The paper seems to address this. It notes that some loop constructs could be made available but it would make the resource consumption estimation unnecessarily

The paper seems to address this. It notes that some loop constructs could be made available but it would make the resource consumption estimation unnecessarily conservative. Note that you can do things like unroll a loop entirely and build a Merkle tree of the possible unrolled versions, just revealing at redemption time the one actually used.