This time, you omitted the overrides to prevent padding and
staff-padding from influencing the test:
%%%%
\version "2.19.45"
notes = \fixed c''' { \ottava 1 f4 4 4 \ottava 0 f, }
{
% Make sure these properties do not influence
% anything while testing outside-staff-padding.
\override Staff.OttavaBracket.padding = 0
\override Staff.OttavaBracket.staff-padding = 0
\override Staff.OttavaBracket.outside-staff-padding = 0.56
\notes
\revert Staff.OttavaBracket.outside-staff-padding
\override Staff.OttavaBracket.extra-offset = #'(0 . 0.1)
\notes
}
%%%%
This gives a clearer overview, thanks, but I would not use it, even if it produces the wanted result.
I did not omit them, given that I did not want to reset any property, as I explained in other posts. In practice, you are only making reset of all the properties that add more space between the grob and the reference point. At the end, considering that you have to add a .46 value, you get even a worse procedure than \override Y-offset, while using the same logic (which is: distance from a reference point, and not *offset the result*). So: I think it's better to use a ruler from the middle of the staff.
No other way, if you want to preserve the avoid-collision algo.
Best,
Paolo