However, the important thing is this -- say you've chosen the mathematically optimal strategy for a 1 in 1,000 chance of losing. You play that strategy and win. And then you play it again. Clearly, your odds of losing are no longer 1 in 1,000. But you played the mathematically optimal strategy for a 1 in 1,000 chance of losing, not for your new, lower odds. Thus you didn't actually play a mathematically optimal strategy and you will ultimately take a higher risk of going broke than you would have had you honestly specified your risk tolerance in the first place.
But don't blame that on the system because you didn't follow it. When you computed your betting strategy, you put in a risk tolerance that was lower than your actual risk tolerance, so of course it produced a sub-optimal strategy.
If you accurately specify your risk tolerance and your bankroll, you can compute a progressive betting system that precisely matches that risk tolerance. But it will only be optimal if you specified your actual risk tolerance -- repeating it will produce suboptimal results and you're always better off specifying your actual risk tolerance in the first place.
You don't really pick a specific strategy. Strategies like d'Alembert and Martingale were popular historically because they're easy for humans to do. But we all have access to computers. So it's not difficult to figure out a near-optimal strategy that maximizes the win amount and reports your odds accurately. You can then decide whether the profile is acceptable to you and, if so, execute it.
I can only add that with this level of sophistication, one had better go hunt elsewhere. I mean these strategies require so high an intellectual development and mental discipline that make the effort simply not worth it (if only for scientific ends)...