I initially thought along those lines, but I realized that if your Bayesian update includes your own strategy, you can very quickly converge to playing optimally against yourself without an explicit handshake. See my thought process here.
There are 5^N possible strings for a handshake that lasts N turns. Select the handshake strong randomly. If the handshake is successful, the probability that the bot it is playing against is itself is (1 − 5^-N).
Using N = 3, we can establish with a Pr(0.992) that we are playing against a copy of ourself within 3 turns.
2 turns would be spent renormalising if we are not playing against ourself.
I initially thought along those lines, but I realized that if your Bayesian update includes your own strategy, you can very quickly converge to playing optimally against yourself without an explicit handshake. See my thought process here.
There are 5^N possible strings for a handshake that lasts N turns. Select the handshake strong randomly. If the handshake is successful, the probability that the bot it is playing against is itself is (1 − 5^-N).
Using N = 3, we can establish with a Pr(0.992) that we are playing against a copy of ourself within 3 turns.
2 turns would be spent renormalising if we are not playing against ourself.