How to use the engine like a filter instead of a marketing tool
The fastest path to better deployment is learning to reject more strategies earlier.
- Do not tune until the bad period disappears.
- Do not compare systems with incompatible capital assumptions.
- Do not treat one high-performing run as robust evidence.
Lock the strategy before the run
Freeze the logic and capital assumptions before you inspect performance so the test reflects the original thesis instead of a moving target.
Run broad enough scenarios
Test multiple periods, symbols, and market states to see whether the system's behavior survives outside the one obvious window.
Review stress, not just summary stats
Drawdown duration, loss clustering, and recovery quality usually matter more operationally than the headline return number.
Decide pass, revise, or kill
A clean workflow ends with an explicit decision. Strategies that stay forever in maybe-mode consume time and add noise to the desk.