I'm open-minded about this. If the project has a believable path to a useful open source contribution in the future, then I think it's fundable. This could even include high variance experiments that might end up totally failing.