Account for look‑ahead bias, survivorship bias, and outlier handling. Use bootstrapped resampling to test fragility across regimes. Reserve clean validation windows, avoid hyperparameter binges, and publish settings so future you remembers why a past you trusted these numbers.
Run the system live without money, noting latency, rejects, and calendar quirks. Keep a journal of unexpected alerts and manual temptations. If you felt compelled to override, evolve the rule rather than indulging adrenaline. Celebrate boring days; that’s stability arriving.