Verb-initial languages tend to have polypersonal agreement; that is, the verb 
agrees with both the subject and (at least one, maybe more of) the object(s). 

Verb-final languages tend to have rich case systems; that is, the subject 
noun-phrase is case-marked as subject and the object noun-phrase(s) is(are) 
case-marked as object (and/or indirect object etc.). 

The explanation for this is that languages like to establish the roles as early in 
the phrase as possible. 

But, if that were so, wouldn't SVO languages case-mark the subject and have 
the verb agree with the object? But mostly they don't; rather, they tend to 
have the verb agree with the subject (if anything), and to case-mark only the 
object (if either grammatical relation). 

Am I wrong? If I'm right, why do SVO languages usually case-mark the object 
but not the subject, and also usually have the verbs agree with the subject 
but not the object, instead of the other way 'round? 

1% of the world's languages are OVS; if the "establish roles as early as 
possible" idea is correct, the object (rather than the subject) should be case-
marked and the verb should agree with the subject (rather than with the 
object). Is that what happens in those languages?