Strategy Forge
15-agent pipeline: propose → code → backtest → critique (12 specialists in parallel) → decide. Every run either GRADUATEs (rare), REFINEs (most common), PIVOTs (rare), or ABANDONs.
Loading…
Click a run row below to drill in. Look for runs marked GRADUATE — those are the strategies the council says are ready to deploy. ABANDON verdicts are also instructive — they tell you what doesn't work.
A strategy proposer drafts a hypothesis. The code-writer agent translates it into a tested Python module. The backtester runs it on real historical data. Then twelve specialist critics — each looking at a different failure mode — independently vote on whether to graduate, refine, pivot, or abandon. The arbitrator synthesizes their verdicts into a single decision and writes the next round's brief.
Historically < 5% of runs graduate on round 1. Most are REFINEs that go 3–5 rounds before reaching a final verdict. ABANDONs are decisive: the council saw a fundamental flaw that no parameter tweak will fix.