The optimal n is sensitive to both α and the cost ratio τ_draft/τ_target.
The first experiment should measure α at n=5, which then tells you analytically whether it is worth going higher. If the measured acceptance rate is above 0.75, n=8 would gain meaningfully; if it is below 0.5, reducing to n=3 might actually be better.
The optimal n is sensitive to both α and the cost ratio τ_draft/τ_target.
The first experiment should measure α at n=5, which then tells you analytically whether it is worth going higher. If the measured acceptance rate is above 0.75, n=8 would gain meaningfully; if it is below 0.5, reducing to n=3 might actually be better.