Ah! But that wasn't my point really - and we've seen on BIP68 activation that there's a lot of "SPV-mining" risk anyway, where early templates are being pre-made by some script that just builds on top of the last known header. I don't know how common that still is; I should spend more time with b10c's observer.
I'm not worried about multiple valid implementations, I'm worried about this: you soft-fork in your proof-of-chips OP_CHECKLAYSVERIFY upgrade and when your activation fails with 20% support it activates. People start verifying their chips and then someone on a non-supporting node (80% of the network) spends an utxo without verified chips. Now 20% of the network hardforked off.
OP_CHECKLAYSVERIFYupgrade and when your activation fails with 20% support it activates. People start verifying their chips and then someone on a non-supporting node (80% of the network) spends an utxo without verified chips. Now 20% of the network hardforked off.