hmmm not sure about this extension, sometimes the best thing to do is just burn the ephemeral anchor value by itself, which results in a 65 vbyte txn. I think in the end it’s the same risk as people paying transaction accelerators so they can use fewer inputs in an RBF?