Embed Notice

@bjb so one mechanism is to use multiple comparisons. Let's say that objects "really" compare by alphabetical order, and the comparator's accuracy is about 2/3. Here are some comparisons between A and E:

(A, E, A)
(B, E, B)
(C, E, E) <- incorrect
(D, E, D)
(A, B, A)
(A, C, A)
(A, D, D) <- incorrect

So, the evidence we have for A > E is:

(A, E, A)
(A, B, A) and (B, E, B)

These don't help:

(A, C, A) and (C, E, E)
(A, D, D) and (D, E, D)