At the risk of once again showing my lack of aptitude for regexes, below is what I have currently settled on. It handles all the cases you list, accepting or rejecting them as appropriate.

This seems like a good opportunity to get another set of eyes to look it over.

    my $num    = qr!
        [+-]?
        (?: \d*? )?
        (?: (?<=\d)\.? | \.?(?=\d) )
        (?: \d*?)?
        (?: \d[Ee] (?: [+-]?\d+ ) )?
    !ox;
[download]

I'm sure there is something wrong with it, but so far it has passed everything I have throw at it.

It was ponted out to me (thankyou sauoq) that this only works if used in a context where it is tightly anchored on both ends.

Minimally, m/^$num$/o.

Note: \b wont work.

For my current learning exercise, that is fine as it will always be used as a part of a larger regex that will so constrain it.

If anyone wants to point out how to modify it to make it work without anchoring, I like to see and note the changes for my ed.

For 'real work' I would use Regex::common of course, especially given it heritage and guardianship. Now if only there was a Regex::Uncommon that did everything else, I wouldn't have to learn regexes at all:)