users have had the option of previewing pages in one
or more browsers loaded on their machine.
You don't understand the difference between seeing one example run
once on one platform, and in testing something properly.
Who cares if it _looks_ right ? That tells me very little - there are
any number of coding problems where it will look right in one
particular set of circumstances. To validate it for the whole web, I
need to either look at it in every possible context (which just isn't
practical, given the great number of devices out there) or to
_validate_ that the code is _correct_ against some objective standard
and then hope that the devices catch up to somewhere near that
standard.
Web design is in a similar position to steam engine building in 1830,
or petrol car engines in 1900. Look at the approaches taken by
Maudslay or Cadillac to improve interoperability. The early Cadillac
wasn't as good a car as the competing Daimler Benz, but it had one
major advantage - in a famous test the RAC (Royal Auto Club)
dismantled two Cadillacs, mixed the parts and re-assembled them - they
then ran. The Daimler Benz was a better built car (the work of the
"fitter" in engineering terms) and the engine ran better - but this
was achieved by fitting each piston to each cylinder individually. In
their own context, they worked fine. Cadillac's approach was to make
each part ot a standard size and _not_ to fit them individually. They
didn't run quite so well, but they would run equally well no matter
which car you atatched that part to -- the "gold standard" for a
part's dimension wasn't its mating part, it was the objective standard
set by the blueprint.
Even the "look at it in all the browsers" approach requires 5 desktop
machines and a boxful of phones. You need a Mac, at least one flavour
of Unix, and three PCs (I don't know any way to get all versions of
IE live on a single machine, and hardware is cheaper than
multi-booting).