Which Ones Broke?

A weather volunteer typed a week of rainfall readings, in millimetres, into a spreadsheet. Most are fine, like "12" or "3.5". A few aren't: "n/a", "3,5" typed with a comma, "twelve", or an empty "".

Before fixing anything, the volunteer wants to see exactly which entries pandas can't read, so they can look them up in their notebook.

Task: write unreadable(readings), where readings is a list of strings.

• Convert the readings to numbers, turning anything that can't be read into a missing value.
• Return the original text of every entry that couldn't be read, in the order it appears. Keep repeats: if "n/a" appears twice, it's in the list twice.
• If every entry could be read, return [].

There is always at least one reading. No reading has spaces in it.