The thing is, seq-cst does not guarantee correctness, far from it. And you write down a proof, using the happen-before rules, the acquire/release edges are fair

The thing is, seq-cst does not guarantee correctness, far from it. And you write down a proof, using the happen-before rules, the acquire/release edges are fairly obvious. So I don't think seq-cst buys you a significant amount of safety.