Every wallet checks your address before it lets money near it, and the check is not a lookup. It is arithmetic on the string itself. No server, no registry, just a re-computation of what the tail characters should be if the rest is real.
Old 1- and 3-addresses use Base58Check. The wallet strips the last four characters, hashes the remaining payload twice with SHA-256, and compares the first four bytes of that hash to the characters you supplied. One wrong letter anywhere earlier in the string, and the four characters stop matching. That's why typos in legacy addresses almost never accidentally become someone else's address.
bc1 addresses use bech32 instead. The checksum here is a polynomial computed over every character's value in a deliberately restricted 32-character alphabet. The alphabet excludes 1, b, i, o and 0 on purpose: the letters that get confused on printed paper and dim screens are the ones the format simply refuses to contain. One corrupted character is always detected, never just corrected.
0x addresses carry EIP-55: the correct casing of each letter is a Keccak-256 hash of the address itself. All-lowercase is accepted as a fallback, but mixed-case must match exactly. This one is quietly stricter than it looks, because a single flipped letter from an autocomplete is instantly a mismatch.
None of this proves the address belongs to a live wallet or that anyone can spend from it. It proves the string is a well-formed address on that chain and nowhere else. That's the whole job, and it's the job worth doing: the chain itself only asks whether the signature matches, not whether you meant to send here.
The generator here runs all of those checks before it draws a code, so the wallet that later scans it is reading your arithmetic back to you, not discovering a typo.