip (PF ~0.75); his real account runs PF ~1.78. That gap is discretion — which sweep is clean, which day is worth trading. We proved his edge is real and lives in judgment, so the build is human bias + bot execution, not bottling his brain.