How long should I paper trade a strategy before going live?

Long enough for the strategy to make a meaningful number of decisions — which depends on its tempo, not the calendar. A daily mean-reversion system might complete 15–20 round trips in two months; a golden-cross strategy might not trade at all in a year, so paper trading it teaches you little beyond plumbing. Practical rule: wait for at least 10–20 completed trades or 4–8 weeks, whichever is longer, and check that live-data behavior matches the backtest's character (similar trade frequency, similar win pattern). Then start live at a fraction of intended size. The purpose isn't proving profitability — a month proves nothing statistically — it's proving the machine does what the backtest said it would.