Once getDriftedPosition issue was resolved, a further one appeared with a "wrong" extension node.

The problem with the wrong extension node is that in this case we don’t get the underlying branch and leaf with GetProof/GetStorageProof. If we could get the underlying branch and any of its leaves, we could (with some slight additions) reuse the WrongGadget. However, to obtain one of the underlying leaves we would need to call PrefetchAccount/PrefetchStorage to get this leaf - but we don’t know its preimage (when we are querying for the key that does not exist, we get the extension node, to get the underlying nodes we need to modify the querying key accordingly, but this is not possible because we cannot know the hash preimage).

That means we cannot an underlying branch and a leaf in the proof. An alternative is check that the key RLC up until the end of the extension node is different from the RLC of the first num_nibbles key nibbles. For example:

Wrong extension node E is at path 2 2 2 and it has three nibbles 3 4 5 (in this case num_nibbles = 6).
We are querying the key 2 2 2 5 7 7 …. Such a leaf doesn’t exist, because at path 2 2 2 5 nothing exists and the extension node E is returned (found at 2 2 2).

The problem with comparing the RLC of 2 2 2 3 4 5 and 2 2 2 5 7 7 is that num_nibbles is dynamic (different from case to case). The solution might be to add split the querying key into two rows in the witness generator (two additional rows) - one row containing 2 2 2 3 4 5, the other row the remaining nibbles (compressed into bytes, which brings another problems though - even/oddness of nibbles).

In that case we can compute the RLC of the first num_nibbles using the first row and compare it to be different of E key_data RLC. And we need to ensure that the RLC of the both two rows is the same as the querying key RLC. Note that the “dynamic” problem of num_nibbles is solved this way, because using the RLP bytes (specifying the length) we can check that after some position there are only zeros in the row (using a fixed table and is_two_byte_lookup_enabled in mpt_circuit.rs).

We also need to ensure there RLP bytes correspond to the num_nibbles.