The phenomenology of rules and meta-rules

If I hypothesise that you are following a rule on the basis of your behaviour, then that hypothesis can never be 'proved' - it is always provisional.

If I suggest to you that we follow a rule, then we must know for sure what the rule means.  If we fail to agree here, then we just have confusion - not a counter-hypothesis within the game.  This is because our attempts to negotiate the rule must result either (a) in an unambiguous rule (agreement) or (b) in breakdown.

An 'external' observer might, metalinguistically, hypothesise about the breakdown (in a conversation with an appropriate meta-interlocutor...).  But this hypothesis would be based on behaviour, and so would be corrigible.

If I (or you) 'hypothesise' that our conversation has broken down, this can only be in a meta-conversation which has not broken down - can only be in a conversation with someone else.

We might have a 'private' hypothesis, here, but only in a derivative way:  it would need to be a hypothesis that could, in prinicple, be articulated.  This is a problem for any general 'hypothesis' about the playability of language games:  a negative judgement here could not, in prinicple, be articulated.  It could never be part of a rule in a shared game.

We could not incorrigibly attribute such a private hypothesis to someone else, and we could not entertain it ourselves without a private language in which it could be articulated ...