How often is a "significant" signal actually real?

Your desk tests candidate signals at significance level $\alpha = 5\%$, and your test has power $90\%$ (it flags a truly real signal $90\%$ of the time). From experience, about $20\%$ of the signals you bother testing turn out to be real.

Among signals your test flags as significant, what fraction are actually real? Redo it if only $2\%$ of tested signals are real.