[somat]

Right, the "familiarity breeds contempt" problem with testing alerts too often.

With sirens it would probably make sense to have a different but similar tone/sequence for tests vs production alerts. Everyone would instinctively know that something was wrong.

Which runs right into the problem that when there is a difference between test and production. production will probably fail. Some days you just can't win. Probably the best compromise would be monthly(I agree weekly might be too much) infrastructure test, with a full blown yearly test/holiday where everything goes to hell for a day as you test actual procedures.