Skip to main content
This page is the demo

Quantum error mitigation.
The layer nobody signs — signed.

Every error-mitigation technique splits in two: a hardware front-end that gathers raw counts, and a deterministic numerical back-end — the extrapolation, the matrix inversion, the shadow estimator — that turns those counts into the number you report. That back-end is the whole value, and today it runs in float Python with no provenance over the step. Here it is in pure integer Rust, byte-identical on every machine, each method recovering a hidden ideal expectation from a simulated noisy front-end.

And each step seals a signed, joule-metered wai.quantum.mitigate receipt binding (raw data, noise-model hash, method, seed) → (mitigated estimate ± error). Nobody signs the classical layer between the QPU and the reported number — this does. Honest boundary: a simulated front-end (no QPU); what is not approximate is the reproducibility of the post-processing and the receipt over it.

Zero-noise extrapolation · Richardson

—
—

Readout inversion · M3

—
—

Classical shadows · random Pauli

—
—

Three methods that need no learned noise model — so they are byte-exact today

Zero-noise extrapolation fits the observable measured at several amplified noise scales and extrapolates to zero — pure curve-fitting over measured data. Readout inversion undoes the per-qubit assignment error by a deterministic 2×2 matrix inverse applied qubit-by-qubit (the M3 spirit: work in the observed subspace, never form the 2ⁿ matrix). Classical shadows invert the randomized-measurement channel to estimate a Pauli expectation with a median-of-means. Each recovers a value it was never told, each hashes identically on every machine (no float in the estimator), and each carries a signed, joule-metered receipt whose input_hash content-addresses the exact raw data and whose noise_model_hash pins which model was used — the hook by which the SOTA ceiling (probabilistic error cancellation, tensor-network mitigation) binds an externally-learned model without WAI claiming to have learned it. A mitigated number you can put in an auditor's hands, reproduce from the git repo, and verify with a public key. Reference engine + spec in the open-standards repo, Apache-2.0.