Strictly speaking, a source cannot by itself be described as reliable. A source can only be reliable for verifying a piece of information.
There are two types of statements a source can verify: those that are attributed and those that are not. With the former, editors look for attributes such as independence, peer-review and a reputation for fact-checking. This can indicate it is reliable for such a statement. They also look for counter-considerations, such as contradicting other sources that also have such attributes and a lack of expertise to make such a statement. These considerations are applied just as equally to primary and secondary sources.
How considerations and counter-considerations are weighted, and the determination of reliability for a statement is made, comes down to any consensus editors can form. The community has some preferences for which considerations are more relevant; experienced editors are more able to apply such intangible preferences. If a source meets this, the material can be put in wikivoice.
When a source falls short of this, we can move from using the source to verify the content of what they said, to verifying that they said something. If the source has a credible claim to representing what it purports to be, it is considered a reliable source to verify the attributed claim. An example of a "credible claim": A book written by Donald Trump may say something, but if other sources say it was ghostwritten a distinction must be drawn between Trump saying something and a book credited to him saying something.