We do commit to a set of features, not just to all the details of how they are implemented. For example it was clear from the start we'd have multi sub dispatch, but the exact semantics for figuring out which candidate to call has changed in the past, for very good reasons.

We also have a core of syntax and semantics that has changed very seldomly in the recent years, but we don't commit to backwards compatibility yet.

There are two main reasons for that. The first is that some areas of the language are yet largely unexplored, for example concurrency. While the Perl 6 design team had concurrency in the back of their heads when desiging the language (which is why much more stuff is done lexically, instead of in global variables as in p5), there is little concrete spec on how concurrency works yet, and it might affect some core features in small, backwards incompatible ways.

The second is that each commitment to backwards compatibility is a possible design debt. Perl 5 has way too much design debt already, and we don't want to repeat its mistake. So we rather commit later than sooner to backwards compatibility.

At some point in time we'll have to reconsider that choice, but I don't feel we're quite there yet.