Wikipedia settles a source's reliability the same way it settles almost everything else: through public discussion, an archive, and the standing habit of reopening a question whenever the evidence changes. There is no approved-publisher list to consult, only the record of every time the question has already been argued out.
The venue is called the Reliable Sources Noticeboard, RSN for short. Anyone can open a thread there: a link to the source, a note on what claim it is meant to support, and a short case for or against treating it as trustworthy for that purpose. Other editors weigh in with counter-examples, corrections, sometimes a correction to a correction. An uninvolved editor eventually closes the thread with a summary of where the rough consensus landed. Nobody votes; the closer reads the arguments and writes down which ones held up.
Repeat that process across thousands of sources over two decades and a reference document emerges almost by accident: the perennial sources list, a running compilation of outlets that have come up often enough at RSN to earn a standing note. Four ratings cover most entries — generally reliable, no consensus, generally unreliable, deprecated — each with a line of reasoning and a link back to the discussions that produced it.
Ratings shift when the underlying publication does. A regional trade title can hold 'generally reliable' for a decade on the strength of an independent newsroom, then change hands. A marketing company buys the masthead and starts running sponsored profiles written in the same house style as the news coverage that earned the rating in the first place. The first editor to notice — a byline that once belonged to a staff reporter, now attached to pieces the outlet itself labels 'partner content' three paragraphs down — opens a new thread, and the rating catches up to what the source has become.
Reliability is a property of a specific piece of content, not of a publication as a whole; context, authorship and editorial oversight all matter.
The practical habit this creates is simple, if unglamorous: before leaning on a smaller trade outlet or a regional paper, check whether it already carries a line on the perennial list, and read the discussion behind it. A 'no consensus' rating still supports a citation, provided the citation shows its work rather than leaning on the source's name alone.
The same granularity runs in reverse for a downgraded source. Its older reporting, filed before an acquisition or an editorial shake-up, can still carry weight when cited with a date attached; nothing published after the change carries the same weight until someone reopens the case at the noticeboard. A simple reliable-or-not checkbox could never hold a distinction that fine.
A publisher's standing here is a live discussion, argued in public and revised on evidence, rather than a certificate issued once and filed away. Reading the discussion before quoting the source is most of the discipline.
