Every claim on this site is tiered: demonstrated on live hardware, established in controlled simulation against fairly-tuned baselines, or clearly labeled a projection. This page is the ledger.
107-channel closed-loop regulation (Rigetti Cepheus-1-108Q): the runtime held the excitation probabilities of 107 physical qubits to a hidden measured target under disclosed injected drift, converging in twelve parallel 150-shot acquisitions, in both of two runs. Finite differences at the identical budget probed 11 of the 108 evaluations its first gradient requires: zero completed updates, no improvement. Loop economics from billing metadata: one runtime update = one task, about 20 credits, about 100 ms of machine execution; one finite-difference update = 108 tasks, about 2,180 credits, 10.8 s.
Retained-versus-reset mechanism ablation (IQM Emerald, six qubits, pre-registered): two controllers identical in every component except that one carries the configuration between cycles and the other re-draws it. Five of five committed hypotheses passed; pooled reset-to-retained error ratio 7.41 against a committed threshold of 2.0.
Million-channel regulation: the unmodified production runtime holds 106 simulated coupled, drifting channels flat (~0.055 hold on the interferometer-tap mesh model), locks within seven cycles, and recovers from a full configuration scramble in one cycle. The deployed dither practice, granted oracle-exact corrections, wins at small scale and fails between 105 and 106 channels by staleness arithmetic; both numbers are published.
The direction law: per-channel readout converges at every tested size to 65,536 channels whether the detector is signed or unsigned (plain power detection suffices; direction is recovered from the loop’s own action-response correlation), while a broadcast scalar is pinned at the do-nothing level and SPSA is worse. Estimator extinction: the runtime’s drift-rate frontier is flat from 64 to 32,768 channels; SPSA’s frontier is zero by n=512, scalar finite differences a decade earlier.
Structured-QUBO solving: fed per-variable structure signals (constraint residual gradients, parity syndrome statistics), quality is 0.95 (constraint) and 0.84 (parity decoding) at n=4096 and flat from n=1024, while budget-matched greedy degrades and scalar feedback collapses. Honest rows: unstructured spin glasses trap at 0.70 of the free-digital ceiling, dense arbitrary QUBOs pay a measured quadratic feedback cost, and any landscape a digital computer can evaluate for free belongs to a digital solver.
Two public repositories contain every experiment above that runs without special hardware, with pre-stated verification bars, seed discipline (tuning seeds disjoint from analysis seeds), per-cell checkpointed result JSONs, and free-key runnable cells to n=4096. See the research page for links, the evidence ladder, and the manuscript.