Read The F*ing Manual is in general a good maxim to live by. But, what if the manual is all over the place? Or, if the manual's readers are the ones who in some obscure note, give the info you need?
Faced quite a quandary a while back, trying to setup Twitter cards for my blog. Even knowing how finicky and layered things are with Twitter's DevX, this one did take the cake.
What I presumed was a matter of extending the existing og:<> to twitter:<> ended up becoming an ordeal of an hour. Even with requisite tags, the validator returned a no metatags found back to back. 
Half way through the thread, went back to reading through the documentation. There isn't much there except for the usual, "this is what we need, this is the format" kind of generic. But interesting enough, post midway in the original thread that I looked up, lay the answer.
It appears, without a Content-Type of text/html, Twitter's validator fails.
There are a few questions here - why not use sensible defaults? why not be comprehensible? After all, the only three kinds of content are articles, player videos, and app links. It does make sense to switch to the first as a fail-safe.
Aside, when there is a dependence on the content-type, would it not make it easier for developers to say so? More so, when third party libraries enable this, used by a tech-unaware publisher.
Twitter's DevX isn't the best out there and such trippings do make things even more a pain on a troubled neck.