I believe the study showed that a first-year draw of 4% from a 50/50 portfolio of large-cap US stocks and US treasuries, adjusted annually for inflation, would never have gone to zero over any rolling 30-year period from the 1920s to the 1990s.

However, as you allude to below, batting 1000 on past market conditions doesn't mean you can't fail in the future. Monte Carlo simulations show this approach has approximately a 13% chance of failure if you simulate thousands of possible 30-year periods.

Worse yet, if you increase the simulated retirement period length (as you'd need to for early retirement) the risk of failure grows even larger.

The most recent Morningstar report shows that you'd have a 90% chance of success over 30 years with a 3.7% starting withdrawal rate.