1. the -A option to (o)nsgmls and (o)sgmlnorm, mentionned in the
archform.htm document, are not mentionned in the detailed syntax for
those commands.

2. both "(nsgmls|sgmlnorm) -A foo doc.sgml" exit with status 0 when foo
is not properly declared as an architecture for the document being
validated - nsgmls even outputs a single "C" line, as if the doc was
empty, but still conforming with something.

3. the archform.htm document shows "<?IS10744 ArcBase archname>" as a
synopsis but tells the name of the required attribute is "ArcDTD"

4. it looks like "sgmls -A archname" validates a document against the
named architecture _additionally_ to validating it against the
declared DTD, when there is a DOCTYPE.  This seems to imply that XML
documents without a doctype, but with an architecture declaration,
cannot be validated against an architecture

See below for the example I'm trying to get to work.  You'll note the
2 PIs I've tried to use (I've used others, but these ones were the
most likely) - the one that is active is derived from what I
understood of archform.htm, the one commented out is the syntax
dexcribed in http://www.isogen.com/papers/archintro.html

After browsing quickly through the draft standard, I understand none of
those is correct, but did not find the correct formulation yet (no equals
sign is sure to be better, but something else is wrong - but the software
keeps keeping me ignorant of the problem)

Yann Dirson <ydirson@altern.org> writes:

I wish I could help, but I've never worked with architectural forms.
Have you tried on the SGML usenet list or even the Mulberry DSSSL
list?

I will mark this bug (and the other) forwarded, opening a sourceforge
bug for them.  I can apply patches upstream and such too if you get
that ambitious.

Again, sorry I can't answer your question.

No, but a google search finally gave me a sample, sent by jjc on a
list after sp began to support that.  It's
http://www.oasis-open.org/cover/archEngine.html IIRC (offline
mail-composing sucks).

I attach is a simpler example, which you may want to include in the
doc.

Will remain to fix this behaviour of silently accepting
syntactically-wrong PI, as well as arguments to -A that are not
declared as base architecture for the document.