Reg Edwards wrote:
. . .

Making measurements is an ART rather than an engineering discipline.

I maintain that it's both an art AND an engineering discipline.

. . .

Something similar applies to numerical computer programs. The
reliability of a computer program depends on the programmer's
knowledge of the matter in hand and has nothing to do with the machine
it is running on. Far too much faith is placed on computed results
merely because they are computed.

Agreed. This makes it essential that computed results be compared with
measured results, or results calculated by other established methods, to
gain confidence that the computer calculations are valid.

. . .
Very little extra knowledge is gained by comparing a pair of computed
and measured results because there is no means of knowing how and from
where the inevitable difference arises.
. . .

I've found that to be completely untrue. I've done a good deal of design
work in the areas of very high speed time domain circuitry and microwave
RF circuitry, where direct measurement of internal functioning is
impossible. I and my colleagues used modeling extensively in the course
of our design work. An essential part of the process is to periodically
look at the overall performance of a real circuit and compare it to the
model. The model was adjusted (by varying unknown values, by adding
components previously thought insignificant, and so forth) until the
model adequately matched the measured results. This does, of course,
require confidence in the measured results, which is subject to a host
of potential problems. But the important thing is that agreement between
the model and measurement must be gained in order to have any confidence
in the usefullness of the model. It's not only possible, but critical to
know how and from where the differences arise.

It sounds like you're saying that when presented with computed results
we have no recourse but to put faith in the programmer. That's not at
all so. We have to test the results against measurements or other
calculations and resolve the differences before we can put much faith in
any model or program.

With the rest, I agree.

. . .

Roy Lewallen, W7EL