Your Strategy Deserves More Than a Sandbox
Sandbox script platforms cap your strategy at 500 lines of interpreted code inside a locked-down environment. The allowed imports are basic math libraries. Network, threading, database access: all prohibited. The result is a structural ceiling that no amount of UI polish can fix.
Capability Comparison
The difference is not features. It is architecture.
| Dimension | Sandbox Platforms | StratCraft |
|---|---|---|
| Strategy language | ~500-line pandas / Pine Script | C++23, Python, AI-generated code |
| Execution performance | Interpreted, single-threaded | Compiled C++23, SIMD/AVX-512 vectorized |
| Strategy composition | Single-indicator logic (one signal per strategy) | Combinator pipeline: Factors ML Signals Hypotheses Indicators → Linear / ML combination |
| Data access | Sandbox-restricted (no network, no file IO) | Multi-source, multi-frequency, alternative data |
| AI / ML integration | Chat assistant (no code generation) | Supported BYOK or local-model authoring plus separately admitted ML research |
| Plugin architecture | No plugins (closed sandbox) | Process-isolated plugins via gRPC, hot reload |
| Scalability | 1 strategy at a time | Governed multi-strategy research operations |
| Deployment | Docker + manual env config | Local research workflow; deployment is not a current product claim |
| Factor research | Not possible in sandbox | Alpha factors, expression trees, IC/ICIR — composable inputs to Combinator |
| Market edge potential | Public signals only (no edge) | Inspectable combination with evidence; no market-edge claim |
The Ceiling Problem
Why architecture affects which research operations are possible
What You Can Build in a Sandbox
- RSI overbought/oversold
Every one of these is a public signal. Thousands of free TradingView indicators do the same thing. There is no market edge here.
What Broader Quant Research May Require
- Latency-sensitive market making (microsecond C++ execution)
None of these fit inside a 500-line sandbox. They require compiled performance, unrestricted data access, and extensible architecture.
When a Sandbox Is Fine
Sandboxes have their place for specific use cases.
- You are learning to code trading strategies
- You want simple indicator alerts
- You need a quick prototype for a basic idea
- You are comfortable with public signals
When You Need a Framework
Serious research requires serious infrastructure.
- You need multi-factor models with statistical validation
- You want ML pipelines integrated into your backtest workflow
- You need compiled C++ execution with workload-specific evidence
- You want AI assistance while retaining human review and approval
Build Without a Ceiling
The ceiling should be your strategy idea, not your tool.