Why accuracy and strong backtests can mislead in financial ML—and why reproducibility, leakage-safe validation, and economic evidence matter.