It is clear that there exists an internal grammar of numbers
in Lojban, even though it has not been fully worked out or
made explicit. For example su'o re ci has to be understood
as su'o(reci) and not as (su'ore)ci. re ci ce'i has to
be understood as (reci)ce'i and not as re(cice'i), so the
rule is nothing as simple as left or right grouping. In fact
the digits are usually joined the tightest, which is why I
don't like making ji'i an exceptionally tight digit modifier.
It is hard enough that we have to make out the value of each
digit just from its position. Adding ji'i in the middle makes
this harder by creating interference and breaking the
three-digit groupings that we have to use to get the value
of each one right.
This is clearly incomplete, but it can serve as a starting point:
<pre-sign> = (ji'i|su'e|su'o|me'i|za'u|da'a)... <sign> = ma'u|ni'u <digit> = no|pa|re|ci|vo|mu|xa|ze|bi|so|dau|fei|gai|jau|rei|vai|ki'o|xo <real> = [<pre-sign>][<sign>](ci'i|pai|te'o|[<digit>...][pi [[<digit>...][[ra'e <digit>...]]) <complex> = [<real>][ka'o <real>] <n-tuple> = [<complex>][pi'e <complex>]... <fraction> = [<real>][fi'u <real>] <indef> = ro|so'a|so'e|so'i|so'o|so'u|no'o|rau|du'e|mo'a <quantif> = tu'o | [pi][<pre-sign>][<indef>] | <real>[ce'i] | <fraction>