Parsing behavior is weird, invalid numeral values aren't always NaN

I’m playing around with sanitizing some bulk data and came across this:

numeral('2:23:57').value()
8637
numeral('hello').value()
null
isNaN(numeral('hello').value())
false

This contradicts the documentation here which asserts two things:

1. numeral('2:23:57') should be invalid, but it isn’t. Not sure how it’s coming up with a value of 8637. Could this be a bug?
2. numeral('some invalid string').value() should equal NaN, but is instead null. This is a breaking issue because validating numerals using isNaN() no longer works!

[liubko]

example from docs is broken numeral('2:23:57').value() returns 8637

[Remixful]

Same with 1-18, value comes up as -118.