Draft proposal for levels

Kim and I worked on this -- but all the writing errors are mine.  It 
still needs some more work -- especially in quantifying ease and in 
overall context.

Our goal is to improve the consistency and transparency of the success 
criteria levels.  Here are some of the ideas Kim and I discussed.
__________________________________________________________________________
Determination of UAAG Levels
After a superficial evaluation of the success criteria levels that have 
been assigned over the years, the working group concluded that an “ad 
hoc” level assignment has resulted in inconsistent and potentially 
imbalanced assignment of levels.  The working group wishes to be 
objective, consistent and transparent in the assignment of levels, as 
these levels can have significant impact for both vendors and users.

Evaluation of success criterion importance must strike a balance between 
improving accessibility and the cost to both vendors (and eventually 
consumers) of adding a specific features. Success criteria that solve 
problems that completely block access are easier to assess than those 
that improve the access for groups.  Success criteria that require 
development of new features or technologies must be weighed against the 
overall benefit to accessibility.

Applying an objective evaluation of improvement to accessibility can be 
examined from the severity of the barrier to access for groups of 
disabilities – it is clear that a severe barrier to access is important. 
  It is more difficult to objectively judge the importance of success 
criteria that improve or ease access.  The working group proposes 
evaluating both the severity of the barrier and to a  lesser extent, 
from the number of disability groups it benefits.  A feature that 
improves ease of access for multiple disabilities would get a 
improvement in importance compared to other features of similar severity 
that only benefit one group.  This idea presents a number of challenges: 
protecting the needs of disability groups that may themselves be a 
minority compared to other disabilities, and comprehensively identifying 
groups of disabilities.

Each success criteria was evaluated in 4 areas:
severity of a barrier
how broadly it benefits different disability groups
existing implementations
feasibility of implementation
Barriers
Success criteria are evaluated on the level of the barrier the success 
criteria is designed to solve. Barriers that prevent a task rank higher 
than barriers that make a task more difficult or slow to perform. 
Barriers that slow a task to the point that the task cannot be completed 
(e.g., the task of reading is so laborious that the user has to stop 
from fatigue before finishing the article) count as preventing the task.
Prevent=5 |  Slow=1

Different Disability Groups
This rating attempts to put a fair value of cross-disability use based 
on an assumption that a success criterion that addresses the needs of 
multiple groups should have slightly higher weight than those that 
address the needs of only one group.  It is important to note that we 
are not considering the number of people and that meeting the needs of 
minorities is still an important goal.

[We need a good taxonomy of disability groups. I have looked at several 
this week, but haven't found one that meets our needs yet.  Most are too 
general. ]

For each category that applies to the success criteria, 1 point is 
counted for each group that has a barrier that the SC addresses.
Existing Implementations
For each browser or player that implements the SC, score 1 point.  Each 
extension, addon, or assistive technology that implements the SC scores 
.5 point.

Feasibility
Deterministic = 5 ↔ Inferential = 1
[Wayne wrote a summary of semantic vs. inferential 
http://lists.w3.org/Archives/Public/w3c-wai-ua/2012JanMar/0038.html]. 
This paragraph is extracts of his summary.
Programmatically determined characteristics of data are semantic 
properties of content that software can recognize with 100% accuracy 
every time. Otherwise, a property of data is non-deterministic and a 
program can only recognize it with probability less than one. If content 
properties are not programmatically determined then Artificial 
Intelligence is needed. AI analyzes the probabilities that a given 
content has given properties. Examples of deterministic data properties 
include: alternative text and labels linked to forms through (for / id) 
matches. Their semantic purpose is unambiguous and both can be parsed as 
part of the required syntax and semantics of HTML, normal programming. 
Tasks that are not deterministic include: error correction of missing 
data , and discovery of relationships that are given by presentation but 
no explicit markup. For a user agent to determine semantics of this kind 
data a non-determinist heuristic would be needed, namely AI.

Received on Thursday, 2 February 2012 17:06:21 UTC