Re: General Exception for Essential Purpose

Another reason for ensuring that the data model, as opposed to the final
interface, is available for processing by software under the user's
control is that given Len's groups of users (with different needs,
different devices, preferences etc., namely {U1, U2, U3 ... Un} we know
empirically that n, even if we could agree upon a means of quantifying it,
is very large.

Now consider a number of parallel, but different, user interfaces designed
by the creator of the content, each of which presents the same content but
in a distinct way: (I shall preserve Len's symbolism, though for the sake
of generality I refer to them as user interfaces rather than web sites):
{S1, S2, S3 ... Sm}

For each interface I (an element of S) there corresponds a subset C of U
such that I satisfies the profile of the needs and preferences of those u 
that comprise C.
. Now the quasi-formalistic treatment starts to break down and we
introduce a
number of further empirical observations. Most importantly, if each
element I of S is very concrete and specific (for example it consists of
rendered content) then the relation between S and U defined above becomes
one-to-one, or nearly so (mathematicians reading this will obviously decry
the lack of precision here)--the idea is that each subset C of U, the
needs/preferences of which is satisfied by each interface I, has one
element, or at most a very small number of elements.

 By contrast, if the author provides high-level abstractions which can be
transformed, by software operating under the user's control, into a
multiplicity of distinct interfaces, then a potentially large set of
interfaces can be generated automatically, thereby satisfying the needs
and preferences of many, but perhaps not all, members of U. Of course,
this "interface generation" capability could reside in software operating
under the author's control, but if this is the case, then (to achieve the
same result as when the interface is generated by the user's software) the
author's server-side software will need to be able to generate as many
elements of the set of interfaces satisfying U, as do the various user
agents available to members of U. This is impracticable because n is
large, and because expertise in some of the required interfaces (for
example, tactual or speech-based interfaces) is likely to be concentrated
in user agent rather than server developers.

What the 2.0 guidelines do is to require the higher-level semantics, but
to permit content developers to satisfy some subset of U by
(author-specified) interfaces comprising the set S. Each u (a member of U)
has the choice of either selecting an element of S, or of creating her/his
own interface I given the higher-level semantics and structure available
from the content developer's server, via style sheets and other
mechanisms. Conceptually, we may consider the higher-level semantic option
to be a member of S, which, unlike any other author-supplied interface I
(elements of S) satisfies a large subset of U, including most, if not all,
of those u (elements of U) that aren't satisfied by the remaining elements
of S. That is why we mandate the semantically rich option, but allow
authors to supply other interfaces I which are elements of S.

I apologize in advance for any mistakes in the quasi-mathematical
symbolism which is here being used only as a metaphor.

The usual disclaimer applies.

Received on Saturday, 28 October 2000 23:01:31 UTC