As you are asking open ended questions. What do you think is an appropriate level of deaths per MWh? Can you put a number on it? And how do you engineer a reactor to accurately and repeatedly achieve a certain risk level?
Well, one answer to that is when the lives saved by nuclear balance out the lives cost by fossil fuels. Nuclear currently has a rate of 0.07 deaths per terawatt-hour. Coal has a rate of 24.6 deaths per terawatt-hour. If fully replacing coal with nuclear required making nuclear a hundred times more dangerous than it is now it would still be completely justified.
So imagine a nuclear plant that saves costs by reducing safety margins. You have less redundancy in equipment and staff, less containment, or whatever. How do you accurately predict the risk? And what error would that risk have? Every single factor you are modifying can come together in weird combinations and has an error bar of its own. What happens when there is suicidal engineer + metal fatigue + unusual weather conditions? Ultimately this seems like a fools errand. The belt-and-braces approach is the only way humans have found to safely design complex inherently risky machines. And deviating from that creates Shuttle shaped clouds in the sky and Boeing shaped holes in the ground.
https://www.thelancet.com/journals/lancet/article/PIIS0140-6...