With 'normal' bounce flash and perhaps the bounce card out or the diffuser dome on and a fair amount of 'forward ' light TTL/BL does a good job . But as soon as the flash head is pointed backwards and the distances start changing TTL/BL tends to expose for " the brightest object in the focus point diamond " which means that it under-exposes when there is white in the frame ....
Lots of forward light = good exposure .
Only bounce light = under-exposure with white in the frame .
I still need to do some tests to confirm my suspicion but I'm thinking that perhaps TTL/BL still 'remembers' the distance info from the lens when you tilt the flash head and doesn't totally forget about it .
Perhaps it determines a minimum output for ''if the flash head was still facing forward'' and doesn't drop below that amount while still using part of its program to prevent blown highlights - which may be what backs off the exposure for whites to an average gray .
At a set distance with the flash head facing forward it can 'conclude' that an object 'must be white' if it is reflecting so much light back on the pre-flash since it knows its distance , but when the flash head is tilted and it doesn't know how far the light will travel then it has to start using its ''preserve highlights'' part of the program along with the ''minimum output'' from distance info data [ perhaps ] which allows it to be quite accurate with a decent amount of forward flash .
Perhaps this is why the manual suggests leaving the diffuser dome on at a 60 degree angle ''for best results ' ?
I'm going to have to work out some sort of a test to figure this one out because this is the final link to '' predictability '' of flash with TTL/BL mode .