Actually... the account model is much easier to use to get an accurate view of all balances than the UTXO model. As for rollbacks, Bitcoin had one in 2010 (https://en.bitcoin.it/wiki/Value_overflow_incident) and a chain fork in 2013 (https://bitcoin.org/en/alert/2013-03-11-chain-fork) both decided by those who could make the decision... and accepted by a majority of nodes, same as with the ETH hard fork
I'm not so sure about understandability at protocol level, I do believe Ethereum to be straightforward but then again I've followed its progress over the years
Finally looked at both of those bitcoin events that you are attempting to equate to the DAO fork. I commented about the first one elsewhere here. The second one was indeed a chain fork, which is actually a normal and expected event in Bitcoin when miners disagree (for whatever reason) about which blocks are valid or not.
In this case the disagreement happened because of a backwards incompatible change that was accidentally made to the mining software. Nodes running the old software rejected blocks generated by the new software. The bug was fixed and miners happily stopped using the buggy version of code and the chain fork was resolved, just as designed by the Bitcoin protocol. Nodes that never ran the buggy version didn't have to do a thing.
Like my discussion of the other bitcoin fork, this to me looks like an entirely different category of event than the DAO. Bitcoin fixed a broken promise in both cases. Ethereum broke a promise in the DAO fork.
I'm not so sure about understandability at protocol level, I do believe Ethereum to be straightforward but then again I've followed its progress over the years