Usage of degrees of conformance

Following an internal discussion to the Team, I got the following
opinion on degrees of conformance from Joseph Reagle (a W3C Team
member):

[ Note that "levels" below are to be understood as what we call now
"degrees of conformance"]

-----Message suivi-----
> From: Joseph Reagle <reagle@w3.org>
> I 
> don't think [degrees of conformance] should be encouraged regardless
> though it seems to be 
> popular at the W3C now.)
> [...]
>
> It's based on experience. But first I want to stress I'm not necessarily 
> concerned with what WAI and QA have done, just the "knee jerk" reaction of  
> calling for levels. 

> Many moons ago when I inherited the P3P project we were attempting to 
> address a number of requirements and in my naivety I pursued, for a short 
> time, "levels of conformance." The reason we went down this path was 
> because of an inability to focus on what was going to be implemented, make 
> the hard decisions, and cut down on the complexity. However, seeing the  
> complexity continue to bloom and getting confused about questions such as 
> what does it mean for an OPTIONAL feature, if implemented, to have a 
> MANDATORY sub-feature, with its own OPTIONAL sub-sub-feature, we pulled 
> back. Since then, I consider every optional feature in a spec I author a 
> personal "demerit" <smile/>, levels as an inability to be specify that 
> which will soon be implemented, and I consequently prefer in my exit 
> criteria to include at least one implementation that implements *every* 
> feature [1]. [...] Regardless, I also remember talking to Ian while 
> he was trying to come to terms with the concepts of the levels in WAI and I 
> appreciated that when a specification is as prescriptive as descriptive (if 
> not more so) levels *may* be a useful tool and I think the framework arrived 
> at is very good *if* the situation merits it. 
>
> 1. [1] ° 02.07.11.th | the progress of step-wise stumbles
>   ...
>   Given different implementations, their variance in the 20% each fails
>   to do well causes 80% of the users' headaches (e.g., CSS layering
>   works in Foo, but not in Bar).

The latest editor draft of the spec GL [2] doesn't make any
recommendation in this regard. Maybe it should?

Specifically, the checkpoint doesn't differentiate between a profiled
specification vs a specification using degrees of conformance (ala WAI
or QA). Rather, it does differentiate them, but not enough (in my
opinion): a profile is a specification per se and the conformance clause
belongs to each profile, whereas in the WAI/QA case, one specification
holds the 3 degrees of conformance.
I would suggest that degrees of conformance be strongly discouraged for
anything but guidelines, or maybe more broadly, anything but "foundation
or abstract" specifications [3].

Dom

2. http://www.w3.org/QA/WG/2002/08/qaframe-spec-0804.html#Ck-define-all-levels
3. http://www.w3.org/QA/WG/2002/08/qaframe-spec-0804.html#document-categories
-- 
Dominique Hazaël-Massieux - http://www.w3.org/People/Dom/
W3C/INRIA
mailto:dom@w3.org

Received on Friday, 9 August 2002 08:26:55 UTC