Space Shuttle missions returned their crew alive 98.5% of the time.

Wasn't there something in statistics to describe something like this? i.e. gaussian distributions vs something that's modeled on sparse occurrences, etc?

The author is confusing visitors with customers. Refusing entry to 2% visitors? No, no! Forgetting about non-interesting 2% visitors? Not even a blink!

Relavent XKCD comic: https://xkcd.com/325/

Hover text: You can do this one in every 30 times and still have 97% positive feedback.

I thought this was gonna be about uptime.

Well I wasn't very far off I guess! Perhaps "5 nines" is a good threshold for new CSS features too?

What’s much and what isn’t depends on the context. One hair on your head isn’t much but one hair in your soup is.

And the last 2% is often the hardest part. The low-hanging fruit has been picked, so you're left with these tricky edge cases that may not have a straightforward solution.

We shouldn't go out of our way to support IE11 anymore, sorry

This applies to AI too.

Your classifier might be 98% accurate and it may sound like a lot.

But if it sits inside a car, making thousands of decisions during every trip then you may be in deep trouble.

> Truly robust engineering isn’t about what works for most; it’s about gracefully handling the edge cases.

How do you justify this when you factor in cost and time?