Maple is designed around one consistent path, meant to apply whether the question is a trading strategy or a broader market relationship:
Question → Data → Model → Test → Simulate → Validate → Evidence
Evidence rarely ends an investigation — it usually reframes the next question. That loop, not a single pass through it, is the shape Maple is being designed around.
The engine behind that path
Underneath that conceptual path, the engine Maple is building is meant to work in four stages:
1. Input hypothesis
A hypothesis expressed in plain English, structured logic, or supported code — for example, an entry condition based on a moving-average crossover, or a broader question about how an asset behaves following a specific kind of economic event.
2. Build model state
Turning that intent into a precise internal representation — the step that separates a vague idea from something that can actually be tested.
3. Replay history
Testing the model against historical conditions using point-in-time assumptions and realistic constraints, rather than information that would not have been available at the time.
4. Evidence output
Producing a result designed to expose uncertainty, assumptions, and limits — instead of hiding them behind a single, confident-looking number.
Why a single backtest isn't the destination
A single historical result is not evidence of anything on its own — it's a starting point. Maple's validation layer (see Methodology) is designed to check for overfitting, small-sample unreliability, regime-dependence, and drawdown risk that a single test won't surface on its own. The goal is a considered, honest read on a result — not just a performance number.
Where AI fits
AI is designed to help express a hypothesis in the first place — turning a plain-English idea into structured logic. The quantitative calculations underneath are designed to stay transparent and deterministic: AI assists with expression, it doesn't decide what a result is.