How could it only need 50% accuracy? That gives you a genome of 1 bit. That's completely useless and doesn't explain how this process is going to lead to anything with higher accuracy.
There are thousands of examples of proteins that are less than 50% identical, yet fold into exactly the same shapes and catalyze exactly the same chemical reaction. Biological systems are extremely robust to change. And remember, self replicating systems do not only produce one output, they may produce millions. If only one of those million preserves the critical features, things can continue to evolve.
There's vastly more than 1 bit of information in those proteins. If there was just 1 bit, then that would mean that any randomly generated protein would have a 50% chance of having those properties (which is false.)
The question was, how accurate does replication need to be, in a replicating system with selection. My argument is that if you do not need perfect copies, and if you can make large numbers of copies, replication does not need to be very accurate to produce offspring that can continue exhibiting some property that can be selected for.
It seems like your perspective is more information theoretic, and does not include selection.