2. Free App releases.
4. Developer to Developer
Links to projects is a poor test, as there any limitless cases even here on XDA of "firmware releases" that, when looked at carefully, are really just re-releases of another persons work (or a kang of 2-3 other projects.) For example, it's trivial to release a straight AOSP android version for any device that cyanogenmod supports. Doing so doesn't imply any dev skills whatsoever. (Does XDA have a "recognized kang'er" title?)
Referrals from other devs or even from ERD's can become a popularity test
Having accounts with developer programs is also a poor test: At the company I currently work for, our Product Dev VP holds the accounts in his name, but he's NOT a developer. As well, in most cases, those types of accounts are available to anyone willing to waste the money on them (in some cases to get beta releases of iOS, for example.)
I know that I can sit face to face with a job candidate and determine if they are a faker or not, but I do that by looking at their resume and then asking them completely non-development related questions to figure out if they think like a developer or not. (Even that isn't flawless.) People with the skills to develop tend to think differently.
Another amusing test used to be to ask "What platform was 'Pascal' originally designed to run on?" - the proper answer was: "none - it was a teaching tool" (much as Knuth had his grape80 platform.) An acceptable answer was "I don't know, which platform?" Anyone taking guesses wouldn't be hired: If you can't admit you don't know something, then you won't be able to learn.
Determining who is and isn't a "real" developer is a problem, and I think it is (or perhaps "was") a problem even here on XDA. I discussed this with a few individuals (and is actually one of the reasons I never applied: I didn't want to be grouped with some of the riff-raff who have been given "RD" here.) However, I never brought that up in public as I can't think of any reasonable solution.
(That means I'm not complaining about it here. I'm merely recognizing the issue exists, as well as recognizing that there might not be a good solution.)
Actually, at least for kernel devs, there is a good way to "test" them: Get a gun and shoot them in the foot. If they smile in familiarity, then they are real.