looks like its missing closing )

I like this idea of messages_root - curious if you have thought about the structure at all yet. I think a sparse merkle tree is most simple where the key is the blockhash and the leaf is a vector commitment to the executing (ssz'ified subset of tx data) + initiating messages (all logs emitted). If block hash isn't present then we can also do keccak256(blocknumber)

Would be useful to define a constants table that has split depth + max l1 history

Is this what you mean by the padding function?

def compute_offset(value):
    # Ensure the input is within the range of a byte
    if value < 0 or value > 255:
        raise ValueError("Input must be a byte (0-255).")
    # Convert the byte value to its binary representation, remove the '0b' prefix
    binary_str = bin(value)[2:]
    # Pad the binary string on the left to make it 8 bits long
    binary_str = binary_str.rjust(8, '0')
    # Pad the binary string on the right with 1s to make it 31 bits long
    padded_binary_str = binary_str.ljust(31, '1')
    return padded_binary_str

This is basically saying, the more steps of bisection, go back less in time?

Assumption that all chains have the same blocktime at 2 seconds

I think it would be helpful with typing if we used path[0] == 0b0, ie the 0b prefix where it makes sense, it helps me keep in context what things are

Can you elaborate on this todo?