By hybridizing PoW and PoS, Decred gains continuity across hard forks in a way that is not possible with Bitcoin, so fixing problems in consensus code is much less daunting a task. I have heard chatter that there are some ideas to do an actual test of the hard fork mechanism in the near future. What are the improvements that could potentially be implemented via a hard fork?
The initial hardfork example will be something simple, like updating the all protocol versions to be uint32 instead of int32, or include witness stripped size of blocks in the block header. There are a myriad of hardfork changes you could make, but they will likely end up small in number compared to the number of soft forks since the software is easily extensible due to script versioning. One often requested change is modifying the ticket price algorithm, but that needs to be done with research and care. There will be more on the methodology for forking when we get closer to getting all the recent btcd changes merged in. The backend needs a major overhaul so that forking in breaking changes can be done smoothly and efficient.