The current approach compares the source output by snarkjs on the server against the source returned by Etherscan to check if the circuit source is an acceptable match. This enables small differences between the deployed source code for several practical reasons:

• Plonk output requires modification for deployment fix: remove redundant hardhat/console.sol iden3/snarkjs#464
• Groth16 entropy values change with every compilation

Verifying by bytecode would require an exact match so until these two things are resolved, it would only work for fflonk verifiers.