I tried to file this as a bugreport in the tracker, but it was rejected as possible spam :(
During extraction the results from an extractor plugin are used to create a message which is then added to the catalog. When the Message instance is created it can set the {{{python-format}}} flag if the text matches a specific regular expression. That means that a text like this:
Did you know 50% of all users do not know how to do this?
will be flagged as a python-format string even it does not originate from python code, or from a piece of python code that will never use formatting. This is breaking our translations in a pretty bad way currently.
As far as I can see the extractor is the only thing that should decide the flags for a message. The Message class itself should never try to guess flags or force, especially since extractors have no way to override this behaviour.
Regards,
Wichert.