The following collation is proposed to replace the current collation with type="standard". The collation with type="direct" can be deleted. It is only a binary sort (codepoint order).
<collation type="standard" references=
"McGregor: Oxford Hindi-English Dictionary ISBN:019864339X
Gatzlaff-Hälsig: Handwörterbuch Hindi-Deutsch ISBN:3875481771">
<settings normalization="on" />
The following tailoring is an adjustment of the
DUCET collation order for ANUSVARA, CANDRABINDU,
and VISARGA. Instead of being sorted with secondary
weights, they are collated on primary level before
the independent vowels. CANDRABINDU is secondary
different from ANUSVARA. This gives a sort order
very close to the referenced dictionaries. The sort
order in the dictionaries, which by the way are not
exactly the same, require quite extensive tailoring.