Every quantum vendor grades its own homework, and every buyer takes the results on faith. Our engine measures hardware the way an operator measures a fleet — and answers the question the field keeps fighting about: is this device performing to spec, and is this workload genuinely hard for a classical machine?
We replace the one number with a diagnosis. For every classical explanation of what the device did, we weigh the two prices the diagram plots — cost, the classical resource it must spend (including the hidden bookkeeping a classical model has to invent), and distortion, the error that explanation has to accept. Where the cheapest classical explanation lands against the budget sorts the device into a four-way verdict: healthy and in spec, near a limit, drifting, or genuinely quantum. That last verdict has a hard backstop — once a device violates a Bell test, no classical model can match it, and the result can't be argued down.
The output of this assessment is the product: a four-way health read with a live drift monitor, and a signed certificate that a workload was hard for a classical machine. It's vendor-agnostic and runs on the data the hardware already emits.
Provisional filed (App. 64/070,738).
For any quantum interface we compute two numbers: the cost of reproducing it with a classical (Boolean) model, and the distortion that model introduces. Plot them, and the hardware lands in one of four regimes. Inside the admissible rectangle — low cost, low distortion — a classical machine can keep up. Outside it, the device is doing something genuinely quantum. The verdict comes with a signed margin, so it's robust to calibration jitter, not a coin-flip at the boundary.
The same engine runs as a live monitor. It watches a device's operating telemetry with anytime-valid statistics — alerts that stay sound no matter when you look — flags drift before a run is wasted, and decomposes an excursion into why it happened. Run it across many machines and you accumulate the one asset no one else has: comparative, cross-fleet measurement. That corpus is what turns observability into the neutral certification authority — the "UL / Moody's for quantum compute."
A verdict is only as good as the diagnostics underneath it — so we filed the diagnostics too. Three provisional families, filed July 4, 2026, protect the measurement methods the engine runs on: reading a device's two telemetry streams together, certifying the kind of correlation a channel carries, and attributing the decoherence budget with zero fitted parameters.
Correlate the error-correction syndrome stream with echo/OTOC experiment records: bottlenecks confirmed by both streams, baselines quarantined during drift, one scheduler splitting the measurement budget.
Provisional filed 2026-07-04 (App. 64/105,107).
Track every parasitic channel on the strict correlation ladder — product, classical, discord, entangled, steerable — and catch a channel drifting toward entangling crosstalk while it is still separable. Certification: every spectator channel at or below a contracted rung.
Provisional filed 2026-07-04 (App. 64/105,101).
Match the temperature knees of a device's loss, noise, and coherence channels to a ladder of small integers — only Planck's and Boltzmann's constants enter — and the ratios name the dominant channel. The answer sets the operating point, or screens the wafer.
Provisional filed 2026-07-04 (App. 64/105,099).
The cost-distortion frontier, the four-regime verdict, the classical-embedding test, and the anytime-valid wrapper are grounded in the company's published technical papers. The engine already exists inside our verification codebase, and an MVP runs daily against publicly accessible hardware — productizing it is consolidation, not invention from zero.
