You see, you also think in terms of "should".   But any "should" is a "constitution" that determines what is "good" and what is "bad", and hence, that has been set up by a central entity ; or is dynamically arising, and in that case, you cannot say "should".  It is what it is.

I think, for instance, that the 21 M limit is one of the biggest stupidities in bitcoin ; others think that it is its major value proposition.  

So, essentially, the choices are:

a) central authority that can dictate whatever is the rule set.

b) dynamical happening that determines what is the rule set.

I think the whole concept of a crypto currency is that there is initially a central authority that determines what is the rule set, and that afterwards, if any claim on decentralization makes sense, that from this initial rule set, a dynamical happening follows.  

My personal theory is that the only dynamical rule that guarantees decentralization, is that the original rule set is a Nash equilibrium, making it impossible to deviate from it, apart from going back to (a).   But one could design more subtle systems in which, in a decentralized way, modifications to the rules can be made without collusion.   In fact, hard forking is the standard way of doing that, but maybe there are less violent ways to do so ; or maybe, it is exactly part of the immutability dynamics, that the only decentralized way to change something, is to hard fork.

The whole question, hence, is not so much of what bitcoin "should" do, but rather, whether it is decentralized (and hence, immutable) or whether it is centralized ; and if, in that last case, in who's hands is the central power.  Until very recently, bitcoin was centralized in the hands of Core, being the software monopolist, and hence the de facto master of the rules (backed into the software).  This goes in fact against the consensus mechanism in bitcoin, that puts the power to decide in the hands of those with most hash rate.  These last ones start to get aware of their final say, and it is no wonder that both are fighting - but this exact fighting is exactly what is to be expected in a decentralized system, imposing immutability.  So this would point finally, to the observation that bitcoin is starting to get decentralized from Core.  We'll see.

Note that jbreher is perfectly right that a valid block is simply by definition, that block on which miners decide to build the rest of the chain.  And implicitly, hence, that the valid protocol is the protocol that considers that block as valid.  Whether that corresponds to any white paper or not !

If tomorrow, all miners start mining blocks in which the block reward rises to 200 BTC, and they build blocks on one another, then the bitcoin protocol has de facto changed, and is now such that block rewards are 200 BTC.  Because miners built on top of such blocks, so they decided collectively that that was the valid protocol.