Re: Summary of Thursday's IRC conversation about @summary

On Jun 6, 2009, at 00:42, Shelley Powers wrote:

> In the IRC discussion yesterday, several references were made to it  
> not being "web scale". We assume by "web scale" that it must appear  
> in some portion of the web. I don't know, I'm not sure what "web  
> scale" is.


I think there a two main aspects to a feature working at "Web scale":
  1) Is it used enough to solve the problem it is supposed to solve  
often enough?
  2) Is it used correctly enough to have a positive overall impact?

Suppose there's a problem and a prospective solution. Also suppose  
that the prospective solution "works" in the lab in the sense that it  
has been possible to construct a demo in the lab where the solution  
solves the problem. For now, let's also suppose that whenever the  
solution is applied, users are better off than if the solution hadn't  
been applied.

Given the suppositions so far, any application of the solution  
anywhere makes someone somewhere better off. However, the feature may  
still be a failure on the "Web scale" if it doesn't solve enough of  
the occurrences of the problem as faced by users. Thus, to assess  
whether the feature works on the "Web scale" the frequency of  
application of the solution on pages that would exhibit the problem  
needs to be high enough that users feel that the problem actually gets  
solved often enough. (It's hard to come up with a hard number for what  
"often enough" is. Also note that it matters how often users would  
encounter the problem, so popular sites matter more than a random  
uninteresting page that has been unmaintained the 1990s.)

For example, the solution could be shown to solve the problem  
perfectly on intranets when applied by expert consultants but still  
fail on the "Web scale" if people browsing the Web would virtually  
always find the problem unsolved because the solution is too hard for  
casual Web authors to apply.

The assumption that whenever the solution is applied users are better  
off is not a correct assumption, though, which brings us to the second  
aspect of working on the Web scale. In practice, "solutions" can be  
abused so that users would be better off if there had been no attempt  
to apply the solution compared to the solution having been misapplied.  
For example, if the user follows a long description link only to find  
a larger JPEG image instead of a textual description, the user's time  
has been wasted and the user is worse off compared to no purported  
long description having been offered.

For the second aspect of a solution working of the "Web scale", the  
feature has to be mostly applied correctly when it is applied at all.  
That is, a solution can solve the problem in the lab or in the hands  
of expert consultants on an intranet and still fail on the Web scale  
if the less apt Web authors misapply it so often that the information  
channel provided by the solution becomes a sewer of bogus data to such  
a degree that the expected value of paying attention to the  
information channel is so low (possibly negative) that it's not enough  
to pay back the effort of inspecting the information channel. (For  
example, if the expected payoff of checking a long description is  
lower than "cost" of the effort of checking the long description.)

-- 
Henri Sivonen
hsivonen@iki.fi
http://hsivonen.iki.fi/

Received on Monday, 8 June 2009 06:58:15 UTC