<a class="mention" href="/@joe">@joe</a><a class="mention" href="/@musiq">@musiq</a> I don&#39;t think that&#39;s correct. The only trace of tx1 inside of tx2 is tx1&#39;s txid. Here is tx2: inputs: [ input { input_output_point { tx1_txid, tx1_output_index } input_unlock_script [ push reduced tx2 // tx1 unlock "script" is part of tx2 and contains reduced version of tx2. tx1 lock script enforces constraints on this (tx2) transaction by examining the reduced tx2 ] } ] outputs: [ output { unlock script [ verify_sig_for_tx3 // tx2 lock script is part of tx2 and contains logic to verify validity of tx3 meta-evaluation ] } ] The only blowup I see is if you want to spend multiple outputs using this trick in one transaction, each unlock script could contain n-1 copies of the reduced tx. I think this can be mitigated with the correct choice of SIGHASH_NONE/ONE/ALL.