It's not always sarcastic. It is always used in at least a moderately negative context.

The principle at work here is as follows: compound adjectives unpack backwards. (It's been patented! In the medical sphere and related to probability, sure, but I'd be surprised if they use the words significantly differently than us mortals do in conversation.)

Without going into any of the (somewhat inscrutable) math in here, base probabilities (or goodness) and their negations aren't just p and 1 - p. If you look at Fig. 4, for example, 'always' has a probability of 99%, and 'not always' has a probability of 96%.

Now, comparing perfection to probability is a stretch, but I believe much the same principle applies - 'less than perfect', might very well be in the top 90% or even 95%, but it's not in the top 99%.

It's not always sarcastic. It is always used in at least a moderately negative context.

The principle at work here is as follows: base probabilities are different than their negations. (It's been patented! In the medical sphere and related to probability, sure, but the same principles carry over.)

Without going into any of the (somewhat inscrutable) math in here, base probabilities (or goodness) and their negations aren't just p and 1 - p. If you look at Fig. 4, for example, 'always' has a probability of 99%, and 'not always' has a probability of 96%.

Now, comparing perfection to probability is a stretch, but I believe much the same principle applies - 'less than perfect', might very well be in the top 90% or even 95%, but it's not in the top 99%.