• Recurring discipline eventually fails, even when a team is right almost all the time.
• Architecture should make invalid behavior impossible, not merely discouraged.
• Keep business logic in a shared core and make platform shells deliberately thin.
• Prefer event-driven data flow over polling and manual refreshes.
• Remove compatibility shims when possible so known mistakes do not become permanent structure.
• Pay the structural cost once instead of paying for vigilance forever.