Most systematic traders are training on a coincidence.
The historical price chart is one path the market took. It is not the path. The future could land on any of a thousand others and if your strategy only survives the one timeline you trained on, you do not have an edge.
You have memory.
Backtesting on the historical OHLC is an already known limit. The fix the industry settled on: Monte Carlo, walk-forward, noise testing - all happen AFTER the strategy is built.
By the time you run them, your search algorithm has already memorized the noise. You're mostly checking how badly.
"Do not optimize for the known past. Optimize for the unknown futures."
Robustness is the only thing that compounds in this game. Everything else is one regime away from going to zero.
The fix: train your models and build your strategies on synthetic data not just the historical data. Link below 👇