Dylan is certainly onto something.  I think the notion of "infallibility" may have ambiguated.  (Not too surprisingly--the affix combination "in...ible" indicates universal quantification over possibilities, and such quantification is always restricted in English, so there will always be a choice of restriction class....)  In Dylan's example, where the candidate-subject S is well aware of an uneliminated counterpossibility but is normatively ignoring it, I guess there may be a sense of "fallible" in which S must acknowledge fallibility:  As Dylan says, in describing the example we're saying explicitly that S knows despite there being a counterpossibility S is unable to eliminate, which does sound pretty fallibilist.
     Of course, there's still Williams' sense in which S is infallible: S's evidence eliminates all the counterpossibilities (psst--except those normatively ignored).   (Dylan suggests that tacit appeal to the proviso "is... an implicit acceptance of the Rule of Attention."  I don't yet see that.  Perhaps the idea is that suddenly going sotto voce is something we would do only if we thought that to mention a possibility aloud would make the possibility relevant/unignorable, i.e., in effect, only if we did accept the Rule of Attention.  But that ignores the fact that for the contextualist, the proviso is always there in any knowledge ascription, not only when there is a possibility that the subject is de facto aware of but normatively ignoring.  Probably I'm missing Dylan's point here.)

     But I'm not yet entirely convinced that Dylan's sense of "fallible" is really in play.  His argument does not suffice to show that it is.  He says that on Williams' view, we are entitled to say something like, "S knows that P, even though there is a possibility in which not-P uneliminated by S's evidence, because it isn't worth paying attention to; we could have properly ignored it."  But I don't think Williams' view does license that statement.  Notice the "there is."  "There is" is a quantifier, and in this sentence it is explicitly introducing an uneliminated possibility; so the possibility must fall within the quantifier's restriction class.  But if it does, and if we assume that the contextual parameter does not shift in mid-sentence, then the opening knowledge-clause would have to be false, because its same restriction class would then include the same uneliminated possibility.
     On the other hand, Dylan seems right in saying that on Williams' view, one ought to be able to say things like "S knows P despite not being able to rule out H, because H might have been properly ignored."  No quantifier there, so my objection to the previous argument doesn't apply.  And the statement does sound fallibilist.  So probably Dylan is right after all.

     Here's another example to show that when you are universally quantifying it can be perfectly proper to exclude a de facto salient individual from your restriction class.  Suppose you and I are members of an athletic team.  Our coach is a somewhat controversial figure.  Some of us love her and are loyal to her, others don't and aren't; but the team is getting along OK and having a so-so season.  Then the AD fires the coach and hires a new one.  The new one is a notorious sleaze from (as it might be) VPI, who brings two equally sleazy assistants with him.  Soon we all despise the guy.  I'm asked by someone how the new coach is doing, and I say "Everyone despises him."  Overhearing this, you say, "Not everyone; you're forgetting the two new assistants, who seem to love him."  I rejoin, "Oh, I'm far from forgetting those two scumballs.  Everyone despises them too."  It's obvious that I mean, and will continue to mean, and am well within my rights in meaning, everyone on the team.  I am properly ignoring the two assistants even though they remain all too vividly present to my awareness.

     I agree with Dylan's "side note" that the Rule of Attention is not an ad hoc, theory-driven add-on for Lewis.  It flows from his general theory of conversational scorekeeping.  I'm not yet persuaded that "the Rule is essential for an infallibilist position, and so to the extent one finds anything right about infallibilism one has reason to accept the Rule" (italics mine), because Williams can remain an infallibilist in the other sense aforementioned, even if he is shown to be a fallibilist in Dylan's.