I wish I had enough time to actually check now, but let me just say this: the file they assigned goes. Most of the files on the PDB are from X-ray crystallography; sometimes the crystal structure isn't perfect (parts may be missing), causing a larger sidechain to be registered as an alanine or glycine, or several residues to just... disappear. Unless it's a major error in the file that would cause the protein to lose function if it happened in real life (in this case, is the cysteine supposed to form a disulfide bond?), base your model on the file, not on information you find elsewhere. Chances are, any such discrepancies won't be in anything sufficiently vital to functionality that you'd be modeling it anyway.
Alternatively, sometimes researchers use a particular mutant of a given protein- that is, the protein with one or more sidechains mutated to different residues, added, or deleted- because it is easier to crystallize than the wild-type protein or more relevant for drug discovery/other research applications. This may not be clear from the heading of the PDB file, but that information would probably be in the original publication (which is cited on the PDB page). In that case, the mutated residues might well be relevant to the function, but you should still follow what's in the given file- if there are other crystal structures available, they chose that mutant for a reason (if no other crystal structures are available, it's probably because they can't crystallize the wild-type).
In a couple months I'll have time to start getting back into the details of this event, but in general, you should base your model on the sequence that actually appears in the file specified in the rules.