The forgotten warm-up service still sending 40 emails a day from the old server is such an easy thing to miss after a migration, and the same-week test to 1,783 people from a second address is a clean way to rule out content. I also appreciate that you said plainly you can't separate the warm-up change from the reply-based giveaway. If you ever test it again, sending the reply campaign from a fresh subdomain first might help isolate which one Gmail responded to.