A 1 in 200 chance doesn't make much sense. In (legacy) addresses there's a 4 byte checksum done with sha256, so it should be something like a 1-in-4-billion (1 in 2^32) chance of a typo being valid. bech32 does something even smarter, but I'm not familiar with the details