Skip to main content
🧬 Research · GPU Discovery

Symbiose

What if you paired a defensive strategy (high win-rate, steady gains) with offensive ones (low win-rate, big PnL)? We brute-forced 574 strategy×asset lanes on an RTX 4080 (523 million weighted portfolios) to find combinations whose combined Sharpe beats each individual leg.

⚠️ Honesty first. The numbers shown are the out-of-sample (walk-forward) Sharpe. In-sample the best portfolio reached ~3.2 — but that was overfit. After train/test validation (182d / 182d) it settles at ~1.7. That honest number is what we publish. This is not financial advice nor a promise of returns.

Live Symbiose Arena

Each symbiose trades its own virtual $10k on Binance candles. The goal is simple: beat the best individual leg, without reading the main arena.

paper trading

Loading standalone arena...


    


    

        
Validated symbioses

        
Loading live data…

        

    


    

        
Real-trade discoveries (accumulating)

        
Loading…

        

    


    

        
How we find them

        

            
01
GPU backtest
57 strategies × 11 assets × 365 days of 10-min candles, replayed on an RTX 4080.

            
02
Weighted brute-force
523 million weighted portfolios tested, distinct-asset constraint (true diversification).

            
03
Walk-forward
Weights learned on the first half, Sharpe measured on the second. We keep only what holds out-of-sample.

            
04
9 / 20,515
Of 20,515 combinations tested, 9 confirm (OOS Sharpe > 1, > best leg, correlation < 0.1).

        

        
Data &amp; method openly available via /api/symbiose/status · Full methodology