We're pretty flexible. In general, if a page makes the full text of an article available for free, either directly on the page, through a PDF link, or in an embedded reader, we'll use it as an OA location.
There are some exceptions. We want links to be usable (and re-usable) by anyone, and to have at least some degree of machine readability. There are two places this often comes up:
- Sites requiring registration: ResearchGate is a well-known academic social network (ASN) that requires you to log in before reading articles. We already don't use ResearchGate because we want to make sure all our locations are in compliance with publishing agreements - but aside from that, the registration requirement is a deal breaker.
- CAPTCHAs: By design, these make it very difficult for our automated systems to verify that full text is available. They also make any links we do find less useful, since one never knows if retrieving the page will show an article or a CAPTCHA. SSRN has a lot of articles we'd love to include, but after you download a few you'll notice that you're prompted to solve a CAPTCHA or make an account.