Spec Guideline: subjectivity (re: issue #83)

Issue #83 on the issues list at
http://www.w3.org/QA/WG/qawg-issues-html
argues for removal of subjective, non-testable words from checkpoints
and normative verbiage. That's a great abstract goal, which is
testable in itself by the criterion of repeatability: would 100 people
using the Spec Guidelines (SpecGL) and a spec being evaluated all come
to the same conclusion about whether that spec had "clearly" conveyed
the policies on extensions, discretionary choices, etc., etc.?

We can improve the quality of W3C specs even if the guideline must
employ "subjective" words by presenting the notion of a "reasonable
and qualified" user of the specs. Perhaps a 95% confidence rating is
applicable as well. The spec (subject of the SpecGL) must be "clear" to
at least 95% of its target audiences, not to 95% of all people who can
read it. Target audiences are those implementing the classes of product
enumerated in the spec, those implementing or running conformance tests
on the products, those who must communicate the meaning of the spec to
other audiences, other WGs who want to cite the spec as normative, and
probably others. The criterion for being in a target audience is that
you cannot rely on a second-hand representation of the content, but
rather must directly use the spec for guidance. But any lingering
subjective terms could be considered only for those members of the
target audience are qualified to be in that role. For example, a spec
that describes the operation of a processor must be "clear" to those
implementers who are qualified to create software of that complexity.

This is not to argue the main point that objective verbiage in the
checkpoints would be preferable. I just want to show that SpecGL can
accomplish its goals even if it's not possible to eradicate every last
bit of subjective verbiage.
.................David Marston

Received on Monday, 7 October 2002 13:58:10 UTC