1four samples of one prompt, and the reference

sample A
A
sample B
B
sample C
C
sample D
D
the reference image
reference

2every pair goes to the judge with the reference, and it picks one

A B ref judge A A C ref judge A A D ref judge A B C ref judge B B D ref judge B C D ref judge C

3wins → rank → bonus

A 1st +1.00
B 2nd +0.67
C 3rd +0.33
D 4th +0.00

bonus = (pool size − rank) ÷ (pool size − 1). eight samples means 28 pairs.