As I understand, the problem is that one likes to have the roman text follow the devnagari saMhitaa version closely to help recitation/recognition but it is better to have split words to help analysis. How about using a centered dot · in place of a space.
For example write tv·asti in place of tv asti or tvasti. if this is done systematically, then a computer can easily perform the necessary analysis as well.

In his version of Ashhtaadhyaayii, S.M. Katre had indeed used the symbols - and = to break words, where one indicated a samaasa and the other a sandhi. While logically correct, the symbols are distracting to read and recite. It would be nice, however, to come up with another non distracting symbol.

BTW, the center dot is has ascii code 0183, in standard roman font. They are the same in the csx font.
if you want to type it in. It is possible to use a bold version "•" too,
namely the ascii code 0149, but I don't know if it is sufficiently distinct.

harry spier wrote:

>snip..snip

This is the very problem that I am trying to avoid.  I am preparing a
new edition of the Bhagavad Geeta for publication which will contain
both the devanagari and the roman transliteration. One of the uses of
the edition will be to chant the Geeta.  Some of those who use the text
will have Hindi as a first language and use the devanagari,  and some of
those who use it will have English as a first language.  In general the
majority will not know Sanskrit.

1) Isn't this one of the main uses of written Sanskrit (both now and
historically)? i.e. as an aid to proper recitation for those who do not
know the text by memory.  And shouldn't any standard address this need.
The pronounciation of the written phrase "tvidam" is much more obvious
than the phrase written as "tv idam" etc..  A space intuitively
indicates hiatus.
Snip>..Snip...