What prevents a blockchain snark from using a txn snark of a partially proved zkapp command?

[L-as]

I'm looking at

mina/src/lib/blockchain_snark/blockchain_snark_state.ml

Line 273 in ea6d94c

let%bind txn_snark_input_correct =

and can't tell what prevents me from proving part of the zkapp command. I suppose it would mean that the second pass ledger isn't updated while the target local state is?

[mrmr1993]

Nothing. It's enforced by the consensus layer (in particular, by the scan state).
There's a proposal to include a merkle list of txns in a block's statement, but this hasn't been implemented for the Berkeley hard-fork.