Interpreting ADF and KPSS Together

You run both tests on a monthly revenue series and get:

• ADF: p-value = 0.42 (fail to reject H_0: unit root)
• KPSS: p-value = 0.01 (reject H_0: stationary)

1. What does this combination indicate, and what should you do to the series before fitting an ARMA model?
2. After first-differencing, ADF gives p = 0.001 and KPSS gives p = 0.15. What do you conclude now, and what does the "d" in ARIMA(p,d,q) equal?
3. A colleague argues "just always difference twice to be safe — it can't hurt." Explain what is wrong with that.