I’ve identified useful thunk gadgets in .altinstr_replacement, but the hard part remains register control at the hijack site. I’m going to run the hint binary in the VM now to validate live pointer shapes and then pivot to a concrete second-stage primitive based on what we can actually control there.