Full PLL — all 21 cases to permute the last layer in a single algorithm. One look, one algorithm, solved.
With one algorithm per case, you remove the variance of a second look and get tighter, more predictable times across your averages.