Re[3]: acceptable error rates

[Iain Harrison <iharrison -at- SCT -dot- CO -dot- UK>]

Arlen writes:

>>
Actually, it's pretty darned bad. A .2% failure rate would mean:

There would be nearly one plane crash each and every day at
O'Hare airport; I don't even want to calculate the daily number
of crashes nationwide.

... and more in the same vein.
<<

That's not so. A 0.2% failure rate doesn't mean that the plane
crashes 0.2% of the time. Defects in something as complex as an
aeroplane or nuclear power plant ARE daily occurrences. This
isn't just my guess -- I have an acquaintance who is an airline
pilot, and my brother used to work at Winscale.

The fact that a fault occurs doesn't mean a catastrophe. The
systems are (or ought to be) designed to be fault-tolerant.
Similarly, information is documentation should be presented in
such a way that a typo won't crash a system-critical
application. This also means that in some places eradicating
typos is more important than in others.

For example, a large system batch processing schedule can't
afford a single typo in the parameters listed, whereas in an
introduction for beginner users, a friendly, easy-to-read and
comprehensible style is far more important than eradicating the
last minor typo.

It is all a matter of judgement, my worry is that sometimes a
simplistic, rule-based approach to quality control can prevent a
proper quality assurance strategy.

It is better to assess the important things, rather than the
easy-to-measure things.

Iain
iharrison -at- sct -dot- co -dot- uk