Vision
Keep compounding without blowing up.
Grow a $2–10k account at 15%+ per year, never losing more than 25% from peak, fully autonomous, across all market regimes.
What This Means
- Compound — profits reinvested, money grows on money
- 15%+ CAGR — at least 15% annual growth averaged over years, not months
- <25% max drawdown — the system survives bear markets, flash crashes, and choppy sideways
- Fully autonomous — runs 24/7 with minimal intervention, no babysitting
- All market regimes — works in bull, bear, and sideways — not fitted to one environment
Success Metrics
| Metric | Target | Why |
|---|---|---|
| CAGR | > 15% | Beats passive holding with less stress |
| Max drawdown | < 25% | Survive to compound another day |
| Sharpe ratio | > 1.0 | Edge is real, not just luck |
| Win rate | > 45% | Combined with R:R > 2:1, expectancy is positive |
| Uptime | > 99% | Autonomous means actually running |
What We Are Building
A multi-strategy trend-following portfolio on crypto futures that:
- Finds edge (gordon-lab) — research, validate, and prove strategies via walk-forward ablation before any real money is risked
- Executes edge (gordon-bot + gordon-executor) — run validated strategies with strict risk management, position sizing, and circuit breakers. One code path for both backtest and live
- Controls risk (gordon-risk) — five portfolio-level circuit breakers enforce a halt-latch. Drawdown, Connectivity, VPIN, Macro, Correlation. Explicit resume required to clear
- Monitors everything (gordon-console + gordon-manager) — operator dashboard, WS fanout, backtest replay, bot lifecycle — all through a single BFF surface
What We Are Not Building
- A prediction engine — Gordon reacts to trends that have already started, it does not predict where prices will go
- A high-frequency trading system — timeframes are 1 hour to 1 week, not milliseconds
- A black-box ML system — strategies are interpretable trend-following; ML is an optional overlay, never the base signal
- An over-engineered platform — every service exists because the boundary is load-bearing; speculative architecture is rejected
Decision Filter
When in doubt, ask: "Does this help the system compound more reliably?"
- Adding a feature? Only if it improves CAGR or reduces drawdown, proven by ablation.
- Choosing complexity vs simplicity? Simplicity — complex systems break under pressure.
- Skipping testnet validation? Never. Validate before risking capital.
- Cutting quality to ship faster? No. With AI tooling, doing it right takes the same wall-clock time.