@cfr That's great news. So I can submit once I get some sleep! ;-)

@SašoŽivanović Thanks for all your suggestions. I don't like ignore edge, so I did something a bit different:

but it seems to work OK. Yours maybe better. Indeed, it probably is.

\documentclass[border=10pt,multi,tikz]{standalone}
\usepackage{forest}
\begin{document}
\forestset{
  declare toks={iteration}{},
  declare boolean={is auto iteration}{0},
}
\begin{forest}
  label my edge/.style n args=3{% based on Sašo's suggestion: chat.stackexchange.com/transcript/message/34121933#34121933
    edge label={node [font=\scriptsize, #3] {$\frac{#1}{#2}$}}
  },
  /tikz/put left/.style={anchor=north east, pos=.6, left, xshift=-3.5pt},
  /tikz/put below/.style={midway, below},

@cfr It surely does work. I actually think this is a better idea than ignore edge.

Two little ideas to improve readability and speed. (i) Instead of computing x and y manually, you can simply set s and l (but before computing xy, of course). (ii) Instead of testing for condition >O_={!s.n children}{0} (cool processing, by the way!) at each w_x node, you could simply walk a nodewalk to w_p. Code follows.

@SašoŽivanović Thanks. I didn't realise you could use .option like that ;). At least, I don't think I did. I did try using l etc. rather than y etc., but it didn't work at the time, which was why I used x/y. I don't know what I was doing wrong.

@SašoŽivanović So the nodewalk for every w_? is more efficient that the test?

@SašoŽivanović Can you explain the while nodewalk valid bit? The first argument is the nodewalk (to the first child). The second is the keylist (2). So does that become a step in the outer nodewalk? So it ends up at the last second child? Then to the parent and then to the first child? But that can't be right ...

Why doesn't it need to be delayed? I guess for the same reason for tree doesn't?

@cfr Since v2.1 ;-)

@SašoŽivanović ;)

@SašoŽivanović Why not e.g.

  !rLs.before computing xy={% Sašo: chat.stackexchange.com/transcript/message/34150084#34150084
    l/.option=!r1.l,
    s/.option=!r1.s,
  },

@cfr It is ... although, in this case, probably not much, I guess. I didn't profile it, I must admit. I thought to show you the technique, as I know you're (sometimes too) fond of trees made of ifs. ;-) Imagine you have a large tree and want to target a single node: walking there will usually be cheaper than testing a condition at each node.

@cfr That's right.

@cfr Yes. The structure is already there when the preamble is executed.

@SašoŽivanović Probably true :(.

@cfr I believe w_1 is the last leaf.

@cfr ;-)

@SašoŽivanović Add blue, typeset node.