position() test for processing boundaries

This is the first time I write to this list, since I have carefully
tried to check the archives for an answer to my question. I am
struggling with finding ways to delimit processing of descendant nodes
based on start and end positions.

A previous posting
(http://www.biglist.com/lists/xsl-list/archives/200009/msg00181.html)
comes close, but leaves me confused about one aspect. Basically, it has
to do with a (seemingly?) difference in the way [position() > ...]
and [position() < ...] predicates are evaluated.

<p>
 <q n="1>
   <app n="2">
     <rdg n="3">rdg1</rdg>
     <rdg n="4">rdg2</rdg>
     <rdg n="5">rdg3</rdg>
   </app>
   [text]
   <anchor n="6"/>
   <app n="7">
     <rdg n="8">rdg1</rdg>
     <rdg n="9">rdg2</rdg>
     <rdg n="10">rdg3</rdg>
   </app>
 </q>
 [text]
 <app n="11">
   <rdg n="12">rdg1</rdg>
   <rdg n="13">rdg2</rdg>
 </app>
 <anchor n="14"/>
 <q n="15">
   <app n="16">
     <rdg n="17">rdg1</rdg>
     <rdg n="18">rdg2</rdg>
     <rdg n="19">rdg3</rdg>
   </app>
   [text]
   <anchor n="20"/>
   text
   <app n="21" TEIform="app">
     <rdg n="22">rdg1</rdg>
     <rdg n="23">rdg2</rdg>
   </app>
   [text]
   </q>
</p>

I am trying to find ways for delimiting the processing of descendants of
<p>, so that all and only those elements between a certain start
position and a certain end position are processed. Experimenting with
ways to express such constraints, I managed to identify a basic problem
(in my understanding, probably). I found following differences:

Given following example XSLT fragment triggering processing from START
position 3:

<xsl:template match="p">
 <xsl:copy-of select="(.//*)[position() &gt; 3]"/>
</xsl:template>

This one correctly (or, at least, expectedly) copies all descendant
nodes of <p>, starting from <rdg n="4"> up to the end.

However, when I use that same position as END position in following
fragment:

<xsl:template match="p">
 <xsl:copy-of select="(.//*)[position() &lt; 3]"/>
</xsl:template>

All direct children of <p> are returned that occur before the 3rd direct
child:

<q n="1">
 <app n="2">
   <rdg n="3">rdg1</rdg>
   <rdg n="4">rdg2</rdg>
   <rdg n="5">rdg3</rdg>
 </app>
 [text]
 <anchor n="6"/>
 <app n="7">
   <rdg n="8">rdg1</rdg>
   <rdg n="9">rdg2</rdg>
   <rdg n="10">rdg3</rdg>
 </app>
</q>
<app n="2">
 <rdg n="3">rdg1</rdg>
 <rdg n="4">rdg2</rdg>
 <rdg n="5">rdg3</rdg>
</app>

This seems to suggest that the position test in case (A) is applied at
the level of the deepest descendant, while in case (B) at the level of
the first generation of descendants.

Could anyone explain what makes the difference? Any suggestions for the
right way to delimit processing of descendant nodes up to a certain
position (in this example only the nodes appearing before <rdg n="3">)?

--
Ron Van den Branden
Wetenschappelijk attachi
Centrum voor Teksteditie en Bronnenstudie (CTB)
Koninklijke Academie voor Nederlandse Taal- en Letterkunde (KANTL)
Koningstraat 18 / b-9000 Gent / Belgium
e-mail : ron.vandenbranden@xxxxxxxx
http://www.kantl.be/ctb/staff/ron.htm